Understanding how land-use change reshapes terrestrial carbon storage (CS) is important for China’s dual-carbon goals in rapidly urbanizing regions. Taking Beijing as a megacity case, we integrated GIS-based land-use analysis with the InVEST CS module to quantify CS dynamics from 1980 to 2020. We then used the Geodetector to identify dominant drivers of spatial heterogeneity and coupled a Markov chain with the PLUS model to simulate land use and CS in 2030 under alternative development pathways. Results show that cultivated land, forest land, and construction land dominated Beijing’s land-use structure. From 2000 to 2020, cropland decreased by 2,240.25 km2, while forest land and construction land increased by 177.22 km2 and 2,130.58 km2, respectively. Total CS followed a “decline–rebound” trajectory, decreasing from 2.9462 × 108 t (1980) to 2.7828 × 108 t (2010) and then rising to 2.8352 × 108 t (2020). Spatially, CS exhibited significant positive spatial autocorrelation throughout the study period (Global Moran’s I = 0.883–0.900), with high–high clusters concentrated in the mountainous northwest and low–low clusters mainly distributed across the built-up plains in the center and southeast, as further confirmed by LISA and hot/cold spot patterns. Geodetector results indicate that topography and climate dominate CS heterogeneity, with slope showing the strongest explanatory power (q = 0.62). Scenario simulations suggest that CS would increase more under a carbon-sink priority scenario (+2.71%) than under an inertial-development scenario (+1.76%) by 2030, highlighting the value of ecological-priority land-use strategies for carbon-sink enhancement.
Beijingcarbon stock estimationgeodetectorInVEST modelland-use dynamicsMarkov-PLUS simulationThe author(s) declared that financial support was not received for this work and/or its publication.section-at-acceptanceLand Use DynamicsIntroduction
Global climate change has intensified, and many countries are facing challenges such as greenhouse gas emissions, ecosystem degradation, and unbalanced land use (LU) (Lei et al., 2023). Addressing global warming often relies on two simple steps, cutting human emissions and increasing ecosystem carbon uptake. These ideas are reflected in the Paris Agreement and the United Nations Sustainable Development Goals (SDGs). Within the SDG system, Goal 13 focuses on climate action, Goal 11 focuses on sustainable cities, and Goal 15 focuses on protecting and using terrestrial ecosystems in a sustainable way. Together, these goals show that healthy urban ecosystems and optimized LU structures are central to global climate governance (Seto et al., 2012; Filho et al., 2023; Arora and Mishra, 2024).
Carbon peaking and carbon neutrality are now seen as essential pathways to control climate risks. China has articulated a two-stage national climate goal-emissions are to reach their maximum before 2030, followed by the realization of carbon neutrality by mid-century (2060) (Liu et al., 2025). Terrestrial ecosystems are vital for reaching these goals, as plant carbon sequestration and soil carbon storage (CS) together form the basic structure of the land carbon cycle (Jiao et al., 2024; Pan et al., 2024). In this study, carbon storage (CS) means the amount of carbon stored in terrestrial ecosystems at a given time. It is a carbon stock. It is usually reported by summing major carbon pools, such as aboveground biomass, belowground biomass, dead organic matter, and soil carbon (Nowak and Crane, 2002; Penman et al., 2003).
Land-use and land-cover change (LUCC) strongly affects regional CS. LUCC affects CS because land categories have different carbon densities. When ecological surfaces—such as croplands, forests, or grasslands—are replaced by built-up areas, regional carbon stocks often drop. Rapid urbanization also increases energy consumption and greenhouse gas emissions through population agglomeration, building expansion, and infrastructure construction, creating an amplified coupling effect between LUCC and carbon emissions (Gong et al., 2022; An et al., 2024; Gu et al., 2025). Therefore, identifying how LUCC influences urban CS is fundamental for understanding urban carbon balance and designing low-carbon planning strategies.
Beijing is a typical example because it is China’s capital and a rapidly developing megacity. Since the late Reform and Opening-up period, rapid outward growth of built-up areas has been accompanied by a steady reduction in ecological land cover, which in turn has markedly weakened the region’s carbon-sequestration capacity (Kang et al., 2025). In response, Beijing has launched some of the country’s most ambitious ecological restoration initiatives - among them the Three-North Shelterbelt Project, the Beijing-Tianjin Sandstorm Source Control Project, and extensive plain afforestation - thereby creating substantial new carbon sinks within the metropolitan area (Zhang and Zhang, 2024). This “urban expansion plus ecological restoration” trajectory makes Beijing a useful case for testing how land transitions and policy actions reshape CS in a megacity. Beijing is not chosen simply because it has experienced land-use change, which is common across cities. It is chosen because it concentrates two processes that are central to urban carbon governance: rapid land conversion driven by megacity growth, and sustained ecological interventions implemented under strong planning and environmental regulation (Bai et al., 2018). These co-occurring processes make it possible to examine how development pressure and ecological protection jointly shape carbon storage in a single urban system. In addition, Beijing has a clear mountain–plain structure, which is typical of many North China cities and creates strong contrasts in land cover and ecosystem conditions across space (Jin et al., 2025). Therefore, the insights from Beijing are expected to be informative for other rapidly urbanizing regions that face similar trade-offs between built-up expansion and ecological conservation (Jiang et al., 2020).
Although many studies have explored LUCC impacts on CS at national and regional scales (An et al., 2024; Zhang H. et al., 2024), studies focusing on megacities are still limited. Most existing studies examine either long-term CS patterns or specific restoration projects, but fewer studies connect long-term LUCC, driver mechanisms, and scenario-based futures in one framework for megacities (Dixon et al., 1994). As global urbanization accelerates, megacities have become key nodes of global carbon cycle change and climate governance. Building a transferable urban CS prediction framework can deepen scientific understanding of rapidly urbanizing regions such as Beijing and support the coordinated achievement of SDG 11, SDG 13, and SDG 15 (Keenan et al., 2015).
Building on this context, the present research takes Beijing as an empirical setting and combines long-term LU monitoring, carbon-storage estimation, and scenario-based simulation to analyze how LUCC reshaped carbon stocks between 1980 and 2020, while also forecasting possible future trajectories under varying development pathways. The objectives are threefold: (1) to clarify how CS has evolved across space and time in the interplay between rapid urban growth and ecological conservation efforts; (2) to measure the extent to which specific LU conversions contribute to shifts in regional carbon sequestration capacity; and (3) to provide feasible pathways for low-carbon urban development through scenario simulations. The findings offer a scientific basis for understanding carbon sink mechanisms in megacities and provide policy references for other rapidly urbanizing regions.
Literature reviewResearch on CS assessment using the InVEST framework
CS within terrestrial ecosystems represents a fundamental component of how the Earth system moderates climate and circulates carbon. It helps mitigate the greenhouse effect and supports carbon neutrality goals. Forests, grasslands, and wetlands are important carbon sinks, and forests are especially significant. Although forests account for only around thirty percent of Earth’s land area, they nonetheless hold close to half of the carbon present in terrestrial ecological systems (Dixon et al., 1994; Keenan et al., 2015). Since the Industrial Revolution, human activities have changed LU patterns through large-scale deforestation, agricultural expansion, and rapid urbanization. Human-driven transformations have reshaped ecosystem composition and functioning, creating growing risks to the stability of carbon budgets within terrestrial environments (Yuan X. et al., 2025). Gaining a clearer understanding of CS and the ways in which LU dynamics influence it is thus essential for assessing regional carbon-sink potential and guiding environmentally informed LU planning.
Research on CS has progressed through three general stages. In the late 20th century, only a few studies used field measurements and traditional statistical methods to estimate forest CS. In the early 21st century, attention to climate change increased, and studies began to estimate CS at the global scale. For example, Fang et al. (2007) estimated the carbon sequestration of China’s main terrestrial vegetation from 1981 to 2000 and found that ecosystem carbon sequestration offset about 20.8%–26.8% of industrial carbon emissions during that period. During the last 20 years, with advances in remote sensing and modeling, scholars have applied models such as the CASA model, InVEST model, and various coupled models to calculate CS at multiple scales, examine differences across regions and ecosystems, and analyze carbon sequestration benefits under historical land use change (LUC) (Nayak et al., 2010; Babbar et al., 2021; Sarathchandra et al., 2021).
The InVEST model-an analytical framework designed to appraise ecosystem-service values and land-management impacts-has become a standard tool in related research fields. Within this framework, the CS and Sequestration component classifies CS into four categories, belowground biomass, aboveground biomass, soil carbon, and dead organic matter carbon pools (Ismaili Alaoui et al., 2023). InVEST is often used together with LUC analysis or scenario simulation to estimate LUCC impacts on CS. For instance, Lin et al. (2022) used InVEST and the PLUS model to evaluate CS change in the karst region of Guizhou from 2010 to 2020 and simulated spatial patterns for 2030 and 2050. Their results showed that ecological restoration policies significantly increased CS. Zhu et al. (2022) examined long-term LU transitions and CS dynamics across China’s coastal zone for the period 1980–2050. Their analysis showed that intensified urban growth fragmented areas with high carbon density, leading to notable CS decline, whereas ecological-protection scenarios helped mitigate part of this reduction.
The InVEST framework has likewise seen extensive adoption in research conducted internationally. Adelisardou et al. (2022) combined InVEST with scenario simulation to assess the impacts of LUCC on CS and sequestration value in the Jiroft Plain in Iran from 2000 to 2019, providing evidence for land-management decisions. Babbar et al. (2021) integrated Markov chain LU simulation with the InVEST to evaluate changes in carbon sink capacity under multiple LU scenarios in the Sariska Tiger Reserve in India.
