Impact of Land Use Change on Carbon Storage Dynamics in the Lijiang River Basin, China: A Complex Network Model Approach

Abstract

As a typical karst landform region, the Lijiang River Basin, located in Southwest China, is characterized by both soil erosion and ecological fragility. The transformation of land use, driven by long-term intensive human activities, has exacerbated the degradation of ecosystem services, threatening the region’s carbon sink function. To clarify the coupling mechanism between land use and land cover change (LUCC) and carbon storage, this paper integrates complex network theory with the PLUS-InVEST model framework. Based on land use data from five periods, i.e., 2001, 2006, 2011, 2016, and 2021, the key transformation types are identified, and the evolution of carbon storage from 2021 to 2041 is simulated under three scenarios, namely, inertial scenario, ecological protection scenario, and urban development scenario. The paper finds that (1) land use transformation in the basin exhibits spatial heterogeneity and network complexity, as evidenced by a significant negative correlation between the node clustering coefficient and the average path length, revealing that land type transitions possess small-world network characteristics. (2) The forested land experienced a net decrease of 196.73 km² from 2001 to 2021, driving a 3.03% decline in carbon storage. This highlights the inhibitory effect of unregulated urban expansion on carbon sink capacity. (3) Scenario simulations indicate that the carbon storage under the ecological protection scenario will be 1.0% higher than under the inertial scenario and 1.5% higher than under the urban development scenario. These suggest that restricting impervious land expansion and promoting forest and grassland restoration can enhance carbon sink capacity. Therefore, this paper provides a quantitative basis for optimizing territorial spatial planning and coordinating the “dual carbon” goals in karst regions.

1. Introduction

The rising global social costs triggered by global warming have compelled countries to accelerate the development of systematic emission reduction strategies. This urgent need has led to the Baku Climate Solidarity Agreement [1], adopted at the 29th United Nations Climate Change Conference (COP29). By linking carbon reduction targets directly to economic loss mitigation, this agreement marks a critical shift in climate governance—from scientific warnings to action-based accountability [2]. As a key carbon sink, terrestrial ecosystems maintain the dynamic balance of carbon storage through two primary mechanisms, that is, reducing carbon emissions and enhancing carbon sequestration within terrestrial ecosystems. The terrestrial carbon storage system helps regulate the climate by absorbing and releasing atmospheric greenhouse gases through plants and soil. The conversion of different land cover types can cause fluctuations in terrestrial ecosystem carbon storage [3,4]. Therefore, a comprehensive assessment and prediction of carbon storage associated with land use is crucial for the overall sustainable development of a region.

In recent years, scholars both domestically and internationally have conducted extensive research on carbon storage in terrestrial ecosystems, primarily employing field survey methods [5] and model simulation methods [6]. However, field surveys cannot effectively capture carbon storage changes across spatiotemporal gradients. Consequently, many researchers have turned to mathematical modeling for visualizing and assessing carbon storage. Among existing models, the InVEST model is the most widely used due to its computational efficiency and low data requirements. For example, M. Imran et al. [7] utilized the InVEST model to conduct geospatial mapping of carbon storage across mountain forest gradients. With the deepening of research, some scholars have noted that the InVEST model is primarily applied in the assessment of ecosystem services, while often overlooking the simulation and evaluation of land use dynamics. At present, various models are available for predicting land use change, including but not limited to CLUE-S [8], CBM-CFS3 models [9], Markov chain models [10], and the integrated CA–Markov models [11]. For instance, Kohestani et al. [12] applied the Markov model to perform spatiotemporal modeling of land use patterns in the Nour-rud River Basin in northern Iran. Rediet Girma et al. [13] employed a neural network-coupled CA–Markov model to predict land use/land cover (LUCC) changes in the Gidabo River Basin, whereas Ines Grigorescu et al. [14] adopted the CLUE-S model to forecast and evaluate LUCC dynamics in Romania. However, the CLUE-S model [15], despite its widespread use, exhibits several limitations, including an overreliance on static parameters and clustering algorithms. These constraints restrict its capability to simulate complex dynamic interactions and multi-scale feedback mechanisms. The FLUS model, which is based on the ANN-CA framework, also suffers from certain drawbacks, such as being highly sensitive to training data bias and inflexible in incorporating bidirectional ecosystem feedbacks in long-term simulations [16]. The Markov chain model assumes a temporally static transition probability matrix, thereby overlooking spatial dependencies and dynamic driving forces [17]. As a result, it struggles to capture the nonlinear characteristics and spatial heterogeneity inherent in land use change. In contrast, the PLUS model, as a new-generation land use change simulation tool, demonstrates significant advantages in mechanism representation, spatial accuracy, and adaptability over the FLUS, Markov, and CLUE-S models [18]. It is currently regarded as one of the most effective approaches for simulating the spatiotemporal dynamics of land use change. Therefore, the temporal forecasting capability of the CA–Markov model can be effectively integrated with the spatially explicit simulation strengths of the PLUS model. By incorporating multi-source data to dynamically adjust transition probabilities, such an approach enhances the accuracy and scenario adaptability of land use change predictions. Consequently, a growing number of studies have employed coupled land use change models to estimate carbon storage. For instance, Li et al. [19] used the PLUS-InVEST model to predict carbon storage in Hangzhou by 2030, while Zhang et al. [20] applied the same model to assess carbon storage under four different scenarios for the Chengdu metropolitan area by 2050.

