Spatiotemporal Evolution of Carbon Storage and Driving Factors in Major Sugarcane-Producing Regions of Guangxi, China

Abstract

Objectives: The major sugarcane-producing regions of Guangxi represent a critical agricultural zone in China. Investigating the mechanisms of land use change and carbon storage dynamics in this area is essential for optimizing regional ecological security and promoting sustainable development. Methods: Employing the land use transfer matrix, the InVEST model and the Geodetector model to analyze carbon storage changes and identify key driving factors and their interactive effects. Results: (1) From 2011 to 2022, Guangxi’s major sugarcane-producing regions experienced significant land use changes: reductions in cultivated land, grassland and water bodies alongside expansions of forest, bare land and construction land. (2) The total carbon storage in Guangxi’s major sugarcane-producing regions has increased from 2011 to 2018 by 0.99%, representing 1627.03 and 1643.10 million tons, while it has decreased by 0.1% in 2022 (1641.47 million tons) compared to 2018. (3) Cultivated land proportion and forest coverage rate were the primary drivers of spatial heterogeneity, followed by average slope and land urbanization rate. (4) Interaction analysis revealed strong synergistic effects among cultivated land proportion, forest coverage rate, NDVI and average slope, confirming multi-factor control over carbon storage changes. Conclusions: Carbon storage in the Guangxi sugarcane-producing regions is shaped by land use patterns and multi-factor interactions. Future strategies should optimize land use structures and balance urbanization with ecological protection to enhance regional carbon sequestration.

1. Introduction

Terrestrial ecosystem carbon storage plays a pivotal role in global carbon cycling [1], with its substantial carbon sequestration capacity and dynamic regulatory mechanisms representing an economically viable and ecologically sustainable approach to climate change mitigation [2]. As a critical nexus between atmospheric and geological carbon pools, the stability of terrestrial carbon reservoirs directly influences the equilibrium of global climate systems [3]. Land use transitions have emerged as the second-largest anthropogenic carbon emission source after fossil fuel combustion [4]. Land use transition refers to the spatiotemporal conversion between different land use types driven by both human activities and natural processes [5], while carbon reservoirs represent storage compartments for carbon elements within terrestrial ecosystems, encompassing vegetation carbon pools [6], soil carbon pools [7], and other components. Distinct land use types exhibit varying carbon storage capacities, and land use/land cover (LULC) changes, such as forest conversion to cropland or grassland, can significantly disrupt the balance of ecosystems, climate, and carbon cycling processes [8]. Under rapid urbanization, significant cropland loss and construction land expansion [9] have profoundly disrupted forest ecosystem integrity [10], agroforestry system functions [11], and agricultural activities [12]. These changes subsequently alter vegetation carbon sequestration capacity and soil carbon input/output balance, exerting substantial impacts on regional carbon storage dynamics. Consequently, investigating the mechanisms linking land use transitions and carbon pool responses will not only enhance understanding of terrestrial ecosystems’ role in global carbon cycling but also facilitate the development of scientific climate change adaptation strategies and promote sustainable ecosystem management.

Early research on carbon storage predominantly depended on forest resource inventories [13] and soil survey data [14], concentrating on the impacts of singular land use changes such as deforestation and afforestation, employing biomass expansion factor methods to estimate static carbon storage. However, such studies were constrained by data resolution limitations in quantifying the composite effects of multi-category land use transitions and dynamic carbon pool processes. With recent advancements in remote sensing technology and ecological modeling, researchers have begun integrating multi-source data, including Landsat imagery [15] and soil carbon density grids [16], utilizing land use transition matrices to quantify land category conversion trajectories while combining various approaches: the CENTURY model based on biogeochemical cycle principles [17], the CA–Markov model focusing on spatial simulation of land use evolution [18], and the InVEST model for integrated ecosystem multifunction assessment [19]. Concurrently, numerous studies have employed multi-model coupling to achieve complementary advantages, such as combining the 3-PGS model with the Bookkeeping model [20], overcoming the limitations of single models in characterizing carbon cycle processes and predicting terrestrial ecosystem carbon storage changes. The PLUS model [21] can integrate land expansion simulation with spatial competition mechanisms and has been used for land use change prediction in complex terrains; the IBIS model [22] can simulate vegetation dynamic carbon density under different LUCC or climate scenarios; the FLUS model [23] can simulate land use evolution pathways under various policy or development scenarios; when combined with the InVEST model through localization of carbon module parameters [24], it enhances regional carbon storage estimation accuracy and strengthens multidimensional analytical capabilities for land use transition and carbon pool response mechanisms. In terms of research scale, macro-level studies primarily focus on national [25], provincial administrative regions [16] or typical urban agglomerations [23], analyzing the macro-level impacts of land management on carbon storage, while meso-scale studies concentrate on typical agricultural areas [7] or key forestry carbon sink regions [19].

Current research has achieved some progress in understanding carbon cycling within cash crop production areas. For jute cultivation zones, studies primarily focus on the depletion effects of continuous cropping on soil carbon pools [26]. Rubber plantation research emphasizes the impacts of area expansion and tree age on carbon storage [27] while tea plantation studies mainly concentrate on soil organic carbon stock estimation and analysis of driving factors behind spatial variations [28]. In the field of sugarcane cultivation carbon sinks, existing research predominantly based on field experiments analyzes the impacts of agricultural practices like fertilization [29], expanded planting areas [30], and leaf burning [31] on soil carbon pools; measures sugarcane’s carbon sequestration capacity through biomass calculations [32]; and calculates the carbon emissions arising from agricultural inputs and mechanical inputs during sugarcane production [33]. Forestry carbon sink research emphasizes regional forest carbon storage estimation [34] or explores the carbon sequestration enhancement benefits of afforestation projects [35]. Sugarcane production regions, characterized by intensive agricultural activities and urbanization [36], still present knowledge gaps regarding the comprehensive impacts of land use transitions on regional carbon pools, which limits exploration of their carbon sequestration potential.