Beyond Asia, InVEST-based CS assessment has also been applied in Europe and North America to support land management and conservation planning (Benez-Secanho and Dwivedi, 2021). These studies often combine InVEST with classified land-cover maps and policy-relevant scenarios to compare carbon outcomes under alternative land-management options (Adelisardou et al., 2022). They highlight two practical advantages of InVEST for decision support. First, the model is transparent and data-efficient when detailed process-based parameters are not available. Second, it provides spatially explicit CS maps that can be aligned with zoning or land-management units (Hamel et al., 2021).
At the same time, international work also points to shared limitations that affect result comparability across regions (Sharma et al., 2024). The most common issue is the use of fixed carbon-density values by land category, which can introduce uncertainty when carbon pools vary across climate zones, vegetation types, management intensity, and soil conditions (Lyu et al., 2019). Many studies therefore recommend sensitivity analysis, careful calibration of carbon densities, and clear reporting of carbon-pool assumptions when InVEST results are used for policy evaluation (Zhou H. et al., 2025).
Studies on CS drivers using the geodetector
The Geodetector can analyze spatial heterogeneity. Its core function is to test the spatial consistency between geographic variables and observed phenomena, and to identify and quantify their driving effects. Because spatial heterogeneity is a fundamental feature of geography, the Geodetector is widely regarded as an effective causal-analysis method. Since it was proposed by Wang et al., in 2010, the method has gained increasing global attention, with 3,123 published studies from 48 countries using it (Liang and Xu, 2023). This wide uptake indicates that the method has been tested across both developed and developing contexts and across multiple spatial scales (Chi et al., 2021).
The Geodetector can conduct both single-factor and interaction detection. It has been commonly applied to studies on climate impacts, environmental partitioning, population distribution, forest carbon monitoring, and urban livability (Zhang et al., 2022; Deng and Chen, 2025; Ni et al., 2025). In the field of LUC and CS, it has produced meaningful results. For example, Cai et al. (2024) evaluated CS patterns in the Pearl River Delta Urban Agglomeration by combining the PLUS model, the InVEST toolkit, and GeoDetector. Their results indicated that biophysical variables—including NDVI and terrain steepness—together with human-related factors such as population size and economic output (GDP), exerted substantial influence on CS variation. Their interactions showed a two-factor enhancement or nonlinear enhancement effect, which greatly improved the explanation of spatial carbon-storage patterns. Similarly, Chen et al. (2024) used an Optimal Parameter Geodetector to analyze county-level CS along the Daiyun Mountain region. They found that CS was jointly influenced by natural factors such as elevation, and annual temperature, and by socioeconomic factors such as population density. Moreover, interaction effects were stronger than single factors, highlighting the dominant role of multi-factor coupling in shaping carbon distribution.
Scenario simulation studies on future CS
In recent years, substantial research has examined LU evolution by probing the mechanisms driving LUCC and by constructing simulations of potential future trajectories. Existing LU modeling tools generally fall into two broad groups, one set—including approaches like the Markov chain model and various neural-network techniques—focuses on forecasting the magnitude of LU change, whereas another set—represented by cellular automata (CA) and the CLUE-S framework—concentrates on reproducing the spatial configuration of LU patterns.
To improve simulation accuracy, many scholars now integrate multiple models to combine their strengths. For example, Zhang et al. (2024c) linked the GMOP, PLUS and InVEST models to project LU evolution and CS dynamics in the Hexi Corridor up to 2035, and reported that ecological conservation scenarios produced markedly higher CS than business-as-usual conditions. Xu and Li (2024) combined the SD, PLUS and InVEST models to simulate LU change and CS in Nanjing between 2020 and 2040, showing that ecological protection and high-quality development pathways yielded greater gains in carbon sequestration than a natural-growth scenario. Using a PLUS-InVEST framework, Zhou J. et al. (2025) examined LU reconfiguration and CS change in the Yangtze River Delta from 2020 to 2030 and found that CS declined under all scenarios except balanced development, underscoring the importance of cultivated land conservation and ecological protection policies for climate mitigation. Moreover, Lei et al. (2024) integrated CA-Markov with InVEST to simulate multi-scenario LU patterns and CS trends on Hainan Island from 2000 to 2050, further confirming the practicality of coupling “future scenario simulation” with CS assessment for policy-oriented analysis.
Internationally, scenario-based LU simulation has been widely used to evaluate future ecosystem service outcomes, including carbon storage, under alternative urban-growth and conservation pathways (Pickard et al., 2017). Studies in Europe and North America often combine CA-based or CLUE-S–type models with ecosystem-service assessment tools to compare how compact growth, sprawl control, protected-area expansion, or agricultural retention may affect carbon outcomes (Sallustio et al., 2015). Work in developing regions also uses similar scenario logic, but it often focuses on rapid peri-urban conversion and the resulting carbon losses in fast-growing metropolitan corridors (Adelisardou et al., 2022).
Across these international studies, two shared practices are relevant for improving robustness (Guo et al., 2023). First, scenario narratives are usually tied to explicit planning constraints, such as strict protection for ecological land, limits on urban expansion, or priorities for restoration. Second, model validation is commonly reported using map-to-map agreement metrics (e.g., Kappa or related indices) before future projections are interpreted (Dong et al., 2024). These practices help reduce the risk that scenario results reflect model artifacts rather than plausible policy pathways (Marey et al., 2024).
Materials and methodsStudy area
Beijing is located in northern China, ranging from 115°20′E to 117°30′E and 39°28′N to 41°05′N. The municipality covers an area of approximately 16,410.54 km2. Beijing borders Tianjin to the east and is surrounded by Hebei Province on all other sides (Figure 1). It lies at the core of the Beijing–Tianjin–Hebei (BTH) economic region, and this position strengthens its role in regional development and land-use management. Beijing also shows a clear mountain–plain structure, with mountainous areas in the west and north and lowland plains in the center and southeast (Zhang et al., 2016; Pan et al., 2022). This spatial contrast is common in many cities in North China and creates strong differences in land cover and ecosystem conditions within one administrative unit (Zhang et al., 2016). These features make Beijing an appropriate study area for examining how land-use transitions relate to carbon storage under the joint influence of development pressure and ecological protection.
Map of Beijing.
Topographic map of a region divided into districts, labeled with abbreviations, showing elevation variation from low (−38 meters, gray) to high (2,289 meters, green), with district boundaries marked in white and a north arrow provided for orientation.Data sources and preprocessing
This research draws upon several datasets, including LU information and a set of driving variables (Table 1). The LU data originate from a 30 m–resolution global land-cover product issued by the Resource and Environment Science and Data Center (RESDC) (Chen et al., 2015). For analytical purposes, these data were reorganized into six LU classes.
Data source description.
Data type
Data name
Resolution
Year
Source
Base data
LU data
30 m
1980, 1990, 2000, 2010, 2020
http://www.resdc.cn/
Natural factors
DEM
30 m
2020
http://www.gscloud.cn/
NDVI
30 m
2020
https://doi.org/10.12199/nesdc.ecodb.rs.2021.012.
Precipitation
1 km
2020
http://www.resdc.cn/
Temperature
1 km
2020
Slope
30 m
2020
Derived from DEM
Socio-economic factors
GDP
1 km
2020
http://www.resdc.cn/
Population
1 km
2020
Accessibility factors
Railway distance
—
2020
http://www.openstreetmap.org/
Highway distance
—
2020
Water distance
—
2020年
The driving-factor dataset included four natural factors, two socioeconomic factors, and three accessibility factors. Slope information was extracted from a 30 m–resolution DEM obtained via the Geospatial Data Cloud, followed by spatial processing procedures (Farr et al., 2007). Data on major highways and national rail lines were sourced from the OpenStreetMap (Haklay and Weber, 2008), and Euclidean distance tools in ArcGIS were then used to generate indicators representing proximity to roads and railways.
Research methodsCS module in the InVEST framework
The InVEST framework is commonly employed to quantify ecosystem services and to inform land and resource management decisions. In this work, we use its CS component to track how terrestrial CS varies over time under different LU categories. Within the module, ecosystem CS is represented as the sum of four carbon pools, belowground/aboveground biomass carbon, dead organic matter/soil carbon. The amount of carbon per unit surface area is termed carbon density (Liang et al., 2024; Shi et al., 2024), and the overall CS of the study area is obtained by aggregating these pools according to Equations 3.3 and 3.4:Cj=Cj‐above+Cj‐below+Cj‐coll+Cj‐deadCtotal=∑j=1∞Cj×Sjwhere Cj denotes the overall carbon density associated with LU category j; Cj-above, Cj-below, Cj-soil, and Cj-dead correspond to the carbon densities of belowground/aboveground biomass, dead organic matter/soil for the same LU classy. Sj indicates the areal extent of LU type j; Ctotal refers to the aggregated CS for the entire study region.
Determination of carbon density
Carbon density refers to the carbon stored per unit area or per unit economic output (Nowak et al., 2013), and it is used to measure the concentration of carbon or the efficiency of carbon sequestration. In this study, the LUCC classification system was adopted, and LU categories were reclassified into six types. As a result, corresponding carbon-density indicators were generated from the four pools for each LU type.
To reduce the uncertainty introduced by using a single parameter source, we compiled carbon density values reported in previous studies for Beijing and then used the arithmetic mean for each land-use type and carbon pool. Specifically, the candidate values were extracted from Dang et al. (2025), Wang P. et al. (2024), Lutong et al. (2025), etc., which all applied the InVEST carbon storage module (often coupled with PLUS) in Beijing or its ecological conservation areas. The averaged carbon density values used in this study are reported in Table 2.