The LUCC analysis methods, such as transition matrices and type proportion analysis, effectively quantify the cross-sectional characteristics of land use changes [21]. However, these methods are limited in their ability to dynamically analyze systemic cascade effects and structural evolution. Based on complex adaptive system theory, land type conversion exhibits node interaction, path dependency, and emergence [22], which are intrinsically linked to the small-world and scale-free topological properties of complex networks. Currently, complex network methods are primarily applied to research on characteristics of land use change and ecological network optimization [23]. For example, the urban agglomeration land use transmission network in the middle reaches of the Yangtze River was developed in [24], where the weights and directionality of the dominant pathways were quantified, revealing spatial differentiation patterns in the urbanization process. Ji et al. [25] found that high-coverage grasslands, low vegetation cover surfaces, and deciduous–evergreen mixed forests are the key land types in the Yellow River Basin’s land system by using the complex networks. Wang et al. [26] identified priority areas for ecosystem restoration and protection based on complex network and circuit theory. By introducing complex network methods to construct a land transition topology model, it is possible to couple macrostructural characteristics with micro-scale transition pathways, uncovering hidden transfer channels and identity key hub types [27]. This multidimensional analytical framework enhances the systematic characterization LUCC dynamic processes and introduces a new quantitative dimension for assessing critical thresholds in land system evolution [28].

However, in the current research trend, the case study area predominantly focuses on urban areas or metropolitan regions defined by administrative boundaries, while paying relatively little attention to China’s karst geomorphological regions. As a representative karst landscape area, the Lijiang River Basin constitutes an essential component of the southwestern Chinese ecosystem and serves as a significant carbon reservoir in northern Guangxi. Unfortunately, the region is characterized by the coexistence of severe soil erosion and ecological fragility. Furthermore, under the long-term and intense impact of human activities, the transformation of land use has accelerated the degradation of ecosystem services and posed a serious threat to the region’s carbon sequestration capacity. Therefore, this study adopts a complex network approach to explore the historical evolution of land use in the area. The overall structural properties of the network are then analyzed to assess the stability of the land use system. Subsequently, this study simulates landscape patterns for 2031 and 2041 under different scenarios using the modified PLUS model. Additionally, the InVEST model is used to assess ecosystem carbon storage under three scenarios, namely, inertial development, ecological protection, and urban expansion. The results offer a coordinated development pathway that balances spatial optimization and carbon sink enhancement, contributing to the implementation of regional carbon neutrality goals and watershed management strategies.

2. Materials and Methods

2.1. Overview of the Research Area

As shown in Figure 1, the Lijiang River Basin serves as a crucial ecological corridor, connecting the Xijiang River system of the Pearl River Basin, is located in the northeastern Guangxi Zhuang Autonomous Region, with Guilin City as its core area, extending to Xing’an and Yangshuo, which are characterized by karst landforms. The region has a subtropical monsoon climate, with an average annual temperature of approximately 20 °C and an average annual precipitation of 1949.5 mm [29]. The river stretches 214 km and covers an area of approximately 5768.29 km² [30]. It is worth noting that karst landforms refer to surface and subsurface features primarily formed by the dissolution of soluble rocks due to the erosive action of water, supplemented by mechanical erosion processes such as scouring, undercutting, and collapse caused by flowing water [31]. In this paper, for the convenience of research, areas without karst landforms will be considered ordinary landforms, namely non-karst landforms. Karst landforms, which are concentrated in the middle and lower reaches of the basin, form the geological foundation of the “Guilin Scenery” and also serve as a core ecologically sensitive zone. However, this landscape is highly susceptible to human activities and is difficult to restore after environmental degradation, which constrains the direction of regional economic development [32]. In 2021, the Lijiang River Basin developed an economic structure dominated by the tertiary sector, with a total GDP of CNY 231.1 billion. The COVID-19 pandemic caused Guilin’s tourism revenue to plummet by 99.2%, exposing the structural risks of excessive reliance on a single industry. Meanwhile, regional development disparities remain prominent, with an urbanization rate of only 53.4%, while municipal districts account for 75% of the total economic output.

2.2. Source of Data

The land cover dataset of the Lijiang River Basin is derived from the annual land cover dataset of China published by Professor Huang [33]. We cropped the Lijiang River Basin and classified it into seven land use types: cropland, forest land, shrubland, grassland, water bodies, impervious land, and barren land. We considered both natural and socio-economic factors (

Table 1. Introduction to the data.

Data Type	Name	Source	Resolution
Topography and Landforms	Digital elevation model	ALOS (accessed on 1 March 2022, https://search.asf.alaska.edu/#/)	12.5 m
	Slope
	Aspect
	Curvature
Climate	Surface runoff	ERA5 (accessed on 15 March 2022, https://cds.climate.copernicus.eu/)	0.1° (11,132 m)
	Temperature
	Precipitation
	Evapotransiration	MOD16A2 (accessed on 20 March 2022, https://lpdaac.usgs.gov/)	500 m
Distance	Distance to river	National Geomatics Center of China (accessed on 5 April 2022, https://www.ngcc.cn/)	30 m
	Distance to artificial
	Distance to town
	Distance to road
	Distance to rural residential area
Economic	Population density	WorldPop (accessed on 25 April 2022, https://www.worldpop.org/)	100 m
	Rural per capita net income	Social and Economic Statistical Yearbook (accessed on 10 June 2022, http://tjj.gxzf.gov.cn/tjsj/tjnj/)	1000 m
	Fiscal revenue
Tourism	Tourist point density	(accessed on 5 August 2022, https://flight.qunar.com/)	\
	Hotel density
Else	Rocky desertification	Combined with slope climate vegetation and other factors to calculate (accessed on 20 October 2022)	30 m

). All datasets were collected between 1 March 2022 and 31 October 2022 and uniformly resampled to a 30 m × 30 m resolution to ensure consistency in the analysis.