The major sugarcane-producing regions of Guangxi, located in South China, encompass core production areas including Nanning, Chongzuo, and Laibin, accounting for over 58% of China’s total sugarcane cultivation area and serving as the nation’s pivotal sugar industry base [37]. As a critical gateway connecting the China–ASEAN Free Trade Area, Guangxi’s sugarcane sector holds dual significance for both national sugar security strategies and regional ecological security coupled with low-carbon development [38]. Recent years have witnessed dramatic land use transformations driven by urbanization and agricultural restructuring, with intensive conversions among forest, cultivated, and construction lands exerting substantial impacts on regional carbon storage, necessitating urgent scientific assessment of carbon response mechanisms. This study integrates 30m resolution Landsat data with the InVEST model to systematically elucidate spatiotemporal response patterns of carbon pools to land use transitions during 2011–2022, while employing Geodetector to quantitatively identify key driving factors. The findings aim to provide scientific support for optimizing land resource allocation and enhancing carbon sequestration capacity, facilitating Guangxi’s dual objectives of agricultural sustainability and carbon neutrality.

2. Materials and Methods

2.1. Study Area

The major sugarcane-producing regions of Guangxi encompass 46 counties/districts (Figure 1), covering approximately 12.18 million hectares (51.26% of Guangxi’s total area), located in western South China (106–110° E, 20–25° N) and bordering Wuzhou to the east, the Beibu Gulf to the south, Yunnan to the west, and Guizhou to the north. The terrain exhibits a distinct northwest-high, southeast-low gradient, dominated by mountains and hills (>70% of total area) with limited plains and terraces [39], while karst landscapes span 70 counties, constituting 41% of Guangxi’s territory [40]. The region features a subtropical monsoon climate with warm temperatures (mean annual ~22.5 °C) [41], high humidity (frost-free period > 330 days) [42], and concentrated summer rainfall (~1800 mm annual precipitation) [41] with ~1500 h annual sunshine [43].

Guangxi, China’s premier sugarcane production region, cultivates approximately 1 million ha of sugarcane, representing 65% of the nation’s total cultivation area [44]. Data from the Guangxi Statistical Yearbook reveals the combined sugarcane acreage across 46 key producing counties was 963,200 ha (2011), 923,800 ha (2014), 785,700 ha (2018), and 760,700 ha (2022), with an 858,300 ha average, underscoring these core regions’ decisive contribution. Elite cultivars like Guitang 42, renowned for high yield, superior sucrose content, and stress resilience, now dominate plantings and account for over 60% of productivity gains [45]. In 2014, Guangxi initiated pilot projects for high-yield, high-sugar sugarcane base construction, promoting scaled operations, improved varieties, mechanized production, and modernized irrigation [46]. The elevation of Guangxi’s sugarcane industry to national strategic status in 2015 facilitated concentrated investments in policies, capital, and technologies [47], consolidating its leading position in China’s sugar industry.

2.2. Data Sources

The research dataset comprises land use, natural factors, and socioeconomic data spanning the period 2011–2022. The land use data features an overall accuracy of 80% with a spatial resolution of 30 m × 30 m, projected in the GCS_WGS_1984 coordinate system. Detailed data are presented in

Table 1. Types of study data and their sources.

Data Type	Data Name	Year(s)	Data Source
Land use data	Land use type	2011, 2014, 2018, 2022	Remote Sensing Information Processing Institute (http://irsip.whu.edu.cn/ (accessed on 27 March 2025))
Natural factors	Average slope	2022	Resource and Environment Data Center, CAS (http://www.resdc.cn/ (accessed on 8 April 2025))
	Forest coverage rate	2011, 2014, 2018, 2022
	DEM	2022	Geospatial Data Cloud (http://www.gscloud.cn/ (accessed on 8 April 2025))
	NDVI	2011, 2014, 2018, 2022
	Annual sunshine hours	2011, 2014, 2018, 2020	China Meteorological Data Network (http://data.cma.cn/ (accessed on 8 April 2025))
	Mean annual temperature	2011, 2014, 2018, 2022
	Annual precipitation
Socioeconomic factors	Sugarcane planting area		Guangxi Bureau of Statistics/Statistical Yearbook (http://tjj.gxzf.gov.cn/ (accessed on 20 March 2025))
	Sugarcane yield
	Population
	Primary industry GDP
	General public budget expenditure
	Number of industrial enterprises above designated size
	Land urbanization rate
	Road network density	2013, 2014, 2018, 2022	National Geographic Information Resource Directory Service System (http://www.webmap.cn/ (accessed on 8 April 2025))
	Nighttime light index	2011, 2014, 2018, 2022	Resource and Environment Data Center, CAS (http://www.resdc.cn/ (accessed on 11 April 2025))
	Distance to adjacent cities	2022

.

2.3. Methods

The research methodology begins with preprocessing Landsat satellite imagery combined with natural and socioeconomic driving factors, enabling comprehensive analysis of land use spatiotemporal dynamics. Subsequent application of the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model facilitates precise quantification of regional carbon stocks. Geographic detector analysis then reveals the principal mechanisms governing carbon storage variations.

2.3.1. Land Use Transfer Matrix

The land use transfer matrix [48] serves as a quantitative analytical method for examining spatiotemporal transition patterns of regional land use types. This methodology establishes a two-dimensional matrix of land use conversions between different time periods, systematically elucidating the evolutionary characteristics of land use structure and spatial transformation trajectories, thereby providing scientific evidence for optimizing land resource allocation and ecological conservation. The computational formula is presented below:

$$S_{mn}=\begin{array}{cccc}{S}_{11}& S_{12}& \dots & S_{1i}\\ S_{21}& S_{22}& \dots & S_{2i}\\ \dots & \dots & \dots & \dots \\ S_{i1}& S_{i2}& \dots & S_{ii}\end{array}$$

(1)

The formula is expressed as: S represents land area (km²); S_mn denotes the area converted from land use category m to category n between the initial and final study periods (m = 1, 2, …, i; n = 1, 2, …, i), where i indicates the total number of land use types.

2.3.2. InVEST Model

This study employed the Carbon Storage and Sequestration module of the InVEST model [19,49] to estimate carbon storage in Guangxi’s major sugarcane-producing regions. Based on four-phase land use/cover data, the model calculated carbon storage across different time periods.