Carbon density value of each part of different LU types (t/ha).
LU type
Aboveground
Belowground
Soil
Dead organic matter
Total carbon density
Cultivated land
4.07
1.54
85.36
3.65
94.62
Forest land
65.55
74.93
142.74
3.01
286.23
Grassland
37.09
41.62
92.13
7.95
178.79
Water bodies
3.05
0.00
20.64
0.00
23.69
Construction land
3.44
2.96
25.44
0.10
31.93
Unused land
4.24
0.85
19.57
0.00
24.66
Geodetector
The Geodetector method is a spatial analytical tool designed to evaluate how strongly various determinants shape a particular geographic pattern. In this research, it is employed to pinpoint the factors that govern the spatial variability of CS within Beijing. Five biophysical variables, two socioeconomic indicators, and three measures of spatial accessibility are incorporated as explanatory inputs. The explanatory variables were selected to represent three complementary mechanisms that commonly shape CS patterns: biophysical constraints (topography and climate), human pressure and land management intensity (socioeconomic activity), and spatial accessibility that mediates where development and land conversion concentrate (distance-based proximity indicators). The technique operates by estimating the degree to which each factor-and combinations of factors-accounts for observed spatial differentiation through single-factor and interaction tests. The computation process follows Equation 3.5 (Cai et al., 2024):Q=1‐∑h=1L‐1nhσh2Nσ2
In this formula, L indicates the number of spatial divisions created for analytical purposes; σh2 refers to the variance of CS within subregion h; nh is the number of observations contained in that subregion; N denotes the total sample size for the entire study area, while σ2 represents the overall variance of CS across all samples. The Q statistic falls between 0 and 1, with larger values implying that the factor under examination provides greater explanatory strength for the spatial differentiation of CS.
Markov chain–based prediction
The Markov approach characterizes a probabilistic system in which the state at the following time step is determined solely by the present state, without any influence from earlier conditions (Kumar et al., 2014; Hamad et al., 2018). Among various Markov models, the Markov Chain is the most commonly used to describe state transitions within a system.
By extracting transition patterns from historical LU data, the Markov framework estimates the likelihood of future LU demand. These transition probabilities enable projections of how LU configurations may evolve under alternative development pathways, thereby offering valuable support for scenario-based LU planning and CS simulation.
Patch-generating land use simulation (PLUS) model
The PLUS model is a LUC simulation method designed to better represent spatial pattern evolution in LU transitions. The model integrates driving factors, land-transition probabilities, spatial features, and human activities. Compared with traditional models, PLUS has higher simulation accuracy, and its outputs can provide scientific support for regional LU planning and policy development.
The PLUS framework operates by identifying how various LU categories expand between two time points and then employing stochastic simulation techniques to determine the factors that drive these expansion processes. This allows the model to estimate development trends and occurrence probabilities of each LU type under various scenarios. Based on these probabilities, and combined with corresponding driving factors, future LU distributions can be simulated.
The PLUS framework is composed of two core components, a rule-mining module derived from analyzing LU expansion patterns, and a multi-type stochastic patch-generation Cellular Automata (CA) model that simulates the spatial evolution of different LU categories (Wang et al., 2023; Marey et al., 2024). The model structure is as follows:
This module transforms LU transition rules into a binary classification problem that can be solved using data mining methods. It uses stochastic algorithms to extract LU transition drivers by sampling expansion areas between two time points. From this process, the development probability of each LU type is obtained, along with the influence of driving factors on corresponding land types at each period. The random algorithm used in this strategy is expressed in Equation 3.6:Pi,kdX=∑n=1MIhnX=dMwhere P denotes the likelihood that LU category k appears at spatial unit i. A value of d = 1 signifies that competing LU classes transition into type k; whereas d = 0 indicates a shift from type k to another LU category. X refers to the vector of driving variables used in the model. I represents the indicator function derived from an ensemble of decision trees, and M corresponds to the total number of trees contributing to that ensemble.
The CA is a discrete spatiotemporal mathematical model used to simulate dynamic changes in complex systems. In LU simulation, the CA represents different LU types through cell states and interprets transition rules through changes in cell states. In this way, the global LU evolution can be simulated based on simple local rules across raster cells on a map.
To determine whether LUC within a raster cell are governed by the overall transition probability, the CA model follows the principle expressed in Equation 3.7:OPi,kd=1=Pi,kd=1×Ωi,kt×Dktwhere OPi,kd=1 expresses the overall likelihood that LU class k will appear at cell i during time t; Ωi,kt captures the neighborhood influence exerted by LU type k around location i at the same moment; Dkt reflects the persistence or inertia associated with LU category k at time t. The neighborhood influence is derived through the computation shown in Equation 3.8:Ωi,kt=concit‐1=kn×n‐1×wkwhere concit‐1=k refers to the count of grid cells classified as LU type k within an n × n neighborhood centered on location i during the previous time step (t-1); wk specifies the weight attributed to the neighborhood influence of each LU category.
Scenario setting
Based on the objectives of this study, LU simulation for Beijing in 2030 is carried out under two scenarios:
Inertial Development Scenario (IDS): The IDS is defined as a baseline pathway that extends the observed land-use transition regime from the historical period into the near future, without introducing additional policy constraints beyond the continuation of current practices. Under this scenario, land-transition probabilities are derived from historical LU transition matrices and are used to project the expected land-demand change for each LU category. The purpose of IDS is to provide a neutral reference for comparison, so that the incremental effects of policy-oriented constraints can be evaluated against a common baseline (Mutale and Qiang, 2024).
Carbon Sink Priority Scenario (CPS): The CPS introduces a “dual-carbon” policy orientation and explicitly aligns the scenario assumptions with Beijing’s future development planning and ecological protection requirements (e.g., the Beijing Urban Master Plan (2016–2035) and related ecological protection initiatives). This scenario emphasizes the protection and expansion of ecological land. In this case, land-transition rules restrict large-scale conversion of ecological land into construction land while increasing the expansion probability of forests and grasslands, with the goal of achieving higher carbon sequestration capacity. In other words, CPS represents a planning-guided pathway, while IDS serves as the baseline for benchmarking policy effects.
Parameters of the Markov-PLUS modelNeighborhood weights
The neighborhood-weight coefficient describes how readily each LU category tends to expand in space. It is normalized to the interval [0,1], classes assigned larger values are more prone to outward growth and, at the same time, less likely to be converted into other LU types. In this work, the simulation procedure is organized into two stages, a model validation stage and a subsequent forward-projection stage.
During validation, the LU pattern from 2010 is taken as the initial state to generate a simulated map for 2020, which is then compared with the observed 2020 LU distribution to evaluate the robustness and credibility of the model. The neighborhood-weight settings used in this step follow Zhang et al. (2024b), who calibrated these parameters with the FLUS model in representative hilly–mountainous regions (Liu et al., 2017) (Table 3). This choice helps reduce subjectivity in parameter tuning and provides a benchmark setting that has been validated in LU simulation practice (Wang et al., 2019).
Neighborhood weight setting for the model validation stage.
LU type
Cultivated land
Forest land
Grassland
Water bodies
Construction land
Unused land
Neighborhood weight
0.32
0.99
0.96
0.85
0.01
0.67
In the future prediction phase, neighborhood weights were determined by dimensionless processing based on LU data of Beijing (Table 4). The calculation is expressed in Equation 3.9:X′=X‐XminXmax‐Xminwhere X′ represents the normalized result produced through deviation standardization; X refers to the area associated with each LU category at both the beginning and end of the study period; Xmin is the smallest observed change in land area among all LU types, whereas Xmax corresponds to the largest area change across those categories. By linking neighborhood competitiveness to Beijing’s observed LU changes, this setting makes the forward simulations consistent with the local land expansion context and improves scenario interpretability.
Neighborhood weight setting for the future prediction stage.
Scenario
Cultivated land
Forest land
Grassland
Water bodies
Construction land
Unused land
IDS
0.27
0.99
0.95
0.79
0.01
0.62
CPS
0.18
0.99
0.87
0.68
0.01
0.56
Transition cost matrix
In LU simulation, conversions among different categories do not occur arbitrarily; rather, they are shaped by a set of biophysical, socioeconomic, and policy constraints. To represent these constraints quantitatively, LU models commonly employ a transition cost matrix, which specifies how feasible it is for one LU class to be transformed into another (Mutale and Qiang, 2024; Zhou et al., 2024). A value of one in the matrix signifies that a given conversion is allowable, whereas 0 indicates that such a transformation is prohibited or implausible.
Consistent with the treatment of neighborhood-weight parameters, this study establishes two separate transition-cost matrices—one tailored to the model-validation stage and another designed for future-scenario simulations (Tables 5 and 6). In the model-validation stage, we used a permissive matrix (all entries = 1) to avoid imposing additional subjective restrictions, so that the simulation can primarily be driven by the empirically learned transition rules and neighborhood effects; model performance is then assessed by comparing the simulated 2020 map against observations.
Transfer cost matrix for the model validation stage.
LU type
Cultivated land
Forest land
Grassland
Water bodies
Construction land
Unused land
Cultivated land
1
1
1
1
1
1
Forest land
1
1
1
1
1
1
Grassland
1
1
1
1
1
1
Water bodies
1
1
1
1
1
1
Construction land
1
1
1
1
1
1
Unused land
1
1
1
1
1
1
Transfer cost matrix for the future prediction stage.