2.3. Research Framework

The framework of this paper is mainly divided into four parts, which is shown in Figure 2. Firstly, the Gephi (0.10.1) was used to create a weighted directed network where treating land use types as the network nodes and the land use transitions as the network edges, the stability of Lijiang River was analyzed from the perspectives of node in-degree, out-degree, and centrality. Then, we set three different development scenarios by adjusting the transition probabilities of land use types, and the Markov–PLUS model was used to predict land use patterns. Finally, the ecosystem carbon storage was assessed by combining the carbon density data with the InVEST model.

2.4. Methods

Firstly, we used the land use transition matrix from 2001 to 2021 to construct a weighted directed land use network where the nodes represent different land use types, the direction of the edges indicates the transition relationship between land types, and the weight of the edges is the ratio between the area of transition from land type 1 to land type 2 and the total transition area. The thickness of the edges represents the magnitude of the weight. The degree of nodes can partially represent its importance, and betweenness centrality allows the assessment of node importance from the perspective of overall network connectivity. The evaluation indicators used to analyze the structural characteristics of the LUCC transition network over the entire period mainly included in-degree and out-degree, betweenness centrality, average path length, and clustering coefficient [34]. The following concepts are used to evaluate the stability of the land use system in the Lijiang River Basin.

(1)

Degree: The total number of edges connected to each node, including both out-degree and in-degree, which are given as follows:

$${in\_deg}_m={\sum }_{n\ne m,n\in N}{L}_{n\to m}$$

(1)

$${out\_deg}_m={\sum }_{n\ne m,n\in N}{L}_{m\to n}$$

(2)

where ${in\_deg}_m$ is the number of edges entering the node; ${out\_deg}_m$ is the number of edges departing from the node.

(2)

Betweenness Centrality: The ratio of the number of times a node is passed through by the shortest path from other nodes to the total number of shortest paths in the graph.

$$C^B\left(\mathrm{v}\right)={\sum }_{s\ne v\ne t\in V}{\displaystyle \frac{{\sigma }_{st\left(v\right)}}{{\sigma }_{st}}}$$

(3)

where ${\sigma }_{st\left(v\right)}$ is the number of shortest paths from $s\to t$ that pass through node $v$, reflecting the importance of node a as a bridge.

(3)

Closeness Centrality: The closeness between a node and other nodes in the network. If a node is close to all other nodes, it does not need to rely on other nodes to transmit information, indicating that this node is important.

$$C_V={\displaystyle \frac{\left|V\right|-1}{{\sum }_{i\ne v}{d}_{vi}}}$$

(4)

where $V$ is the total number of nodes in the network; $d_{vi}$ is the shortest path length between node $v$ and node $i$. It is a variant of closeness centrality, typically used to handle situations where nodes are unreachable in a graph. Unlike traditional closeness centrality, harmonic closeness centrality avoids penalizing unreachable nodes by taking the reciprocal, thus ensuring that all nodes have a defined centrality value.

Then, we used the Markov–PLUS model to simulate. This model comprises two parts, that is, the Land Expansion Analysis Strategy (LEAS) rule-mining framework and the CA model based on multitype random seeds (CARS) [35]. The LEAS employs the random forest algorithm to analyze growth probabilities and their drivers [36]. This framework assesses land use development potential and quantifies the contribution of each driving factor to expansion. The function is shown as follows:

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

(5)

where $P_{i,k\left(x\right)}^d$ is the development probability of $k$ land use type in unit. There are land use types that convert into the $k$-th land use type when d = 1, and others when d = 0. $x$ is a vector composed of multiple driving factors, $i$ is the indicator function of the decision tree, $M$ is the total number of decision trees, and $h_n\left(x\right)$ is the prediction type of the $n$-th decision tree of vector $x$.

The CARS, combined with random seed generation and a threshold-decreasing mechanism, allows the PLUS model to automatically generate spatiotemporal dynamic simulation patches under development probabilities constraint. Using the Monte Carlo method, when the land neighborhood effect $k$ is 0, the probability surface ${OP}_{i,k}^{1,t}$ for each land use type is as follows.

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

(6)

where $k$ is the type of land use, ${\Omega }_{i,k}^t={\displaystyle \frac{con(c_i^{t-1}=k)}{n\times n-1}}\times w_k$ is the neighborhood effect of the unit which means that it depends on the gap between the current number of land use type $k$ and the target demand, $r$ is the random variables from $0$ to $1$, and ${\mu }_k$ is the threshold for generating new land use patches for the $k$ type of land use. $D_k^t$ is an adaptive drive coefficient, satisfying

$$D_k^t=\left\{\begin{array}{c}{D}_k^{t-1}, if\left|G_k^{t-1}\right|\le \left|G_k^{t-2}\right|\\ D_k^{t-1}\times {\displaystyle \frac{G_k^{t-2}}{G_k^{t-1}}}, if 0>G_k^{t-2}>G_k^{t-1}\\ D_k^{t-1}\times {\displaystyle \frac{G_k^{t-2}}{G_k^{t-1}}}, if G_k^{t-1}>G_k^{t-2}>0,\end{array}\right.$$

(7)

in which $G_k^{t-1}$ and $G_k^{t-2}$ are the difference between the current demand and future demand for land use type $k$ during the $t-1$ and $t-2$ iterations. It is worth noting that, in this paper, we assume seeds can generate new land use types and grow into a series of new patches.