The carbon storage calculation formula is as follows:

$$C_{{\text{\_}}_{total}}=C_{{\text{\_}}_{above}}+C_{{\text{\_}}_{below}}+C_{{\text{\_}}_{soil}}+C_{{\text{\_}}_{dead}}$$

(2)

where C_{_total} represents the total ecosystem carbon storage; C_{_above} denotes aboveground vegetation carbon storage; C_{_below} indicates belowground biomass carbon storage; C_{_soil} refers to soil carbon storage; and C_{_dead} represents dead organic matter carbon storage. Considering the relatively minor contribution of dead organic matter to regional carbon cycling [50,51], this component was excluded from the accounting framework in this study.

Total carbon storage by land use type in the study area can be calculated by combining carbon density data with land use data:

$$C_{\text{\_}total}={\sum }_{j=1}^i{C}_j\times A_j\left(j=\mathrm{1,2},\dots ,i\right)$$

(3)

where C_{_total} represents the total ecosystem carbon storage in the study area (t); C_j denotes the carbon storage per unit area of the j-th land use/cover type (t/km²); A_j indicates the spatial distribution area of the j-th land use type (km²); and i represents the total number of land use categories in the classification system.

The carbon density data were obtained from published academic research in relevant fields. Within the same climate zone, carbon density values for similar land use types show minimal variation [52]. Therefore, this study selected reference carbon density values from adjacent regions with comparable conditions, and further refined the data based on the study area’s actual characteristics and land use classification system. Specific carbon density values are detailed in

Table 2. Carbon density of various land use types in the study area (t/hm²).

LULC_Name	C_above	C_below	C_soil	C_dead	Source
Cultivated land	13.57	2.65	47.4	1	[53,54]
Forest	58.3	14.58	96	3.5	[54]
Grass	3.01	13.53	60	1	[55]
Water	2.8	2.4	0	0	[55]
Barren	3.4	0	31.4	0	[53,54,55]
Construction land	11.45	0.93	31.4	0	[54]

.

2.3.3. Spatial Autocorrelation Analysis

Spatial autocorrelation analysis is suitable for measuring whether the distribution of spatial variables is agglomerative. Global and local spatial autocorrelation can well reflect the relationships between geographical objects [56]. The Global Moran’s I focuses on evaluating the overall spatial autocorrelation characteristics of the entire study area, while the Local Moran’s I decomposes the global statistics into each spatial unit, enabling the identification of local spatial clustering areas and outliers [57].

The formula for global spatial autocorrelation is as follows:

$$Global Moran’s I={\displaystyle \frac{n{\sum }_{i=1}^n\sum _{j=1}^n{W}_{ij}\left(x_i-\mu \right)\left(x_j-\mu \right)}{({\sum }_{i=1}^n{\sum }_{j=1}^n{W}_{ij}){\sum }_{i=1}^n{\left(x_i-\mu \right)}^2}}$$

(4)

where n represents the number of study units; x_i and x_j denote the carbon storage values of units i and j, respectively; μ is the mean carbon storage value; and W_ij stands for the spatial weight value. The value range of Global Moran’s I is [−1, 1], when I > 0, it indicates positive spatial autocorrelation; when I < 0, it indicates negative spatial autocorrelation; and when I = 0, it means the spatial distribution is random [58]. The significance of the index is determined by the Z-test. If |Z| > 1.96 (p < 0.05), it can be considered that there is significant spatial autocorrelation [59].

The study area is divided at the county level, and the LISA map of local spatial autocorrelation is used to represent the agglomeration characteristics of carbon storage in local spaces. The formula for local spatial autocorrelation is as follows:

$$Local Moran’s I_i={\displaystyle \frac{X_i-\mu }{{\sum }_{i=1}^n{X}_i-{\mu }^2}}{\sum }_{j=1}^n{W}_{ij}(X_i-\mu )$$

(5)

The value range of Local Moran’s I_i is [−1, 1]. I_i > 0 indicates the existence of spatial clustering, among which High–High (HH) clustering refers to high-value units surrounded by high-value units (hotspot areas), and Low–Low (LL) clustering refers to low-value units surrounded by low-value units (coldspot areas); I_i < 0 indicates the existence of spatial anomalies, among which High–Low (HL) anomaly refers to high-value units surrounded by low-value units, and Low–High (LH) anomaly refers to low-value units surrounded by high-value units [60].

2.3.4. Geodetector

The Geodetector [61] is a statistical tool based on spatial heterogeneity theory, primarily used to analyze the spatial differentiation characteristics of geographical elements and their driving mechanisms. Its fundamental premise states that if a driving factor significantly influences the target variable, its spatial distributions should exhibit strong coupling.

This study selected 17 driving factors across five categories for geographical detection based on the previous literature, incorporating regional geographical location, climatic conditions, and economic development while ensuring data accessibility. The factors include climate [62,63] (annual sunshine hours, mean annual temperature, annual precipitation), topography [62] (average slope), vegetation [64] (NDVI, forest coverage rate), land use [65,66] (proportion of cultivated land, proportion of sugarcane planting area), and socioeconomic factors [62,64] (sugarcane yield per unit area, population size, primary industry GDP, general public budget expenditure, number of industrial enterprises above designated size, land urbanization rate, nighttime light index, road network density, and distance to adjacent cities), covering both natural environmental elements and human activities.

The analysis employed two functional modules of the Geodetector: (1) the single-factor detection module calculates the q-value (range [0, 1]), where values closer to 1 indicate stronger explanatory power of the factor for spatial differentiation [67]; (2) the interaction detection module compares q-value differences between factor combinations (X1 ∩ X2) and individual factors, classifying interaction effects into five types (e.g., nonlinear enhancement and bivariate enhancement) (

Table 3. Interaction mode of independent variables on dependent variables.