LU type
IDS
CPS
Cultivated land
Forest land
Grassland
Water bodies
Construction land
Unused land
Cultivated land
Forest land
Grassland
Water bodies
Construction land
Unused land
Cultivated land
1
1
1
1
1
1
1
1
1
1
1
1
Forest land
1
1
1
1
1
1
0
1
1
1
0
0
Grassland
1
1
1
1
1
1
0
1
1
1
0
0
Water bodies
1
1
1
1
1
1
1
1
1
1
1
1
Construction land
1
1
1
1
1
1
1
1
1
1
1
1
Unused land
1
1
1
1
1
1
1
1
1
1
1
1
For the forward-projection stage, the IDS keeps the permissive conversion setting to represent a continuation of historical transition tendencies without introducing additional policy constraints. By contrast, the CPS encodes an ecological-priority constraint by restricting the conversion of high-carbon ecological land (forest land and grassland) into cropland, construction land, and unused land (matrix entries set to 0) (Yuxin et al., 2024; Yuan L. et al., 2025). This design operationalizes the scenario assumption that ecological land is more strictly protected under carbon-sink–oriented development, while still allowing conversions among other land categories where appropriate (Yuxin et al., 2024; Yuan L. et al., 2025).
Model validation
The Kappa coefficient is a statistical indicator used to evaluate the accuracy of classification models and is commonly applied in spatial simulation and LUC modeling for accuracy assessment (van Vliet et al., 2011; Wang et al., 2012). Unlike overall accuracy, the Kappa coefficient not only measures the agreement between predicted results and observed data but also accounts for agreement that may occur by chance. Its calculation is expressed in Equation 3.10:k=po‐pe1‐pewhere po represents the proportion of actual agreement, referring to the proportion of predictions that match the observed values; pe denotes the expected proportion of agreement that may occur by chance under a random distribution.
Based on the above principle of the Kappa coefficient and the parameter settings of the Markov-PLUS model, this study used LU data from 2010 to simulate the LU pattern in 2020 during the validation phase. The calculated Kappa coefficient of the simulation result is 86.66%, which falls into the category of “almost perfect agreement.” This indicates that the validation process meets the required accuracy, and thus the model and corresponding parameter settings can be reliably applied to multi-scenario LU prediction in this study.
ResultsSpatiotemporal changes in LU in Beijing from 1980 to 2020Temporal characteristics of LUC
Table 7 presents the changes in LU area and proportion for each LU type in Beijing, reflecting historical LU transition trends. Table 8 shows the LU dynamic degree, indicating the intensity of LU transitions over time. Based on the combined analysis of these datasets, Beijing experienced substantial changes in its LU pattern. The evolution of LU development can be divided into two main stages dominated by different trends: the rapid urbanization stage from 1980 to 2010 and the slower urbanization stage from 2010 to 2020.
The area and proportion of land types (1980–2020) (km2, %).
Data type
Year
Cultivated land
Forest land
Grassland
Water bodies
Construction land
Unused land
Area
1980
5,900.69
7,300.79
1,358.75
401.11
1,430.49
1.13
1990
5,850.82
7,312.22
1,364.3
397.13
1,467.36
1.13
2000
4,909.73
7,431.18
1,295.09
511.12
2,244.72
1.13
2010
3,796.65
7,316.73
1,112.42
325.6
3,839.87
1.7
2020
3,660.24
7,478.01
1,254.74
422.15
3,561.07
16.76
Percentage
1980
36
44.54
8.29
2.45
8.73
0.01
1990
35.69
44.61
8.32
2.42
8.95
0.01
2000
29.95
45.33
7.9
3.12
13.69
0.01
2010
23.16
44.63
6.79
1.99
23.42
0.01
2020
22.33
45.62
7.65
2.58
21.72
0.1
The LU dynamic degree of Beijing.
Data type
Year
Cultivated land
Forest land
Grassland
Water bodies
Construction land
Unused land
Area change
1980–1990
−49.87
11.43
5.55
−3.98
36.86
0
1990–2000
−941.09
118.96
−69.21
113.98
777.36
0
2000–2010
−1,113.07
−114.45
−182.67
−185.52
1,595.15
0.58
2010–2020
−136.42
161.28
142.33
96.55
−278.8
15.06
Dynamic degree
1980–1990
−0.08
0.02
0.04
−0.1
0.26
0.02
1990–2000
−1.61
0.16
−0.51
2.87
5.3
−0.03
2000–2010
−2.27
−0.15
−1.41
−3.63
7.11
5.09
2010–2020
−0.36
0.22
1.28
2.97
−0.73
88.33
To more intuitively illustrate LUC from 1980 to 2020, a bar chart of LU proportions was generated (Figure 2). The findings indicate that construction land expanded dramatically under the influence of accelerated urban development. Between 1980 and 2010, its extent rose from 1,430.49 km2 (8.73%) to 3,839.87 km2 (23.42%), representing a 168% increase. This substantial growth mirrors the surge in infrastructure needs and urban spatial expansion that accompanied population increases and economic transformation in the decades following China’s Reform and Opening-up.
Bar chart of LU area by category in Beijing from 1980 to 2020.
Stacked bar chart showing land use area from 1980 to 2020 with categories: cultivated land, woodland, grassland, water, construction land, and unused land; cultivated land decreases, construction land increases.
The dynamic degree of construction land change reached 5.30% during 1990–2000 and 7.11% during 2000–2010, indicating that urban expansion was most intensive during these 2 decades. During the same period, the areas of cultivated land, forest land, and grassland continuously declined, decreasing by 30.4%, 18.1%, and 18.8%, respectively. Their dynamic degrees also reached high values between 1990 and 2010. The expansion of roads and infrastructure required for urbanization continuously encroached on agricultural and ecological land at the urban fringe, accelerating the pace of urban growth.
However, constrained by topographic conditions, urban expansion was mainly concentrated in the central and southeastern plains of Beijing, while forest land in the western and northern mountainous regions was less affected. From 1980 to 2010, forest land in mountainous areas showed only a marginal increase of 0.002%, with a dynamic degree below 0.16%.
During the slower urbanization stage from 2010 to 2020, the area of construction land decreased to 3,561.07 km2, representing a decline of 7.3%. In contrast, the areas and proportions of ecological land types—including forest land, grassland, and water bodies—showed an overall increase. The dynamic degree of grassland and water bodies reached 1.28% and 2.97%, respectively. Compared with the intense urban expansion from 1980 to 2010, this shift indicates that rapid urbanization in Beijing has reached a saturation point, leading to a slowdown in land expansion.
Spatial configuration of LU patterns
Drawing on the spatial arrangement and temporal shifts of various LU categories from 1980 to 2020, a synthesized map illustrating the evolution of LU in Beijing was generated (Figure 3). The map highlights the spatial impacts of the two major historical development stages identified previously.
Schematic diagram of LU evolution in Beijing.
Set of six land use maps from 1980, 1990, 2000, 2010, and 2020 showing a region with increasing red areas representing construction land spreading mainly in the central and southern sections, while yellow cultivated land and green woodland areas decrease. A legend indicates yellow for cultivated land, green for woodland, light green for grassland, blue for water, red for construction land, and gray for unused land, along with a forty kilometer scale and north arrow.
From 1980 to 2010, urban expansion originated from the central urban area and spread outward into surrounding ecological land in a radial pattern. During this period, several secondary urban centers gradually emerged in the outer suburbs, forming smaller nodes that became spatially connected to the Beijing metropolitan core.
After 2010, Beijing’s urbanization strategy shifted toward a slower-growth model. Urban construction began to move in two main directions: (1) upgrading and redeveloping the old urban areas, including infrastructure renewal and conversion of developed land into urban green space; and (2) relocating part of the central urban population and urban functions to suburban secondary centers, thereby strengthening the socioeconomic spillover effects of the urban core while alleviating land pressure in the central districts.
Features of LU conversion processes
The magnitudes of LU transformations across the study region are summarized in Table 9. These tables display the specific transfer quantities between different LU types at 10-year intervals, illustrating the mutual conversion patterns among various land categories in Beijing.
LU transition area in Beijing from 1980 to 2020.
LU type
Year
Cultivated land
Forest land
Grassland
Water bodies
Construction land
Unused land
Cultivated land
1980–1990
5,830.13
21.03
1.01
11.16
37.36
0
1990–2000
4,870.73
130.38
13.2
93.62
742.89
0
2000–2010
3,239.63
141.98
45.72
31.52
1,449.72
1.15
2010–2020
3,196.57
164.15
28.14
83.37
316.36
8.06
Forest land
1980–1990
2.81
7,289.55
8.13
0.07
0.23
0
1990–2000
18.79
7,260.52
14.95
0.27
17.68
0
2000–2010
219.47
6,962.29
30.96
28.44
190
0.02
2010–2020
50.5
7,191.86
19.78
19.13
34.96
0.5
Grassland
1980–1990
2.86
1.48
1,351.01
3.38
0.03
0
1990–2000
17.06
36.84
1,266.89
28.6
14.9
0
2000–2010
44.17
170.64
1,018.65
6.38
55.24
0.01
2010–2020
5.39
20.47
1,073.95
3.8
8.8
0.01
Water bodies
1980–1990
14.18
0.06
4.12
382.52
0.23
0
1990–2000
2.5
3.29
0.02
388.59
2.73
0
2000–2010
118.55
19.74
9.07
251.88
111.82
0.06
2010–2020
15.69
13.2
19.38
261.93
11.72
3.66
Construction land
1980–1990
0.85
0.1
0.03
0.01
1,429.5
0
1990–2000
0.64
0.14
0.02
0.04
1,466.51
0
2000–2010
174.82
21.98
7.97
7.36
2032.6
0
2010–2020
391.31
88.3
113.49
53.91
3,189.05
3.8
Unused land
1980–1990
0
0
0
0
0
1.13
1990–2000
0
0
0
0
0
1.13
2000–2010
0.03
0.09
0.05
0.01
0.49
0.46
2010–2020
0.76
0.02
0.01
0.01
0.17
0.74
Across the 4 decades from 1980 to 2020, two LU transitions surpassed 500 km2 in magnitude: the shift from cropland to built-up areas during 1990–2000 (742.89 km2) and again during 2000–2010 (1,449.72 km2). Three medium-scale transitions, falling within the 201–500 km2 range, were identified-namely, the conversion of woodland to cropland between 2000 and 2010, as well as the two-way exchanges between cropland and construction land from 2010 to 2020. Changes of 100–200 km2 occurred in several situations, including the transformation of cropland into forest between 1990 and 2000, various reciprocal conversions among cropland, woodland, water bodies, and built-up land during 2000–2010, and the cropland-to-forest increase observed in 2010–2020.