In order to ensure the sustainability of carbon storage resources in the Lijiang River Basin, this paper sets three scenarios: the natural development model, the ecological priority model, and the urban development model. The specific settings are as follows:

(1)

Inertial Scenario: Given the assumption that the evolution of landscape patterns in the basin is unaffected by the new policy and that the parameters for neighborhood factors and transfer cost matrices of each land use type remain unchanged. Based on the evolution of landscape pattern in 2011, 2016, and 2021, the Markov model in the PLUS model is used to predict land demand.

(2)

Ecological Protection Scenario (EP): In this scenario, future government planning documents are primarily referenced. Considering the reduced demand for urban impervious land by residents, ecological civilization construction is prioritized and serves as the guiding ideology for urban development. The management of ecological protection areas such as forest land, grassland, and water bodies is strengthened, and the conversion of land use types within park green spaces and ecological protection areas into impervious land is curbed. The trend of uncontrolled expansion of impervious land is strictly controlled, and the decline of forest land and cropland is slowed down. Based on the natural development model, the conversion rate of forest land, shrubland, and grassland into impervious land is reduced by approximately 50%, while the conversion rate of cropland into forest land is increased by 30%.

(3)

Urban Development Scenario (UP): The New Urbanization Plan for Guangxi (2021–2035) indicates that the level of urbanization development in Guangxi is currently below the national average. In this scenario, the conversion rate of cropland, forest land, shrubland, and grassland into impervious land increases by 20%, while the conversion rate of impervious land into cropland, forest land, shrubland, and grassland decreases by 20%.

2.5. Carbon Storage Based on the InVEST Model

We use the carbon storage and sequestration part of the InVEST model [37] to access the change in carbon storage of Lijiang River. Based on data of land use and carbon density, the terrestrial ecosystem carbon storage is divided into vegetation aboveground biomass carbon storage, vegetation underground biomass carbon storage, and soil carbon storage and dead organic carbon storage, the function is as follows.

$$Ci\_tot=Ci\_above+Ci\_below+Ci\_soil+Ci\_dead$$

(8)

where $i$ is the land use type, $Ci\_tot$ is the total carbon density in the ecosystem, $Ci\_above$ is the aboveground carbon density of vegetation, $Ci\_below$ is the underground carbon density of vegetation, $Ci\_soil$ is soil carbon storage, and $Ci\_dead$ is the carbon density of dead organic matter. By combining the vegetation biomass distribution map of China (with a spatial resolution of 500 m) published on the PIE-Engine platform and in the relevant literature [38,39,40,41], we obtain the carbon density of land use landscape types in Lijiang River, which are shown in

Table 2. Carbon density of land use landscape types in Lijiang River Basin (t/hm²).

Types	Ci_above	Ci_below	Ci_soil	Ci_dead
cropland	13.50	2.70	96.59	1.00
forest	73.59	25.20	207.33	3.50
shrub	18.96	5.69	9.40	2.47
grassland	5.01	13.53	117.06	1.00
waters	0.21	0.00	0.00	0.00
barren land	19.52	3.90	0.86	0.00
impervious land	1.20	0.93	12.48	0.00

.

3. Result

3.1. Analysis of LUCC Characteristics Based on Complex Networks

Based on the land use change data for each period, we obtained the weighted and directed networks for 2001–2006, 2006–2011, 2011–2016, and 2016–2021, respectively. The average path lengths are 1.262, 1.214, 1.19, and 1.214, and the clustering coefficients are 0.824, 0.84, 0.857, and 0.836. As shown in

Table 3. Complex network node in-degree and out-degree in land use conversion.

Land UseType	2001–2006 s			2006–2011 s			2011–2016 s			2016–2021 s
	Input	Output	Output/Input	Input	Output	Output/Input	Input	Output	Output/Input	Input	Output	Output/Input
cropland	6	6	12	6	5	11	6	6	12	6	6	12
grassland	6	5	11	6	6	12	6	6	12	6	6	12
shrub	3	4	7	4	4	8	4	3	7	4	5	9
impervious	5	4	9	5	5	10	5	6	11	6	4	10
forest	5	6	11	6	5	11	5	6	11	5	6	11
barren	3	2	5	1	4	5	4	3	7	3	4	7
water	4	5	0	5	4	9	4	4	8	4	3	7

, there exists heterogeneity in the in-degree and out-degree of nodes across different time periods in the basin. Moreover, forests, shrubland, water bodies, and barren land exhibit characteristics of output land use types, while impervious land exhibits characteristics of input land use types. The various types of land use have different degrees of conversion into impervious land, with the largest conversion being from cropland, totaling 67.486 km², accounting for 88.46% of the total converted area.