Description	Interaction
q (X1 ∩ X2) < Min (q (X1), q (X2))	Weaken, nonlinear
Min (q (X1),q (X2)) < q (X1 ∩ X2) < Max (q (X1), q (X2))	Weaken, uni-
q (X1 ∩ X2) > Max (q (X1), q (X2))	Enhance, bi-
q (X1 ∩ X2) = q (X1) + q (X2)	Independent
q (X1 ∩ X2) > q (X1) + q (X2)	Enhance, nonlinear

) to reveal synergistic mechanisms among driving factors [68]. The specific algorithm is as follows:

$$q=1-{\displaystyle \frac{\sum _{h=1}^L{N}_h{\sigma }_h^2}{N{\sigma }^2}}=1-{\displaystyle \frac{SSW}{SST}}$$

(6)

$$SSW={\sum }_{h=1}^L{N}_h{\sigma }_h^2; SST=N{\sigma }^2$$

(7)

In the model, h = 1, 2, …, L represents the stratification of variable Y or factor X; N_h and N denote the number of sample units in the h-th stratum and the total number of units in the study area, respectively; σ_h and σ represent the variances of Y-values in the h-th stratum and the entire region. Here, SSW indicates the sum of within-stratum variances, while SST represents the total variance of the entire region.

2.3.5. Land Use/Cover Classification

This study adopted the current land use classification (GB/T 21010-2017) [69] and utilized ArcGIS 10.8 software to classify land use types from remote sensing images, generating four-period land use distribution maps (2011, 2014, 2018, and 2022) for major sugarcane-producing regions of Guangxi. The land was categorized into six types: cultivated land, forest land, grassland, water bodies, bare land, and construction land (

Table 4. Classification of land use types.

Land Use Type	Description
Cultivated land	It refers to land used for growing crops, including mature cultivated land, newly reclaimed land, rotation land, fallow land, as well as land for agro-fruit intercropping, agro-forestry complexes, etc., covering types such as paddy fields, irrigated fields and dry land.
Forest	It refers to land where arbors, shrubs and bamboos grow, as well as coastal mangrove land, including natural forests and artificial forests.
Grass	It refers to land mainly covered by herbaceous plants (with a coverage rate of ≥5%), including natural grasslands, artificial pastures and shrub grasslands, etc.
Water	It refers to natural land waters (rivers, lakes, etc.) and land for water conservancy facilities, such as reservoirs, ditches, etc.
Barren	It refers to unused or hard-to-use land, including sandy land, bare rock, saline-alkali land, etc.
Construction land	It covers urban land (built-up areas of large, medium and small cities and county towns), rural residential areas (independent rural settlements), as well as other construction land such as factories, mines and transportation roads.

).

Figure 2 shows the sugarcane planting area distribution across counties in the study region for the years 2011, 2014, 2018, and 2022. The data reveal that Chongzuo and Laibin consistently remained the most significant production areas. This spatial distribution pattern reflects, to some extent, the correspondence between cultivated land classification in land use systems and agricultural production activities, providing crucial contextual support for subsequent analysis of carbon stock variations in cultivated areas.

3. Results

3.1. Evolution of Land Use Patterns

3.1.1. Spatiotemporal Characteristics of Land Use

Table 5. Area and percentage of land use types in the study area during 2011–2022.

Land Use Types	2011		2014		2018		2022
	Area/km2	Percent/%	Area/km2	Percent/%	Area/km2	Percent/%	Area/km2	Percent/%
Cultivated land	37,138.581	30.483	36,329.046	29.818	35,394.517	29.051	35,731.644	29.328
Forest	81,982.671	67.290	82,648.001	67.836	83,518.456	68.551	83,272.032	68.349
Grass	80.771	0.066	83.938	0.069	64.929	0.053	45.320	0.037
Water	1532.788	1.258	1541.544	1.265	1436.712	1.179	1216.842	0.999
Barren	0.599	0.0005	0.561	0.0005	1.038	0.0009	2.186	0.0018
Construction land	1098.976	0.902	1231.296	1.011	1418.735	1.164	1566.362	1.286

shows the area of land use types in Guangxi’s major sugarcane-producing regions during the study period. From 2011 to 2022, forest and cultivated land dominated the regional land use pattern and were widely distributed. As shown in Figure 3 and Table 5, forest remained the largest landscape type, consistently accounting for over 67.29% of the total area with a distinct altitudinal distribution pattern. Other land use types ranked by area proportion were construction land > water bodies > grassland > barren land. Cultivated land area decreased significantly with a net reduction of 1406.94 km² (3.79%) over 11 years; grassland area shrank by 43.89% and water bodies decreased by 20.61%, showing continuous decline trends; forest area increased by 1.57%, reflecting the implementation effects of ecological protection policies; construction land expanded most rapidly (42.53%), indicating rapid urbanization; while barren land showed a large percentage increase (265.26%), its absolute growth was limited (1.587 km²). Notably, a significant spatial coupling existed between construction land expansion and cultivated land reduction, a phenomenon likely related to urban expansion.

Land use dynamics reflect the magnitude and rate of land type changes within a specific time period in the study area [41].

Table 6. Changes in land use type area and dynamic degree in the study region during 2011–2022.

Land Use Types	2011→2014		2014→2018		2018→2022		2011→2022
	Area Change/km2	Dynamic/%	Area Change/km2	Dynamic/%	Area Change/km2	Dynamic/%	Area Change/km2	Dynamic/%
Cultivated land	−809.535	−2.180	−934.529	−2.572	337.127	0.952	−1406.937	−3.788
Forest	665.329	0.812	870.455	1.053	−246.424	−0.295	1289.361	1.573
Grass	3.166	3.920	−19.009	−22.646	−19.609	−30.201	−35.452	−43.892
Water	8.756	0.571	−104.832	−6.800	−219.870	−15.304	−315.946	−20.612
Barren	−0.038	−6.316	0.477	85.072	1.148	110.668	1.588	265.263
Construction land	132.321	12.040	187.439	15.223	147.627	10.406	467.386	42.529