Between 1980 and 2020, the dominant LU exchanges in Beijing were the two-way conversions between cropland and urban built-up areas, which stand out as the most significant transitions (Figure 4). This pattern reinforces the earlier interpretation that, in the initial phases of urban growth, cropland situated around the metropolitan periphery was progressively absorbed into expanding built-up areas. During 2010–2020, the two-way exchanges between cropland and construction land once again surpassed 300 km2, revealing that in the later stage of urbanization, planning and policy priorities moved away from outward expansion and toward ecological repair, including the reinstatement of agricultural land.
The Sankey diagram of LUC from 1980 to 2020. Note: The Sankey diagram visualizes LU flows, where the width of each branch is proportional to the corresponding LU area.
Sankey diagram showing land use changes from 1980 to 2020. Categories are cultivated land, woodland, grassland, water, construction land, and unused land with colors matching the legend above. Notable shifts from cultivated and woodland to construction land occur after 2000.
In addition, the bidirectional conversions among cultivated land, forest land, grassland, and water bodies since the beginning of the 21st century suggest that, under urbanization pressure, agricultural land had to extend toward ecological land due to demand. However, this phenomenon was alleviated after the implementation of ecological restoration programs promoted by the Beijing municipal government. Overall, policies such as transforming the old urban area into urban green space, returning urban fringe land to cultivated land, and converting cultivated land back to forest or grassland were implemented in a progressive manner to achieve a balance between ecological restoration and urbanization, ultimately contributing to the progress toward the “dual-carbon” goal.
Spatiotemporal dynamics of CS in Beijing (1980–2020)Temporal evolution of CS
Using the CS module of the InVEST framework, this study quantified CS associated with various LU categories in Beijing for each decade between 1980 and 2020. A stacked bar visualization (Figure 5) illustrates both the total CS for each time point and the relative contributions of individual LU types. The estimated totals for the five periods were 2.9462 × 108 t, 2.9468 × 108 t, 2.9070 × 108 t, 2.7828 × 108 t, and 2.8352 × 108 t, respectively. Overall, CS followed a pronounced “V-shaped” trajectory-gradually declining through 2010 and experiencing a modest rebound thereafter.
Stacked bar chart of CS by LU type in Beijing from 1980 to 2020.
Stacked bar chart displaying carbon storage by land use types from 1980 to 2020, with woodland contributing the largest portion, followed by cultivated land, grassland, construction land, water, and unused land.
Examining CS by LU category reveals that cropland CS fell from 5.583 × 107 t in 1980 to 3.463 × 107 t by 2020, a reduction of 37.97%, making cropland loss the dominant driver behind the long-term decrease in total CS. Forest and grassland CS exhibited patterns similar to the aggregate trend, with slight increases observed during 2010–2020. CS associated with built-up areas rose steadily from 1980 to 2010 but showed a minor decline in the final decade.
In addition, forest land, water bodies, and unused land showed relatively small overall fluctuations. Among them, forest land experienced a slight increase in CS from 2010 to 2020 and accounted for 75.5% of the total CS, making it the main carbon carrier of Beijing’s terrestrial ecosystem. Due to the low carbon density of water bodies and unused land, their CS contributions each accounted for less than 1% of the total, having little impact on the overall trend. However, it should be noted that both natural and artificial water systems are closely linked to ecological land types through interactions and mutual dependence. Therefore, when analyzing the spatiotemporal variation of CS in Beijing, the indirect influence of water bodies should not be overlooked.
Spatial pattern of CS distribution
The mapped CS distribution (Figure 6) reveals a clear gradient, with higher values concentrated in the western and northern sectors of Beijing and markedly lower values across the central and southeastern plains. The high-CS zone aligns with the continuous mountain system composed of Xishan to the west, Jundushan in the northwest, and the broader Yanshan Mountains along the northern boundary. These regions maintain extensive forest and grassland cover, together with water bodies, forming a relatively intact ecological belt with strong carbon-sequestration potential.
Spatial distribution pattern map of CS in Beijing from 1980 to 2020.
Set of five maps depicting spatial distribution of carbon storage in a region from 1980 to 2020 in ten-year increments, with dark blue indicating high carbon storage and light blue indicating low; legend provides scale, direction, and value range from 2.13 to 25.76 tonnes.
Conversely, low-CS values dominate the central and southeastern lowlands, radiating outward from the metropolitan core. Additional scattered low-value patches appear in certain mountainous settlements such as Miyun, Pinggu, and Yanqing. These areas largely overlap with the distribution of built-up land, where limited vegetation cover suppresses CS capacity.
The long-term spatial evolution of CS mirrors Beijing’s urbanization pathway, reflecting a close coupling between CS patterns and LU configuration. Built-up areas consistently correspond to low-value CS zones, whereas forests and grasslands underpin the high-value regions. Consequently, LU planning and policy design should better integrate the spatial interplay between CS and LU patterns, ensuring that the contributions of various LU categories are balanced so as to support a sustained increase in regional CS.
Spatial clustering of CS
To examine whether CS shows spatial clustering, we first conducted a global spatial autocorrelation analysis using Global Moran’s I. The results indicate strong and stable positive spatial autocorrelation across all years. Moran’s I was 0.885 in 1980, 0.883 in 1990, 0.889 in 2000, 0.900 in 2010, and 0.890 in 2020. These consistently high values suggest that CS was spatially clustered rather than randomly distributed during 1980–2020.
Given the strong global autocorrelation, we further applied Local Indicators of Spatial Association (LISA) to identify local clusters and spatial outliers (Figure 7). The LISA results reveal a clear and persistent clustering structure. HH clusters were mainly concentrated in the western and northern parts of Beijing and formed relatively continuous patches across years. LL clusters were mainly distributed in the central and southeastern parts and also appeared as large contiguous areas. By contrast, HL and LH patterns were limited in extent and occurred as scattered patches. These outliers were mostly located near the transition zones between HH and LL areas. Across the five time points, the main HH and LL cluster zones remained stable, while the spatial extent and internal continuity of clusters changed only moderately from decade to decade. Areas classified as Not Significant were mainly distributed around the major clusters and as small scattered patches.
LISA cluster maps of CS in Beijing for 1980 to 2020. Note: HH and LL represent high–high and low–low clusters, respectively; HL and LH indicate spatial outliers; non-significant cells are shown in grey.
Grid of five maps displays spatial clustering classification changes in a geographic region from 1980 to 2020, using red, pink, blue, and gray to show HH, HL, LH, LL, and not significant areas, with a legend and north arrow for reference.Hotspot–coldspot patterns of CS
To further characterize the spatial concentration of high and low CS, we conducted a Getis–Ord Gi* hot spot analysis for each decade from 1980 to 2020 (Figure 8).
Getis-Ord Gi* hot spot and cold spot maps of carbon storage in Beijing (1980–2020).
Grid of five geographic heat maps labeled by year from nineteen eighty to twenty twenty shows spatial distribution of hot and cold spots at various confidence levels, with red for hot spots and blue for cold spots. The legend explains the confidence levels for both spot types and indicates nonsignificant areas in white, with a north arrow and scale bar provided. Increase in hot spots over time is notable.
Across all five time points, the Gi* results show a stable and contrasting hot–cold configuration. Hot spots are consistently concentrated in the western and northern parts of Beijing, forming contiguous patches with multiple high-confidence (95%–99%) cells. In contrast, cold spots are mainly distributed across the central and southeastern parts of the municipality and also exhibit large continuous areas, including many cells reaching 95%–99% confidence.
Temporal comparisons indicate that the overall locations of hot and cold spots remain broadly consistent from 1980 to 2020. The hot-spot belt in the western–northern sector persists across decades, and the cold-spot core in the central–southeastern sector is also maintained. Meanwhile, Not Significant areas are mainly located in the transition zone between hot and cold clusters, and appear as fragmented patches surrounding the major hot/cold regions.
Influence of LU dynamics on CS
Table 10 summarize the magnitudes of LU conversions across the study period, detailing the exchanges among different LU categories at each decade interval.
Impact of LUC on CS in Beijing (1980–2020) (Ten thousand tons).