It is worth noting that the average shortest path reflects the stability of the network. A smaller average shortest path indicates that the nodes in the network can more easily connect with each other, suggesting that the network is more active and the land system is less stable. As shown in Table 3 and Figure 3, land use transitions within the basin were primarily concentrated among cropland, forest land, and impervious land, while other land categories exhibited relatively minor changes in area. Overall, the ratios between input and output for cropland, grassland, and forest land remained relatively stable, indicating a state of dynamic equilibrium. Notably, the input of impervious land showed a continuous increase, reaching a ratio of 11 in 2016, suggesting an accelerated urbanization process. During the period from 2016 to 2021, the output of impervious land was 4, while the input rose to 6, reflecting a sustained trend of expansion. In contrast, the input of bare land dropped sharply to 1 from 2006 to 2011, while the output increased to 4, resulting in a ratio of 5. This period may correspond to specific development activities or ecological restoration projects. It can be found that the average path length of the land transfer network does not exceed 1.3, indicating that the overall stability is relatively poor, and the land use types are easily convertible during the four time periods. Furthermore, the average path length showed a slight downward trend, indicating that the land transfer speed in the watershed accelerated from 2001 to 2011. Between 2006 and 2011, the in-degree of forest land was smaller than its out-degree, influenced by the policy of returning cropland to forest, which restrained the growth of cropland. Due to the disorderly urban expansion and the decrease in the stability of the watershed ecosystem, the average path length dropped to 1.19 from 2011 to 2016. In recent years, the Chinese government and local authorities in the watershed have strengthened the regulation of ecological protection, resulting in a slight increase in the average path length from 2016 to 2021.

Besides the average shortest path, we also focused on the average clustering coefficient. According to [42], we can know that a higher average clustering coefficient means that most nodes in the network tend to form tightly connected communities. In contrast, a lower clustering coefficient indicates that the nodes in the network form fewer tight-knit groups, and the network structure is more loosely connected. From Figure 3, it can be found that the clustering coefficient increased to 0.857 from 2001 to 2016. However, from 2016 to 2021, the clustering coefficient decreased, indicating that the previously tight network structure is gradually transforming. Overall, the land ecosystem in the Lijiang River Basin exhibits a tight land use conversion relationship, with frequent and rapid reciprocal conversion relationships between different land use types. However, we also found that the trend in the clustering coefficient of the land ecosystem is opposite to that of the average path length from 2001 to 2016. This indicates that urban expansion leads to more direct and frequent connections, creating localized high-density development areas. This change helps improve regional connectivity and efficiency, but it may also bring ecological pressure and environmental issues.

3.2. Analysis of Land Use Transfer Quantity and Spatial Changes

It can be observed from Figure 4 that, in terms of transfer types, the main conversions are from forest land to cropland, impervious land, and shrubland, from cropland to forest land, impervious land, and water areas. In terms of spatial distribution of these transfers, the central, western, and southern parts of the basin have a larger transfer area compared to the northern and eastern parts. As shown in

Table 4. The land use transfer matrix of the Lijiang River Basin from 2001 to 2021 (km²).

2001	2021
	Cropland	Forest	Shrub	Grassland	Water	Impervious Land	Total	Transferred Out
cropland	1115.16	340.06	4.65	2.03	9.86	2.16	1473.93	358.77
forest	139.22	3977.39	16.63	0.63	1.20	0.21	4135.29	157.90
shrub	0.07	5.14	2.31	0.04	0.00	0.00	7.56	5.25
grassland	0.44	0.82	0.30	0.42	0.03	0.00	2.01	1.59
waters	5.07	2.07	0.00	0.26	43.05	1.41	51.85	8.80
impervious land	67.49	6.42	0.01	0.59	1.80	82.66	158.96	76.29
total	1327.46	4331.90	23.91	3.96	55.93	86.44	5829.61	——
transferred out	212.30	354.51	21.60	3.55	12.88	3.78	——	608.62

, all land use types in the Lijiang River Basin experienced transfers, with a total transfer area of 608.62 km², accounting for 10.44% of the total area of the study region. Furthermore, it was also found that a total of 211.3 km² of cropland was transferred out, with the reduction in cropland mainly converted into forest land and impervious land, accounting for 139.22 km² and 67.49 km², respectively. A total of 354.51 km² of forest land was transferred out, which is 2.25 times the amount transferred in. The transferred forest land mainly flowed into cropland, impervious land, and shrubland, with transfer areas of 340.06 km², 6.42 km², and 5.14 km², accounting for 95.92%, 1.81%, and 1.45% of the total transfer out, respectively. Different land use types were transferred into impervious land to varying degrees, with the largest transfer coming from cropland, which accounted for 88.46% of the total transfer area.

It was found that the land transfer in the Lijiang River Basin is primarily reflected in the mutual conversion between forest land and cropland, as well as the transformation of these land types into impervious land, water areas, and others. The conversion between forest land and cropland is widespread across the entire basin, especially in areas such as Tanxia Town, Lingui Town, and Lingchuan Town. In contrast, in the southern areas like Baisha Town, Jinbao Township, and Yangshuo Town, the conversion is more characterized by cropland being transformed into forest land, which reflect a stronger emphasis on ecological restoration or protection in these regions. The conversion of forest land to shrubland is primarily concentrated in the section from Daxu Town to Pingle Town in the middle and lower reaches, while such changes are less common in areas like Lantian Yao Township and Hekou Yao Township in the northwest. The conversion of forest land into impervious land is more commonly observed along the Lijiang River, such as Huajiang Yao Township, Rongjiang Town, and the main urban area of Guilin, highlighting the impact of urbanization on the natural environment. The areas where cropland is converted into impervious land are similar to those of forest land, but the conversion area is larger, especially in the central-western region. The conversion of water areas into cropland is more common in the upper and middle reaches than that in the lower reaches. Due to illegal sand mining activities, leading to deforestation, riverbed destruction, and significant soil erosion, the upper reaches of the Lijiang River experienced severe damage. The conversion of shrubland to forest land is concentrated in the southeastern areas, such as Dabu Township and Yangdi Township, reflecting efforts for vegetation restoration. The conversion of water areas into cropland follows a linear distribution, with a particularly noticeable trend near the Qingshitan Reservoir.