presents the area changes and dynamic degree data of different land use types in the study region during 2011–2022. Each land use type exhibited distinct change characteristics across different phases. During 2011–2014, construction land showed a dynamic degree of 12.040%, indicating significant area expansion that reflected active regional development activities. Grassland expanded moderately with a dynamic degree of 3.920%, while cultivated land and barren land decreased with dynamic degrees of −2.180% and −6.316%, respectively. From 2014 to 2018, cultivated land, grassland and water bodies continued to decrease with dynamic degrees of −2.572%, −22.646% and −6.800%, demonstrating substantial external pressures on these land types. Forest land and construction land maintained growth with dynamic degrees of 1.053% and 15.223%, respectively, with construction land showing particularly strong expansion that indicated accelerating urbanization. Barren land transitioned from decrease to increase with a dynamic degree of 85.072% (relatively small absolute change but large proportional change). During 2018–2022, cultivated land area increased with a dynamic degree of 0.952%; forest land, grassland and water bodies continued to decrease with dynamic degrees of −0.295%, −30.201% and −15.304%, respectively; construction land still expanded but at a slower rate (dynamic degree 10.406%); and barren land persistently increased with a dynamic degree of 110.668%.

3.1.2. Land Use Transfers Analysis

Figure 4 and

Table 7. Land use transfer matrix of study area from 2011 to 2014.

Land Use Types	Cultivated Land	Forest	Grass	Water	Barren	Construction Land	Total
Cultivated land	33,836.438	3089.237	20.660	68.477	0.000	123.770	37,138.581
Forest	2417.180	79,546.946	13.461	0.000	0.000	5.085	81,982.671
Grass	20.522	4.449	48.755	1.997	0.095	4.954	80.771
Water	54.877	7.370	0.920	1464.638	0.110	4.874	1532.788
Barren	0.031	0.000	0.142	0.001	0.356	0.069	0.599
Construction land	0.000	0.000	0.000	6.431	0.000	1092.544	1098.976
Total	36,329.046	82,648.001	83.938	1541.544	0.561	1231.296	121,834.386

,

Table 8. Land use transfer matrix of study area from 2014 to 2018.

Land Use Types	Cultivated Land	Forest	Grass	Water	Barren	Construction Land	Total
Cultivated land	32,599.504	3505.519	20.101	31.734	0.036	172.153	36,329.046
Forest	2649.155	79,989.571	2.083	0.155	0.001	7.037	82,648.001
Grass	20.743	13.019	41.493	1.337	0.644	6.701	83.938
Water	125.034	10.346	1.102	1398.217	0.038	6.808	1541.544
Barren	0.045	0.000	0.151	0.003	0.320	0.042	0.561
Construction land	0.036	0.000	0.000	5.267	0.000	1225.994	1231.296
Total	35,394.517	83,518.456	64.929	1436.712	1.038	1418.735	121,834.386

and

Table 9. Land use transfer matrix of study area from 2018 to 2022.

Land Use Types	Cultivated Land	Forest	Grass	Water	Barren	Construction Land	Total
Cultivated land	31,920.381	3288.267	12.456	28.718	0.068	144.627	35,394.517
Forest	3544.106	79,962.305	2.745	0.616	0.000	8.684	83,518.456
Grass	23.869	5.747	29.592	0.644	0.946	4.130	64.929
Water	243.119	15.710	0.321	1169.828	0.526	7.209	1436.712
Barren	0.100	0.000	0.205	0.004	0.647	0.082	1.038
Construction land	0.070	0.003	0.000	17.033	0.000	1401.629	1418.735
Total	35,731.644	83,272.032	45.320	1216.842	2.186	1566.362	121,834.386

provide a comprehensive overview of land use transitions in the study area during 2011–2022. Figure 4 visually demonstrates the consistent mutual conversion between cultivated land and forest land over this 11-year period, while also revealing frequent conversion of cultivated land to construction land, indicating pronounced urban expansion trends.

Table 7 provides a comprehensive overview of land use changes in the study area during 2011–2014. During this three-year period, cultivated land showed a significant reduction with 3302.143 km² transferred out and 2492.608 km² transferred in. Forest land exhibited substantial expansion, with 2435.726 km² transferred out but 3101.055 km² transferred in. Grassland area increased slightly (32.016 km² out vs. 35.183 km² in), as did water (68.151 km² out vs. 76.906 km² in). Barren land experienced minimal changes (0.243 km² out vs. 0.205 km² in), remaining essentially stable. Construction land showed marked expansion (6.431 km² out vs. 138.752 km² in).

Table 8 reveals the land use transition patterns during 2014–2018. This period saw dramatic reduction in cultivated land (3729.542 km² out vs. 2795.013 km² in), contrasting with significant forest land expansion (2658.423 km² out vs. 3528.885 km² in). Both grass (42.445 km² out vs. 23.436 km² in) and water (143.328 km² out vs. 38.496 km² in) decreased, while barren land increased slightly (0.241 km² out vs. 0.718 km² in). Construction land continued rapid expansion (5.303 km² out vs. 192.741 km² in).

Table 9 demonstrates the land use dynamics during 2018–2022, showing notable trend reversals. Cultivated land area increased (3474.136 km² out vs. 3811.263 km² in), breaking its previous declining trend. Forest showed a slight reduction (3556.150 km² out vs. 3309.727 km² in), contrasting with earlier expansion. Grass (35.337 km² out vs. 15.728 km² in) and water (266.884 km² out vs. 47.014 km² in) continued decreasing, while barren land increased (0.391 km² out vs. 1.539 km² in). Construction land maintained strong expansion (17.105 km² out vs. 164.732 km² in).