LU type
Year
Cultivated land
Forest land
Grassland
Water bodies
Construction land
Unused land
Cultivated land
1980–1990
0
40.3
0.85
−7.92
−23.42
0
1990–2000
0
249.82
11.11
−66.4
−465.75
0
2000–2010
0
272.06
38.49
−22.36
−908.88
−0.8
2010–2020
0
314.54
23.69
−59.13
−198.34
−5.64
Forest land
1980–1990
−5.38
0
−8.74
−0.18
−0.58
−0.01
1990–2000
−36.01
0
−16.06
−0.71
−44.97
0
2000–2010
−420.53
0
−33.27
−74.67
−483.19
−0.05
2010–2020
−96.77
0
−21.25
−50.22
−88.92
−1.29
Grassland
1980–1990
−2.4
1.59
0
−5.24
−0.04
0
1990–2000
−14.36
39.58
0
−44.36
−21.88
0
2000–2010
−37.18
183.34
0
−9.89
−81.13
−0.02
2010–2020
−4.54
21.99
0
−5.89
−12.92
−0.01
Water bodies
1980–1990
10.06
0.17
6.39
0
0.02
0
1990–2000
1.78
8.64
0.03
0
0.22
0
2000–2010
84.08
51.82
14.06
0
9.21
0
2010–2020
11.13
34.66
30.06
0
0.96
0.04
Construction land
1980–1990
0.53
0.25
0.05
0
0
0
1990–2000
0.4
0.36
0.03
0
0
0
2000–2010
109.6
55.88
11.7
−0.61
0
0
2010–2020
245.33
224.55
166.68
−4.44
0
−0.28
Unused land
1980–1990
0
0
0
0
0
0
1990–2000
0
0.01
0
0
0
0
2000–2010
0.02
0.24
0.07
0
0.04
0
2010–2020
0.53
0.04
0.01
0
0.01
0
During 1980–1990, LU adjustments were relatively modest, and the resulting effect on CS was correspondingly minor. A more substantial shift emerged in 1990–2000, when accelerated urban development caused cropland to become the primary LU category encroached upon by expanding built-up areas. Consequently, the reduction in cropland was the main contributor to CS losses in this decade.
The period from 2000 to 2010 marked the peak of urban expansion in Beijing. Not only did cropland continue to decline, but forested areas—critical to maintaining CS in terrestrial ecosystems—were also affected. The decrease in forest area accounted for the largest CS reduction during this decade, followed by the loss of cultivated land. Notably, more than 45% of the forest area that transitioned during this period was converted into construction land, further demonstrating that urbanization-induced LU transitions substantially eroded CS.
After 2010, the contraction of forest land diminished by 74.5%, and cropland shifted from net loss to net gain. These changes indicate that the ecological restoration programs launched by the municipal government produced measurable improvements, contributing positively to progress toward regional “dual-carbon” objectives.
Analysis of driving mechanisms of CS based on the geodetectorSelection and preprocessing of driving factors
Beijing is a megacity with strong land-use contrasts between the mountainous ecological areas and the intensively developed plains. In this setting, the spatial pattern of CS is shaped by both biophysical constraints and human activities. Following prior studies that use factor-based attribution frameworks (Bi et al., 2024; Cai et al., 2024; Zhao et al., 2025), we selected drivers to cover three mechanism groups: (i) natural controls that constrain vegetation productivity and land suitability, (ii) socioeconomic pressure that proxies development intensity and land management, and (iii) accessibility conditions that capture where land conversion and infrastructure expansion tend to concentrate.
Natural factors (dem, slope, precipitation, temperature, and NDVI) describe the biophysical constraints and ecosystem status that set the potential for carbon accumulation. Elevation and slope constrain where urban expansion and farming can occur and they shape vegetation and soil conditions, so they affect CS through LU patterns and carbon pool capacity (Bi et al., 2024). Precipitation and temperature represent water–energy conditions that control plant growth and soil carbon processes, so they regulate spatial differences in CS (Nie et al., 2024). NDVI is a proxy for vegetation cover and biomass, which is directly related to carbon stored in biomass pools and partly reflects ecosystem productivity (Wang X. et al., 2024).
Socio-economic factors (population and GDP) capture development intensity and land-demand pressure. Higher population and GDP often increase the probability of converting ecological land to built-up land and they reshape the LU structure, which then changes CS (Nie et al., 2024).
Accessibility factors (distance to railways, distance to roads, and distance to waterways) reflect the spatial conditions that guide land conversion. Areas closer to roads and railways tend to attract construction and related land transitions, so these distances can explain CS patterns indirectly through LU conversion intensity (Wang X. et al., 2024). Distance to waterways is included because river and reservoir corridors often influence settlement layout, farmland distribution, and ecological land configuration, which can also affect CS through LU structure (Huang et al., 2024).
The preprocessing of the driving factors was conducted using ArcGIS, including metadata checks, coordinate system unification, and raster alignment to a consistent grid. For accessibility indicators, Euclidean distance rasters were generated from road, railway, and water layers. Based on these steps, a set of driving factor maps for the study area was produced (Figure 9).
Geospatial drivers diagram with mechanism pathways.
Nine-panel grid illustrating spatial distributions of DEM, slope, precipitation, temperature, NDVI, population, GDP, and distances to railways, roads, and waterways for a region, with a color-coded legend indicating normalized values from zero (red) to one (blue), and a north arrow and scale bar for geographic reference.Analysis of single-factor detection results
In this study, CS values from the most recent decade serve as the dependent variable Y. Following the previously established criteria for variable selection, ten explanatory factors X were incorporated. These include topographic and climatic variables: Dem (X1), slope (X2), precipitation (X3), temperature (X4), NDVI (X5), population (X6), GDP (X7), distance to railways (X8), distance to roads (X9), and distance to waterways (X10).
The analysis are illustrated in Figure 10. All ten driving factors show q-values higher than 0.1, indicating that each factor is correlated with CS in the study area. However, their explanatory power differs. According to the rule, the descending order of influence is: X2>X1>X4>X5>X3>X7>X6>X9>X8>X10. Among them, slope shows the strongest explanatory power, with a q-value of 0.62. In addition, elevation, precipitation, temperature, and NDVI also have q-values greater than 0.5.
Schematic diagram of single factor detection based on Geodetector.
Bar chart with ten light blue bars labeled X1 to X10 on the x-axis, showing values for q on the y-axis from 0.0 to 0.7; highest bars are X1 and X2, lowest are X8 and X10.
Slope and elevation influence LU types and vegetation retention, while precipitation and temperature affect hydrothermal conditions that regulate plant growth. NDVI reflects vegetation coverage and biomass per unit area. Together, these five factors dominate the distribution of ecological land in Beijing and, in turn, shape the spatial distribution of CS. Their underlying geographic and ecological mechanisms result in relatively high explanatory power in the single-factor detection results.
To clarify the effect direction of each driver and its spatial non-stationarity, we further selected the seven factors with relatively stronger GeoDetector explanatory power (X1–X7) and ran an OLS model as a baseline, followed by a GWR model. In the OLS results (Table 11), slope, precipitation, NDVI, and GDP show positive associations with CS, while DEM, temperature, and population show negative associations; all coefficients are significant (p < 0.01). The collinearity check indicates that all VIF values are below 7.5, so multicollinearity is not severe and the regression results are acceptable for the subsequent GWR analysis. Compared with OLS, GWR improves model fit, with a lower AICc (−2056.02 vs. 4,537.74) and a higher adjusted R2 (0.97 vs. 0.89), suggesting that allowing coefficients to vary across space better captures the spatial structure of CS. The GWR coefficient maps (Figure 11) further show clear spatial heterogeneity: the effects of DEM and temperature are mostly negative but vary in magnitude across the municipality, whereas slope and NDVI are mainly positive and tend to be stronger in areas with more complex terrain and higher vegetation coverage. Precipitation is predominantly positive, but its strength differs across space, while population is mainly negative with localized deviations, and GDP is generally weakly positive with spatially uneven intensity. The residual surface shows limited magnitude and no obvious large contiguous patches, indicating that the fitted patterns are broadly consistent with the observed CS distribution.
OLS and GWR results for the effects of key drivers on CS in Beijing.
Indicators
OLS
GWR
P
VIF
Coefficients
Dem (X1)
−0.01
−0.2
<0.01***
7.43
Slope (X2)
0.54
0.63
<0.01***
3.87
Precipitation (X3)
0.02
0.19
<0.01***
4.87
Temperature (X4)
−0.35
−0.15
<0.01***
7.04
NDVI (X5)
0.01
0.31
<0.01***
3.55
Population (X6)
−0.01
−0.19
<0.01***
6.94
GDP (X7)
0.01
0.11
<0.01***
6.33
AICc
4,537.74
−2056.02
Adjusted R2
0.89
0.97
*** means p<0.01.
Spatial distribution of GWR for the selected drivers.
Grid of eight thematic map panels for a region, each labeled and shaded in orange representing variables: DEM, Slope, Precipitation, Temperature, NDVI, Population, GDP, and Residual. Each map displays a color bar legend with value ranges, north arrow, and scale bar. The bottom right panel is blank.Analysis of interaction detection results
Figure 12 presents the interaction-detection outcomes. The analysis indicates that every pairwise combination among the ten driving variables exhibits either mutual reinforcement or nonlinear amplification effects. No weakening or independent interaction types were observed, indicating that none of the driving factors offset each other; instead, most exhibit synergistic enhancement. Compared with single-factor results, all paired combinations show higher explanatory power than individual factors. Among them, combinations of natural driving factors generally produce higher q-values, followed by combinations of natural and socioeconomic factors, while combinations consisting solely of socioeconomic or accessibility factors show relatively limited enhancement effects.
Schematic diagram of interaction factor detection based on Geodetector.
Heatmap showing the pairwise correlation coefficients between variables X1 to X10, with stronger positive correlations in red and weaker or lower correlations in blue, scale ranges from 0.28 to 0.68.