3.3. Carbon Storage Change Analysis

3.3.1. Temporal Variation in Carbon Storage in the Lijiang River Basin from 2001 to 2021

The carbon storages of Lijiang River Basin in 2001, 2006, 2011, 2016, and 2021 were 149.39 × 10⁶ t, 147.97 × 10⁶ t, 146.38 × 10⁶ t, 144.28 × 10⁶ t, and 144.87 × 10⁶ t. From Figure 5, we can find that, from 2001 to 2016, the carbon storage in the Lijiang River Basin gradually declined. However, from 2016 to 2021, the carbon storage stabilized and showed a slight increase. Overall, the carbon storage in the Lijiang River Basin shows a pattern of significant reduction followed by a slow increase. The overall land use/coverage pattern is developing steadily and positively, which is intrinsically related to land use transitions. There is a significant consistency between the changes in carbon storage and land use types in the Lijiang River Basin. The change in carbon storage is closely related to the conversion of land use types (

Table 5. Carbon storage transfer matrix (10⁶ t), where “-” in the table indicates missing values.

2001	2021
	Cropland	Forest	Shrub	Grassland	Waters	Impervious Land	Total	Transferred Out
cropland	12.69	−6.66	0.04	0.00	0.11	0.02	6.19	−6.49
forest	2.73	123.15	0.45	0.01	0.04	0.01	126.38	3.23
shrub	0.00	−0.14	0.01	0.00	0.00	0.00	−0.13	−0.14
grassland	0.00	−0.01	0.00	0.01	0.00	0.00	0.00	−0.01
waters	−0.06	−0.06	0.00	0.00	0.00	0.00	−0.13	−0.13
impervious land	−0.67	−0.19	0.00	−0.01	0.00	0.12	−0.74	−0.86
total	14.69	116.08	0.50	0.00	0.15	0.15	131.57	-
transferred out	2.00	−7.07	0.49	0.00	0.15	0.03	-	−4.40

). The transition between different land types can cause fluctuations in the carbon storage of ecosystems. The increase in forest, grassland, and shrubland areas is the main reason for the rise in carbon storage in the Lijiang River Basin. The stored carbon dioxide into the atmosphere when converting land from high-carbon-density types to low-carbon-density types can be released. The significant conversion of cropland to forest is the main reason for the increase in carbon storage, with an increase of 2.73 × 10⁶ t of carbon storage. However, the carbon storage gain from cropland being converted to forest is far outweighed by the carbon loss caused by the conversion of forest to cropland. The carbon storage loss from the conversion of forest land to cropland amounted to 6.66 × 10⁶ t, accounting for 94.23% of the total carbon loss. The Lijiang River Basin has undergone rapid urbanization, with the conversion of other land types to impervious land leading to a total carbon storage loss of 0.86 × 10⁶ t.

3.3.2. The Impact of Karst Landforms on Carbon Storage

The Lijiang River Basin is primarily composed of karst and non-karst landforms. The area of non-karst landforms is approximately 3432.46 km², while the area of karst landforms is about 2393.66 km², accounting for 41% of the total area of the basin. There is a clear difference in carbon storage distribution between karst and non-karst landforms (Figure 6). The average carbon storage density in non-karst areas is significantly higher than that in karst areas. Additionally, the total carbon storage in non-karst regions accounts for approximately two-thirds of the total carbon storage in the Lijiang River Basin. Over the past 20 years, both the carbon storage density and total carbon storage in karst and non-karst areas have shown a downward trend. In karst regions, carbon storage decreased by 3.85%, while in non-karst areas, it decreased by 2.57%. From 2016 to 2021, both the carbon storage and carbon density in the karst and non-karst regions of the basin showed a slight increase. The increase in carbon storage was 0.51 × 10⁶ t for karst regions and 0.08 × 10⁶ t for non-karst regions. The average increase in carbon density was 0.19 t/hm² for karst areas and 0.02 t/hm² for non-karst areas. This improvement can be attributed to the integrated management and restoration projects, such as the mountain-water-forest-cropland-lake-grass approach, which focused on protecting the vulnerable ecological areas of karst landforms within the basin. These efforts have effectively enhanced the carbon storage in karst regions.

3.4. Carbon Storage Scenario Prediction of the Lijiang River Basin

3.4.1. Model Accuracy Validation

The land use data for 2021 were obtained using the Markov–PLUS model, and the simulation results were validated using the FOM statistical tool, with the value of FOM approaching 1 typically indicating higher simulation accuracy. The results showed that the Kappa coefficient was 0.82, and the overall accuracy was 0.92, indicating that the Markov–PLUS model has high simulation accuracy and is well suited for the Lijiang River Basin.

3.4.2. Land Use Change Prediction

From 2021 to 2041, the changes in land use types in the Lijiang River Basin show significant differences under various scenarios (

Table 6. Land use change in different situations (10⁶ t).