3.2. Spatiotemporal Characteristics of Carbon Storage

3.2.1. Carbon Storage Distribution Patterns

From 2011 to 2022, the major sugarcane-producing regions of Guangxi maintained moderate overall carbon storage levels, with distribution closely tied to land use types (Figure 5). The spatial heterogeneity was pronounced, exhibiting a northwest-high, southeast-low gradient. The western and northern mountainous areas formed high-value zones, where favorable hydrothermal conditions supported robust forest growth. These regions’ inaccessibility limited human disturbance, preserving extensive forest cover that consistently accounted for >67% of the study area, primarily in higher-altitude northwestern areas. Forests’ high carbon density and sequestration capacity dominated these zones. Moderate-value areas were widespread, dominated by cultivated land (>29% coverage) and grassland, where crop growth and vegetation provided measurable carbon sequestration. The central and southeastern regions represent low carbon storage zones, characterized by flat topography, abundant water resources, and relatively dynamic economic development. These areas contain substantial construction land and water bodies, both exhibiting low carbon density and consequently weak carbon sequestration capacity. During the period 2011 to 2022, the carbon storage level of each district and county was relatively stable on the whole, but there were also changes in some areas. Tianlin County, Rongshui County, Huanjiang County and Du’an County remained at high levels of carbon storage, while six districts and counties, including Xixiangtang District, Jiangnan District and Gangbei District, remained at low levels of carbon storage, and Qintang District and Qinbei District fluctuated.

3.2.2. Characteristics of Carbon Storage Changes

The InVEST model’s Carbon module calculations (

Table 10. Changes of carbon storage in the study area from 2011 to 2022 (×10⁴ t).

Land Use Types	2011		2014		2018		2022
	Carbon Stock	Percent/%	Carbon Stock	Percent/%	Carbon Stock	Percent/%	Carbon Stock	Percent/%
Cultivated land	23,627.563	14.522	23,112.537	14.147	22,517.990	13.705	22,732.470	13.849
Forest	138,452.341	85.095	139,575.950	85.434	141,045.974	85.841	140,629.813	85.673
Grass	61.822	0.038	64.246	0.039	49.696	0.030	34.688	0.021
Water	79.705	0.049	80.160	0.049	74.709	0.045	63.276	0.039
Barren	0.208	0.00013	0.195	0.00012	0.361	0.00022	0.761	0.00046
Construction land	481.132	0.296	539.062	0.330	621.122	0.378	685.753	0.418
Total	162,702.772	100	163,372.150	100	164,309.852	100	164,146.761	100

) reveal that forests consistently dominated carbon storage contributions in Guangxi’s major sugarcane-producing regions during 2011–2022, maintaining >85% shares: 1384.523 million tons (85.095%) in 2011, increasing to 1406.298 million tons (85.673%) by 2022. The period 2011–2018 witnessed extensive conversion of cultivated land to forests, representing a shift from low-carbon-density to high-carbon-density land uses. Conversely, 2018–2022 saw significant forest conversion to cultivated and construction land—a reverse transition pattern. These land use changes directly impacted carbon fluxes, with high-to-low carbon density conversions triggering carbon release [50], and vice versa.

Cultivated land’s carbon storage declined from 236.276 million tons (14.522%) to 225.180 million tons (13.705%) during 2011–2018, before recovering slightly to 227.325 million tons (13.849%) by 2022. Grassland and water bodies showed consistent carbon storage reduction, while barren and construction land exhibited modest increases despite small proportional contributions. The carbon storage ranking across all years remained: forest > cultivated land > construction land > water bodies > grassland > barren land.

Total regional carbon storage fluctuated moderately between 1627.028 and 1643.099 million tons during the 11-year period. Forest expansion enhanced carbon sequestration, while urban expansion at the expense of cultivated land emerged as the primary factor undermining regional carbon storage capacity [70]. Notably, forests and cultivated land collectively contributed >99.5% of carbon sequestration, underscoring the critical importance of understanding land use transition mechanisms for ecological security and sustainable development in these key sugarcane-producing regions.

3.2.3. Spatial Autocorrelation Analysis of Carbon Storage

Global Moran’s I (GMI) indices of carbon storage in the major sugarcane-producing regions of Guangxi in 2011, 2014, 2018, and 2022 were 0.469, 0.423, 0.426, and 0.437, respectively (

Table 11. GMI indices of carbon storage in major sugarcane-producing regions of Guangxi from 2011 to 2022.

Year	GMI	Z	p
2011	0.469	4.900	0.001
2014	0.423	4.451	0.001
2018	0.426	4.506	0.001
2022	0.437	4.578	0.001

), all greater than 0, indicating a significant positive spatial autocorrelation in carbon storage. The Moran’s I scatter plot shows a particularly obvious spatial association agglomeration in the first quadrant (high–high clustering) and the third quadrant (low–low clustering) (Figure 6).

The results of local spatial autocorrelation (LISA clustering) analysis (Figure 7) indicate that the spatial distribution patterns of carbon storage in each year are highly similar. Specifically, the carbon storage hotspots (high–high clusters) are mainly concentrated in the mountainous forest areas in the west and north, where the proportion of forest land is high and carbon density is large, serving as the core contributors to regional carbon storage. In contrast, the carbon storage coldspots (low–low clusters) are mainly distributed in the southeastern plains, where the proportion of construction land is high and continues to expand with low carbon density, making it a key region for improving carbon sequestration potential through optimized land use in the future.

3.3. Driving Factors of Carbon Storage Spatial Differentiation

3.3.1. Factor Detection Results

This study employed the Geodetector model to quantitatively analyze the dominant factors and interaction mechanisms governing carbon storage spatial differentiation, using county-level carbon storage as the dependent variable and 17 indicator factors as independent variables. Among the selected drivers, four factors failed the significance test (p-value ≥ 0.05): annual precipitation (X3), proportion of sugarcane planting area (X8), general public budget expenditure (X13), and road network density (X16). These non-significant factors were excluded from subsequent Geodetector analysis as their q-values became negligible [71].

Figure 8 demonstrates that the remaining factors all influenced carbon storage spatial differentiation to varying degrees. Taking 2011 as an example, the explanatory power (q-value) ranking was as follows: proportion of cultivated land area (X7) > forest coverage rate (X6) > NDVI (X5) > average slope (X4) > land urbanization rate (X10) > nighttime light index (X15) > mean annual temperature (X2) > primary industry GDP (X12) > annual sunshine hours (X1) > distance to adjacent cities (X17). The annual variations in q-value rankings reflect dynamic changes in factor influences.