Given the strong explanatory power of slope in the single-factor results and its coupling effect with other natural factors, the combinations of slope (X2) with elevation (X1), temperature (X4), and NDVI (X5) show the highest explanatory power among all factor pairs, reaching approximately 0.68. Such high qqq-values often indicate complementary or synergistic effects between the paired factors—for example, the elevation–slope combination reflects terrain as one of the dominant controls shaping the spatial distribution of CS in Beijing. Additionally, some combinations involving natural and socioeconomic factors also show synergistic enhancement, which corresponds to key spatial contradictions affecting CS distribution, such as the NDVI–population combination, which reflects how the coupling of ecological conditions and human disturbance intensifies spatial differentiation in CS.
Prediction of LUC and CS in 2030Spatiotemporal patterns of LU across alternative development scenarios
To qualitatively and quantitatively analyze LUC driven by different future policies, this study combines the historical LU transition patterns of Beijing with the government plan outlined in the Beijing Urban Master Plan (2016–2035). Two scenarios were established, the IDS and the CPS. The PLUS framework was employed to generate simulations of LU dynamics and corresponding CS outcomes for both development scenarios.
The IDS is based on the Markov Chain model and primarily considers historical LU evolution, excluding natural and socio-economic influences on future LUC (Table 12). Under this scenario, cultivated land shows a loss of 130.07 km2, while forest land, grassland, and water bodies show gains of 152.90 km2, 134.49 km2, and 71.75 km2, respectively. Construction land decreases by 236.02 km2, and unused land increases by 7.04 km2. The transition pattern in the IDS mainly reflects a shift from construction land and cultivated land to forest land, grassland, and water bodies, although the dynamic degree of such transitions remains relatively low. Compared with 2020, the proportional change of each land type is minor, with construction land decreasing by approximately 1.5% and all other land types changing by less than 1%. In addition, although unused land exhibits the highest dynamic degree, its relatively small change in area contributes little to the overall transition trend.
Scenario-based LU area and proportion in Beijing (2030).
LU type
IDS
CPS
Area/km2
Percentage/%
Area/km2
Percentage/%
Cultivated land
3,530.17
21.53
3,457.24
21.09
Forest land
7,630.91
46.55
7,729.36
47.15
Grassland
1,389.14
8.47
1,433.67
8.75
Water bodies
493.90
3.01
493.90
3.01
Construction land
3,325.05
20.28
3,255.21
19.86
Unused land
23.80
0.15
23.57
0.14
Under the CPS, carbon sink objectives are prioritized, and ecological space protection is more stringent than in the IDS, reflecting the vision of ecological civilization at the national and societal levels and providing a reference for future LU planning in Beijing. In this scenario (Table 13), the “Greening Beijing” policy is assigned the highest level of ecological priority. Cultivated land decreases by 202.99 km2, while forest land, grassland, and water bodies increase by 251.35 km2, 178.93 km2, and 71.75 km2, respectively. Construction land decreases by 305.86 km2, and unused land increases by 6.81 km2.
Dynamic degree of LUC under multiple scenarios (2020–2030).
LU type
IDS
CPS
Area change/km2
Dynamic degree/%
Area change/km2
Dynamic degree/%
Cultivated land
−130.07
−0.36
−202.99
−0.55
Forest land
152.90
0.20
251.35
0.34
Grassland
134.39
1.07
178.93
1.43
Water bodies
71.75
1.70
71.75
1.70
Construction land
−236.02
−0.66
−305.86
−0.86
Unused land
7.04
4.20
6.81
4.06
Comparing the two development scenarios, the overall LU transition pattern in the CPS is generally consistent with that in the IDS, but the magnitude of change is greater. Except for water bodies and unused land, all other land types exhibit proportion changes greater than 1%. In the IDS, construction land and cultivated land account for 64.5% and 35.5% of total land loss, respectively, while the ratio of land gained by forest land, grassland, and water bodies is 1.15: 1: 0.55. In the CPS, construction land and cultivated land account for 60% and 40% of total land loss, while the corresponding ratio of land gained by forest land, grassland, and water bodies is 1.4 : 1: 0.4.
Figure 13 presents the LU projections for the two scenarios derived from the PLUS simulations. While the scenarios show noticeable differences in terms of LU area changes and their dynamic rates, the overall spatial configurations remain broadly comparable. This resemblance suggests that, guided by the “dual-carbon” agenda, LU planning in Beijing has progressed in a relatively consistent and orderly manner. Compared with the 2020 LU map, construction land in suburban areas such as Shunyi, Mentougou, Fangshan, Miyun, and Daxing shows a small-scale reduction, with most of the converted land becoming cultivated land and grassland. In addition, small-scale reductions are also observed in mountainous towns located in the outer suburbs, where construction land is mainly converted into forest land to support the formation of continuous ecological corridors in mountainous regions.
Scenario-based future LU prediction map of Beijing.
Side-by-side map comparison labeled 2030IDS and 2030CPS shows projected land use types with color coding: yellow for cultivated land, dark green for woodland, light green for grassland, blue for water, red for construction land, and gray for unused land, accompanied by a legend and a 40-kilometer scale bar.
Overall, both scenarios show LU transitions in the urban core, suburban areas, and outer mountainous zones that are favorable for enhancing ecosystem carbon sink capacity. Under the CPS, the effects of “urban renovation, conversion of construction land to cropland, afforestation and grassland restoration, and water conservation” are more pronounced in Beijing.
Spatiotemporal changes in CS under different scenarios
Using the projected LU datasets, the InVEST framework was employed to estimate the CS of Beijing in 2030 for both scenarios (Figure 14; Table 14). A comparison with the 2020 baseline reveals that total CS is expected to rise by approximately 4.99 × 106 t under the IDS and by about 7.69 × 106 t under the CPS.
Scenario-based projection of terrestrial CS in Beijing.
Two side-by-side maps labeled 2030IDS and 2030CPS display carbon storage distribution in a geographic region, with darker blue indicating higher storage values. A legend shows storage ranges from 2.13 to 25.76 tons.
Scenario-based projection of terrestrial CS in Beijing (million tons).
LU type
IDS
CPS
Cultivated land
33.40
32.71
Forest land
218.42
221.24
Grassland
24.84
25.63
Water bodies
1.17
1.17
Construction land
10.62
10.39
Unused land
0.06
0.06
Total
288.51
291.21
The variations in CS across individual LU categories - where each pair of values refers first to IDS and then to CPS - are summarized as follows: total CS +1.76%; +2.71%, cultivated land −3.55%; −5.54%, forest land +2.04%; +3.35%, grassland +10.74%; +14.27%, water bodies +17.00%; +17.00%, construction land −6.60%; −8.61%, and unused land +50.00%; +50.00%.
Among all LU categories, unused land has a relatively small base area, meaning that even small changes in CS may result in large variation percentages; therefore, this category is not included in the analysis. In the projected LU transitions from cultivated land and construction land to forest land, grassland, and water bodies, grassland and water bodies show relatively high proportions of CS increase. Although forest land contributes most to the overall increase in CS, its change rate appears less significant than that of grassland and water bodies due to its large CS base in 2020.
From the standpoint of total CS gain, the CPS delivers an increase roughly 54% greater than that produced under the IDS. Despite this numerical difference, the spatial configuration of CS in both scenarios is broadly comparable, zones with high CS remain concentrated in the mountainous terrain to the west and north, whereas low-CS regions continue to occupy the central and southeastern plains. This spatial contrast highlights the distinction between the urban core—characterized by extensive built-up land and limited carbon-sequestration capacity—and the peripheral suburban districts, where ecological land cover predominates and stronger carbon-sink functions are maintained.
Influence of LU transitions on CS
Tables 15 and 16 summarize how LU conversions affect CS under the two projected scenarios. Across both pathways, the principal contributors to CS variation are the transformations in which cropland is replaced by forest or grassland, as well as the reversion of built-up areas into cropland, forest, or grassland. Among these shifts, the conversion of construction land into forest land yields the most substantial CS gains, owing to the pronounced carbon-density contrast between these LU categories. This transition results in increases of 2.1156 × 106 t under the IDS and 3.4542 × 106 t under the CPS.
Impact of LU transition on CS under scenario IDS (ten thousand tons).
2020–2030 IDS
Cultivated land
Forest land
Grassland
Water bodies
Construction land
Unused land
Cultivated land
0.00
124.30
80.26
−36.40
−37.42
−3.97
Forest land
−3.48
0.00
−8.89
−7.25
−4.01
−0.03
Grassland
−4.48
18.17
0.00
−4.64
−1.98
−0.10
Water bodies
2.71
6.10
5.42
0.00
0.11
0.01
Construction land
85.18
211.56
78.98
−2.16
0.00
−0.07
Unused land
0.01
0.14
0.21
0.00
0.00
0.00
Impact of LU transition on CS under scenario CPS (ten thousand ons).
2020–2030 CPS
Cultivated land
Forest land
Grassland
Water bodies
Construction land
Unused land
Cultivated land
0.00
196.84
113.77
−34.38
−40.99
−3.93
Forest land
0.00
0.00
−29.20
−22.16
0.00
0.00
Grassland
0.00
45.64
0.00
−4.23
0.00
0.00
Water bodies
2.69
15.25
5.14
0.00
0.12
0.01
Construction land
94.39
345.42
85.65
−2.23
0.00
−0.07
Unused land
0.01
0.35
0.25
0.00
0.00
0.00
Based on the above simulation results, the LU transfers between construction land and cultivated land, and between these categories and forest land, grassland, and water bodies, have the most significant impacts on CS under both the IDS and CPS scenarios for 2030. In recent years, Beijing has achieved preliminary progress in improving LU structure, increasing CS, and balancing its spatial distribution by implementing the “Greening Beijing” policy and the Beijing Urban Master Plan. However, the comparison between the two scenarios in this study shows that the CPS produces nearly 50% higher CS gains than the IDS, indicating that there is still considerable potential for improvement in LUC and CS enhancement in Beijing.