Types	Inertial Scenario		Urban Development Scenario		Ecological Protection Scenario
	2021–2031	2031–2041	2021–2031	2031–2041	2021–2031	2031–2041
cropland	19.79	13.19	13.01	44.10	−21.88	−21.37
forest	−25.97	−13.06	−18.54	−44.74	18.04	17.49
shrub	−0.55	−0.27	−0.55	−1.07	−0.32	−0.33
grassland	−0.10	−0.20	−0.07	−0.18	−0.07	−0.08
waters	4.26	0.50	3.59	1.08	2.16	2.14
impervious land	2.57	−0.16	2.56	0.78	2.11	2.16

). In the inertial scenario, the area of cropland increases 32.98 km². Other land types experience varying degrees of reduction except for slight increases in water bodies and impervious land areas. In the urban development scenario, the area of forest land decreases by 55.88 km², while the area of impervious land increases by 22.93 km². In the ecological protection scenario, the area of forest land increases, with an increase of 35.52 km². The areas of cropland, shrubland, and grassland experience slight decreases, while the area of forest land increases significantly. The areas of impervious land and water bodies show slight increases.

3.4.3. Carbon Storage Change Prediction

The carbon density in the Lijiang River Basin ranges from 0.0189 to 27.8658 t/hm². In the inertial scenario, the carbon storage in the basin shows a steady decline (Figure 7). The spatial pattern of carbon storage generally shows higher values in the northeast and southeast, while the central and western regions have lower values (Figure 8). By 2031 and 2041, the carbon storage will be 144.53 × 10⁶ t and 144.27 × 10⁶ t, respectively. From 2021 to 2041, the cumulative carbon storage loss will be 0.84 × 10⁶ t, indicating that the basin will be a carbon source in the future. However, the decline trend will slightly slow down between 2031 and 2041. In the urban development scenario, the rate of carbon storage decline in the Lijiang River Basin accelerates over time. From 2021 to 2041, a total of 1.31 × 10⁶ t of carbon storage will be lost. In the last decade, the carbon storage loss will account for 67.94% of the total loss. In the ecological protection scenario, the carbon storage in the Lijiang River Basin shows an increasing trend. From 2021 to 2041, the cumulative carbon sequestration will be 0.61 × 10⁶ t, marking a transition from a carbon source to a carbon sink, with carbon sequestration capacity gradually recovering. The areas with increased carbon density are distributed, mainly in the middle and lower reaches of the Lijiang River Basin, such as Tanxia Town and Lingchuan Town. The areas with decreased carbon density are primarily located in the upper reaches. This suggests that implementing ecological protection strategies contributes to the ecological sustainability of the Lijiang River Basin.

4. Discussions

(1)

Land Use and Carbon Storage Changes in the Lijiang River Basin

We analyzed the changes in average path length and clustering coefficient of land use transitions in the Lijiang River Basin from 2001 to 2021 by using the complex network approached. The results show that the average path length of the land transfer network remained below 1.3, indicating frequent and rapid transitions between different land use types. The ratio between input and output of impervious land remained at a high level throughout the period, suggesting a continued trend of expansion. These patterns reflect the accelerating urbanization process and the increasing intensity of land use conflicts. Notably, changes in impervious land, shrubland, and water bodies highlight the tension between urban development and ecological conservation. Detailed land transition analysis revealed that impervious land have continuously encroached upon cropland and forest land. As a result, high carbon density cropland and forest land have been converted into low-carbon-density construction land, leading to a steady decline in total carbon storage, which is 3.03%, representing a total loss of approximately 2.73 × 10⁶ t. This finding is consistent with the results reported by Zhang et al. [43]. Further statistical analysis comparing karst and non-karst geomorphological zones revealed differing rates of carbon loss: 3.85% in karst areas and 2.57% in non-karst areas. Spatial analysis indicated that most carbon storage loss occurred in the middle and lower reaches of the basin, where karst geomorphology is predominantly distributed. This supports the conclusion that carbon storage in karst regions is more susceptible to external disturbances. These findings align with previous studies. For example, Lin et al. [44] reported that ecological governance in Guilin has lagged behind the pressures imposed by environmental degradation. Similarly, Wei et al. [45] found that the loss of carbon storage in the Lijiang River Basin was largely due to forest land conversion driven by tourism development and urban construction.

(2)