During 2011–2022, the proportion of cultivated land area (X7) consistently ranked first, likely due to significant area fluctuations (Table 6) that persistently highlighted its role as a carbon pool carrier, with this impact being amplified by intensive utilization under policies like the “High-Yield High-Sugar” sugarcane base development. Forest coverage rate (X6) and NDVI (X5) maintained consistently high rankings with alternating fluctuations, potentially linked to concurrent forest loss and afforestation activities that caused woodland area changes [72], demonstrating vegetation’s crucial influence on carbon storage. After 2018, although the growth rate of construction land slowed and the q-value of land urbanization rate (X10) gradually decreased, the continued intensification of human activities (e.g., industrial production) [73] enhanced the explanatory power of the nighttime light index on carbon storage. The q-value of the number of industrial enterprises above designated size (X14) rose to top rankings post-2018, aligning with Guangxi’s increasing trend of industrial energy consumption and carbon emissions [73]. Essentially, these ranking changes directly manifested how multiple processes, including agricultural production, ecological conservation, urbanization and industrialization, alternately dominated the carbon cycle at different stages, providing quantitative evidence for identifying critical nodes in regional carbon management.

Note for Figure 8: Road network density data for 2011 were substituted with 2013 values. q represents explanatory power; X1: annual sunshine hours; X2: mean annual temperature; X3: annual precipitation; X4: average slope; X5: NDVI; X6: forest coverage rate; X7: proportion of cultivated land area; X8: proportion of sugarcane planting area (relative to total crop sown area); X9: sugarcane yield per unit area; X10: land urbanization rate (proportion of construction land); X11: population size; X12: primary industry GDP; X13: general public budget expenditure; X14: number of industrial enterprises above designated size; X15: nighttime light index; X16: road network density; X17: distance to adjacent cities. The same conventions apply hereafter.

3.3.2. Interaction Detection Results

The interaction detection results exhibited two patterns: bivariate enhancement and nonlinear enhancement, with all factor combinations’ interaction explanatory power q (Xi ∩ Xj) being significantly higher than individual factors’ independent effects, confirming that the spatial differentiation of carbon storage in the study area resulted from multiple factors. Temporal analysis (Figure 9a–d) showed that in 2011, the interaction between cultivated land proportion (X7) and forest coverage rate (X6), as well as primary land urbanization rate (X10), and number of industrial enterprises above designated size (X14) was most prominent (q-values of 0.995, 0.992, and 0.992, respectively). In 2014, average slope (X4) and forest coverage rate (X6), as well as cultivated land proportion (X7) and land urbanization rate (X10), demonstrated the strongest explanatory power (q-values of 0.995 and 0.992, respectively). In 2018, the most significant interactions were between cultivated land proportion (X7) and mean annual temperature (X2), cultivated land proportion (X7) and average slope (X4), as well as NDVI (X5) and land urbanization rate (X10) (the q-values of all three were 0.989). By 2022, the interaction between and cultivated land proportion (X7) and nighttime light index (X15) reached its peak (q = 0.994).

The dynamic interactions among factors further demonstrated that the spatial variation of carbon storage in Guangxi’s sugarcane-producing regions is closely associated with regional development policies and ecological environment. Factors including average slope, NDVI, forest coverage and cultivated land proportion consistently demonstrated high explanatory power in interaction effects across multiple periods, indicating that vegetation coverage and agricultural production activities were core elements driving carbon storage changes. Meanwhile, human activity indicators such as land urbanization rate, number of industrial enterprises above designated size, nighttime light index, and distance to adjacent cities also significantly influenced carbon storage spatial patterns through interactions, and the synergistic effects between these factors and natural elements collectively shaped the dynamic evolution characteristics of regional carbon storage.

4. Discussion

This study reveals that during 2011–2022, the carbon storage in Guangxi’s major sugarcane-producing regions exhibited a distinct northwest-high and southeast-low spatial pattern, where the western and northern mountainous areas with favorable hydrothermal conditions for forest growth formed stable natural and plantation forest zones, while the southeastern plains experienced significant carbon loss due to intensive agricultural development and urbanization, resulting in lower carbon storage that closely correlated with regional forest distribution patterns—consistent with findings from comparable studies in other regions [74,75,76]. Forests increased their carbon sequestration contribution from 85.095% in 2011 to 85.673% in 2022, confirming their dominant role in carbon balance and aligning with previous vegetation carbon sink research [77,78]. The four-phase carbon storage data (2018 > 2022 > 2014 > 2011) showed an initial increase followed by a decline, likely resulting from synergistic effects of policy interventions and land use transitions such as (1) the China Ecological Conservation Redline (ECR) strategy proposed in 2011 and nationally implemented by 2017 [79], which restricted forest-to-construction land conversion; (2) the 2014 launch of the second phase Grain-for-Green Program [80] combined with rocky desertification control [81] which improved regional ecology; and (3) concurrent “High-Yield High-Sugar” sugarcane base construction that promoted organic fertilizers and soil-testing formulated fertilization technologies to mitigate agricultural carbon loss [82]. Additionally, some abandoned farmlands were converted to forests due to agricultural conditions and demographic changes [83], becoming another carbon accumulation pathway. Notably, substantial cultivated-to-construction land transitions caused carbon losses, providing quantitative evidence for balancing the tripartite objectives of food security, ecological protection and urbanization.

The Geodetector results identified cultivated land proportion (q ≥ 0.976) and forest coverage rate (q ≥ 0.948) had the greatest impact on carbon storage. Spatial analysis revealed that forest-dominated northwestern areas accounted for 85.673% of carbon storage, while southeastern regions with predominant cultivated and construction lands showed lower carbon density, confirming land use type’s deterministic influence [84]. Anthropogenic factors played substantial roles: human activities accelerated LULC changes [85], with rapid economic development and population density in the central and southeastern regions driving construction land expansion and carbon storage decline [86]; GDP and population density exhibited threshold effects on carbon storage [87], while agricultural intensification exacerbated carbon loss [88]. The significant differences in land use patterns and human interventions between northwestern and southeastern areas collectively shaped regional carbon storage heterogeneity.