Impact of Different Development Scenarios on Carbon Storage

Three development scenarios were established for simulation: the inertial scenario, the urban development scenario, and the ecological protection scenario. The simulation results indicate that different development pathways lead to distinct spatial land use patterns and carbon storage outcomes. Under both the baseline and urban development scenarios, carbon storage in the Lijiang River Basin is projected to decline further. This decline is primarily attributed to the reduction in forest land and the expansion of impervious land. In the inertial scenario, forest land is predominantly converted into cropland. In contrast, under the urban development scenario, both forest and croplands are more frequently transformed into impervious land. From a land use transition probability perspective, land conversion is more likely to be influenced by the surrounding land use types. Since existing impervious land are mainly distributed in the central part of the basin, forest and croplands in these areas are more vulnerable to being converted, leading to a substantial reduction in carbon storage capacity. Moreover, in the latter decade of the simulation period, the rate of forest carbon loss increases significantly under the urban development scenario. Conversely, the forest area expands by 35.53 km² in the ecological protection scenario, a result of stringent forest and water-body conservation measures and restrictions on the expansion of impervious land. Given that forest land possesses the highest carbon sequestration capacity among all land categories, total carbon storage in the basin under the ecological protection scenario is projected to increase by 1.45 × 10⁶ t compared to the baseline scenario and by 2.16 × 10⁶ t relative to the urban development scenario. The findings of He et al. [46] also suggest that restricting the conversion of high-carbon-density land types such as forest, grassland, and cropland into low-carbon-density construction land can effectively mitigate the decline in terrestrial ecosystem carbon storage and contribute to its enhancement. Moreover, the implementation of an ecological protection-first strategy has been widely recognized as an effective approach to mitigating carbon storage loss. Policies such as reforestation, grassland restoration, and the establishment of natural forest conservation areas have been shown to significantly enhance regional carbon sink capacity. For instance, the study by Deng et al. [47] demonstrated that the carbon storage service in Zichang County on the Loess Plateau has improved as a result of reforestation and grassland restoration efforts, with potential carbon sequestration value expected in the future. Similarly, Li et al. [48] reported that reforestation initiatives have enhanced carbon sink functions in northwestern China. In this case, we argue that enhancing terrestrial ecosystem carbon storage in the Lijiang River Basin is essential to maintaining its strategic role as an ecological barrier in Guangxi. To this end, we should act on the principles of prioritizing resource conservation and environmental protection. Specifically, we recommend reducing development intensity in the middle and lower reaches of the basin, reinforcing the ecological barrier in the northern region, and improving forest resource management and forest quality through targeted interventions to enhance ecosystem structural integrity. Additionally, stronger control measures should be implemented in karst landscapes, which would not only improve future carbon storage in the Lijiang River Basin but also provide valuable insights for ecological management in other regions with sensitive or unique geomorphological features.

(3)

Research Limitations and Future Perspectives

In this paper, we systematically investigated the LUCC trends in the Lijiang River Basin using complex network analysis, quantified the spatiotemporal evolution of carbon storage over the past two decades through the InVEST model, and predicted future land use patterns and corresponding carbon storage changes under multiple scenarios under integrated Markov and PLUS models. The integrated approach effectively addressed common limitations of conventional models, particularly in mitigating simulation-driven overfitting and optimizing dominant land type transitions. However, two notable limitations warrant attention: First, while nineteen driving factors were considered in the PLUS model, the study did not thoroughly investigate the underlying mechanisms through which these driving factors influence both LUCC processes and carbon storage dynamics. Second, the scenario projections failed to incorporate specific climate change simulations from advanced climate models such as CMIP6 [49], potentially affecting the reliability of long-term predictions under climate-sensitive scenarios. Furthermore, the limitation of the InVEST model is that it assumes a static carbon density across land use categories throughout the projection period, which oversimplifies carbon cycling processes. Considering that forest land dominates the landscape, and vegetation carbon storage demonstrates significant variability with forest age and type in the Lijiang River Basin, future studies should refine model parameters by incorporating forest age–class differentiation (e.g., young-growth, mid-aged, pre-mature, mature, and over-mature stands [50]) and forest-type classification specific to the vegetation composition of basin, including evergreen broad leaf, deciduous broad leaf, evergreen coniferous forests, and shrublands [51,52]. Such multidimensional parameter optimization could substantially enhance the model’s capacity to capture vegetation-mediated carbon dynamics.

5. Conclusions

(1)

We have identified that the forest land, cropland, and impervious land are the key land types in the Lijiang River basin by utilizing complex networks. Over prolonged periods, the Lijiang River Basin has exhibited frequent and rapid interconversion among different land use types. Notably, the terrestrial ecosystem demonstrates an inverse trend between clustering coefficient variation and average path length changes. Spatially, land use transitions in the middle and lower reaches occur through more direct and intensive pathways, concurrently fostering localized high density development zones. This reflects the increasing pressure that human activities are exerting on critical ecological spaces.

(2)

The carbon storage of Lijiang River Basin in 2001, 2006, 2011, 2016, and 2021 are 149.39 × 10⁶ t, 147.97 × 10⁶ t, 146.38 × 10⁶ t, 144.28 × 10⁶ t, and 144.87 × 10⁶ t, respectively, which reveal an overall declining trend with a cumulative reduction of 4.52 × 10⁶ t. Spatially, carbon storage demonstrated marked heterogeneity, characterized by a northeast high and southwest low distribution pattern that strongly aligned with land use configuration.

(3)

Under the inertial development scenario, the cropland area increased by 32.98 km², while all other land types, except for minor expansions in water bodies and construction land, experienced varying degrees of reduction. The urban expansion scenario triggered a 55.88 km² loss of forest land and 22.93 km² gain in construction land. In contrast, the ecological protection scenario facilitated forest land recovery with a net increase of 35.52 km².

(4)

The carbon storage in the Lijiang River Basin by 2041 is estimated at 144.27 × 10⁶ t under the inertial scenario, 143.80 × 10⁶ t under the urban development scenario, and 145.72 × 10⁶ t under the ecological protection scenario. The ecological priority scenario exhibits a 1.45 × 10⁶ t and 2.16 × 10⁶ t advantage over the inertial and urban development scenarios, respectively. These results demonstrate that adopting ecological protection strategies, regulating conversions from high-carbon-density land classes to low-carbon-density types, and curbing unregulated urban expansion can effectively enhance the carbon sequestration capacity in the basin.