Interaction detection demonstrated bivariate enhancement, where any two factors’ combined explanatory power exceeded individual effects. For instance, the 2022 interaction between forest coverage rate and cultivated land proportion reached q = 0.993, while cultivated land proportion and nighttime light index achieved q = 0.994, aligning with the “ecological-economic coupling amplification effect” theory [62]. This was exemplified by Guangxi’s 2016 Ecological Conservation Redline regulations that simultaneously constrained construction land and enhanced forest carbon sinks, reflecting policy-mediated factor coordination. The human-dominated driving mechanisms and factor coupling effects revealed by Geodetector necessitate multidimensional carbon management strategies beyond uniform approaches [62]. In rapidly urbanizing zones, strict ecological redline enforcement and science-based urban growth boundaries should be implemented with low-carbon infrastructure [89]. In sugarcane cultivation areas, optimized “High-Yield High-Sugar” sugarcane base policies should promote low-carbon farming [12]. For ecological protection, northwestern forests require enhanced compensation and quality improvement [90], while southeastern karst areas need integrated rocky desertification control to restore carbon sequestration functions [91], alongside exploring forestry-agriculture carbon synergy mechanisms. Coordinated land use policies and ecological measures will collectively facilitate regional green transition and sustainable development.

This study quantitatively assessed the carbon effects of land use changes in Guangxi’s major sugarcane-producing regions, providing new insights for tropical agricultural carbon pool dynamics. However, the following several limitations should be noted: (1) The spatiotemporal patterns and driving mechanisms of carbon storage require longer-term monitoring—the 2011–2022 study period was insufficient to fully capture the impacts of land use policy adjustments on carbon storage, and future LUCC scenarios were not simulated; (2) While focusing on land classification and natural/socioeconomic indicators, ecological factors such as microbial decomposition were not included—future studies should expand the indicator system to incorporate more socioeconomic and ecological parameters for a more comprehensive understanding of carbon storage drivers; (3) The interaction detection approach has limited capacity to characterize complex nonlinear relationships—further research is needed to elucidate the integrated pathways through which multiple factors influence carbon storage; (4) Due to the limitations of the accuracy and spatial resolution of currently available remote sensing data, it has not been possible to provide a refined spatial distribution map of sugarcane-producing areas in the study region, with only statistical information on sugarcane planting areas presented. This limitation has resulted in a lack of intuitiveness in the analysis of the correlation characteristics between the spatial pattern of sugarcane planting and carbon storage, making it difficult to accurately reflect their spatial coupling relationship. To address these limitations, we propose four optimization strategies: (1) Developing multi-scenario simulation frameworks aligned with Guangxi’s medium- and long-term development plans to enable long-term carbon storage prediction and targeted management [92]; (2) Establishing a multidimensional assessment system encompassing “soil-vegetation-human activity” by integrating soil carbon monitoring, agricultural policy adjustments, and labor migration data [93]; (3) Creating a multi-scale monitoring system combining Sentinel-2 and UAV data [94] in response to IPCC’s call for improved agricultural carbon accounting methodologies, and exploring coupled modeling frameworks that integrate socioeconomic and ecological process parameters to enhance the predictive accuracy of land use policy effects on carbon dynamics, thereby providing actionable guidance for similar regions to achieve “dual carbon” goals; (4) For future research, we will specifically address the limitations in data on sugarcane-producing areas. We plan to integrate high-resolution remote sensing images with field survey data to map the actual spatial distribution of sugarcane cultivation in the study area, presenting its specific geographical distribution pattern. Meanwhile, combining phenological data with field investigations, we will construct spatiotemporal dynamic maps that include the sugarcane planting cycle (such as key stages like sowing, growth, and harvesting), integrating the spatial distribution characteristics of sugarcane cultivation with the laws of its growth cycle. We will analyze the differences in the impact of sugarcane at different spatial locations and different planting stages on carbon storage, aiming to more accurately reveal the coupling mechanism between the two, thereby providing more spatially targeted scientific basis for regional carbon management and the sustainable development of the sugarcane industry.

5. Conclusions and Recommendations

The important findings that can be concluded from the study are as follows:

(1)

From 2011 to 2022, the total carbon storage in Guangxi’s major sugarcane-producing regions measured 1627.03 (2011), 1633.72 (2014), 1643.10 (2018), and 1641.47 (2022) million tons, respectively, exhibiting a distinct northwest-high and southeast-low spatial pattern. The western and northern mountainous areas formed high-value carbon zones due to favorable forest growth conditions and abundant woodland resources, while the flat southeastern regions with intensive economic activities showed relatively lower carbon storage owing to higher proportions of construction land and water bodies.

(2)

Land use transitions significantly influenced carbon storage dynamics. Forests consistently contributed over 85% of total carbon storage, playing a pivotal role in maintaining regional carbon stocks. Cultivated land showed fluctuating carbon storage corresponding to its initial decrease and subsequent recovery, whereas grassland and water bodies demonstrated consistent declines. Although barren land and construction land accounted for minor proportions, their marked increasing trends reflected growing anthropogenic impacts on regional carbon cycles.

(3)

Geodetector analysis identified cultivated land proportion (q ≥ 0.976) and forest coverage rate (q ≥ 0.948) as dominant factors governing carbon storage spatial differentiation. Other factors, including NDVI, average slope, and land urbanization rate, also demonstrate relatively high q-values exceeding 0.719. The dynamic ranking of q-values reflected the phased characteristics of each driving factor’s influence, as well as revealed that agricultural intensification, ecological conservation, urbanization and industrialization processes alternately dominated the carbon cycle at different stages.

(4)

Interaction detection demonstrated that the interactions between cultivated land proportion, forest coverage rate, NDVI and average slope had the strongest impact on carbon storage changes. All factor interactions showed greater explanatory power than individual factors acting alone, confirming that multi-factor coordination drives carbon storage changes.

(5)

Policy recommendations emphasize optimizing land use structures during urbanization/agricultural development; strengthening forest conservation and ecological restoration; controlling construction land expansion; and establishing dynamic balance mechanisms between economic growth and ecological protection by leveraging factor interactions to enhance carbon sequestration capacity and support China’s dual carbon goals.