DiscussionLU restructuring and the “decline–rebound” CS pathway
Beijing’s LU pattern underwent marked changes between 1980 and 2020, which can be divided into two distinct phases: a period of rapid urban expansion and a subsequent period of slowed expansion. From 1980 to 2010, construction land in Beijing increased sharply from 1,430.49 km2 to 3,839.87 km2 (rising from 8.73% to 23.42%), with a growth rate of 168%. Large areas of cultivated land, forest land, and grassland were converted to construction land during this period (Piao et al., 2009; Hu et al., 2018). After 2010, with urban expansion approaching saturation and ecological protection policies being progressively strengthened (Zhou et al., 2021), LU structure shifted toward a more stable pattern. Construction land experienced a slight decline (a 7.3% reduction from its 2010 peak, falling to 3,561.07 km2), while the areas and proportions of forest land, grassland, and water bodies increased (Hu et al., 2023). This trend is similar to LU restructuring observed in other global megacities during their later development stages, such as Tokyo, Shanghai, and Seoul, which also exhibited a shift from expansion to ecological restoration. Such patterns reflect a transition toward higher resource efficiency and ecological security, aligning with the sustainability principles emphasized in SDG 11 (Wang et al., 2018; Choi and Kim, 2022; Li Y. et al., 2025).
This LU evolution directly shaped the temporal and spatial characteristics of Beijing’s terrestrial CS. Based on InVEST model estimates, total CS in Beijing followed a “decline–rebound” V-shaped trajectory between 1980 and 2020 (Kang et al., 2025; Zhang et al., 2025b). Different land types contributed unevenly to this pattern: the CS of cultivated land decreased sharply from 5.583 × 107 t in 1980 to 3.463 × 107 t in 2020, making it the main driver of overall carbon loss-a finding consistent with studies in the Huang-Huai-Hai Plain (Wang X. et al., 2024). Forest and grassland CS exhibited trends similar to the total CS curve, declining before 2010 and rising thereafter. Due to its high carbon density (Chen et al., 2022), forest land accounted for 75.5% of total terrestrial CS in 2020, giving it the greatest influence on overall carbon change. This result is consistent with findings from other studies applying the PLUS–InVEST framework across different regions in China (Li et al., 2023; Zhen et al., 2023; Chen et al., 2024).
Taken together, Beijing’s “decline–rebound” CS pathway can be interpreted as the net outcome of long-term LU transitions, especially the balance between the expansion of built-up land and the protection/restoration of high-carbon-density ecological land.
Spatial structure, dominant drivers, and policy-related mechanisms
The GeoDetector analysis revealed that spatial differentiation in CS is jointly influenced by natural, socio-economic, and accessibility-related factors. In the single-factor detection, natural factors had significantly greater explanatory power than human-related factors, indicating that the ecological system in Beijing’s mountainous areas—especially regions with high elevation and steep slopes—remains relatively stable (Zhang et al., 2025b) and serves as the city’s primary carbon sink. This aligns with earlier research emphasizing the importance of mountainous and forest ecosystems for carbon sequestration (Zhu et al., 2015). In contrast, low-elevation and low-slope urban areas exhibit high population density but weak CS capacity. The interaction detector further showed that all variable combinations produced synergistic enhancement effects rather than mutual weakening. Natural–natural combinations demonstrated the highest explanatory power, exceeding that of any single factor. Although some studies highlight the role of socio-economic drivers in shaping CS (Li J. et al., 2025; Zhang et al., 2025c), in Beijing, the stability of its mountain ecosystems and effective ecological protection policies intensify the dominant influence of natural factors. Thus, even though heavily developed lowland areas undergo rapid land conversion, their limited vegetation cover restricts potential carbon gains, whereas preserved forest and grassland areas maintain higher and more stable CS.
The newly added spatial autocorrelation results further support this interpretation. Global Moran’s I values remain consistently high (0.883–0.900 from 1980 to 2020), indicating strong positive spatial clustering of CS rather than a random pattern. LISA clusters and Gi* hot/cold spots show that high–high aggregation persists in the mountainous belt, while low–low aggregation remains in the central–southeastern plains. This stable “mountain–plain” clustering pattern suggests that terrain-related constraints provide a structural template for CS distribution, within which urban expansion and ecological restoration mainly change the intensity and marginal extent of low-CS areas rather than reversing the overall spatial gradient.
From a policy perspective, the post-2010 stabilization of construction land and the recovery of forest/grassland/water bodies are consistent with strengthened ecological protection and greening efforts in Beijing (Zhou et al., 2021; Hu et al., 2023). In this context, policies can be viewed as altering the direction and magnitude of net LU transitions, thereby influencing area-weighted carbon density and the spatial continuity of high-CS landscapes.
Policy implications and generalizability
Finally, simulation results under the Markov–PLUS framework for the IDS and CPS scenarios indicate that ecological land expansion will play a central role in future carbon enhancement. The CPS scenario shows a much stronger CS increase than the IDS scenario, consistent with research outcomes in other cities such as Northern Anhui, Hohhot, and the Yangtze River Basin, where ecological prioritization yielded the highest CS potential (Ye et al., 2025; Zhang J. et al., 2025; Zhou J. et al., 2025). For megacities like Beijing, improving the layout of ecological spaces, restricting construction land expansion, and enhancing forest landscape connectivity are key strategies to strengthen future carbon sequestration. The implications extend beyond regional carbon neutrality goals and contribute to coordinated progress toward SDG 11, SDG 13, and SDG 15.
This study advances the field by providing a replicable “assessment–diagnosis–simulation–optimization” workflow rather than a single-model application. Specifically, it integrates (i) long-term LU monitoring and CS accounting using InVEST, (ii) driver and interaction diagnosis via GeoDetector, (iii) spatial-structure validation using Moran’s I, LISA, and Gi* analyses, and (iv) scenario-based LU projection using Markov–PLUS.
The framework is transferable to other megacities because its key inputs are commonly available (multi-period LU maps, consistent carbon-density parameters, spatial driver layers, and policy-translatable scenario constraints). At the same time, Beijing highlights that transferability does not imply identical dominant drivers: in terrain-heterogeneous cities, topographic and climatic controls may dominate CS heterogeneity, whereas in flatter megacities such as Shanghai, socio-economic intensity and land redevelopment can play a larger role (Lyu and Li, 2025). Likewise, mature megacities such as Tokyo often emphasize compact growth and green/blue infrastructure strategies to improve ecosystem services under limited developable land (Wang et al., 2018). Adding these cross-city comparisons clarifies the generalization value, the framework is general, while the dominant mechanisms are context-dependent.
Despite these meaningful findings, the study has limitations. The InVEST model applies fixed carbon density values for each land type and does not capture interannual fluctuations or ecological succession, which may lead to underestimation of carbon gains under strong policy intervention. Additionally, land-transfer probability parameters are derived under static socio-economic assumptions and do not reflect potential impacts of future policy shifts, such as stricter urban boundary controls or land-approval reforms. Future work would benefit from combining multi-source remote-sensing datasets with ground-based carbon observations to refine model parameterization and enhance predictive reliability. Additionally, constructing a broader range of policy-oriented scenario frameworks would provide more diverse and informative projections for long-term CS assessment.
Conclusion
This study develops and tests a transferable, scenario-oriented workflow for urban carbon-stock management by linking (i) long-term land-use dynamics, (ii) driver diagnosis, and (iii) spatially explicit land-use simulation within an integrated Markov–PLUS–InVEST framework. By combining pattern tracking with mechanism screening and future pathway testing, the approach moves beyond “mapping carbon” to “planning for carbon”.
At the megacity scale, Beijing’s carbon-stock pattern is structured by a strong and persistent spatial clustering and a stable terrain-driven gradient, with high-carbon areas anchored in continuous mountainous ecosystems and low-carbon areas concentrated in the intensively urbanized plains. This indicates that land-use carbon stocks in megacities are constrained first by biophysical suitability (topography–climate–vegetation), while human activities mainly operate by reallocating land among carbon-dense and carbon-poor classes rather than overriding the natural template.
The driver and effect-direction analyses jointly suggest that carbon-stock heterogeneity cannot be explained by a single factor, because the dominant controls arise from the coupling of terrain, climate, and vegetation conditions, and their interactions with socio-economic pressures. Therefore, carbon-oriented land management should prioritize “where change happens” (spatial hotspots) as much as “what changes” (land transitions), especially in cities with pronounced mountain–plain contrasts.
Scenario results show that a carbon-sink–priority pathway can deliver larger carbon gains than an inertial trajectory, implying that planning rules that protect high-carbon ecosystems and guide land transitions toward carbon-dense classes are effective levers for carbon enhancement, particularly when they are consistent with existing ecological redlines and growth management. More broadly, the proposed workflow is applicable to other rapidly urbanizing regions where land-use transition data, carbon-density parameters, and key drivers are available, providing a practical template for scenario-based carbon-oriented land-use planning.
Data availability statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.
Author contributions
CL: Conceptualization, Data curation, Formal Analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review and editing. BL: Writing – original draft, Writing – review and editing.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
ReferencesAdelisardouF.ZhaoW.ChowR.MederlyP.MinkinaT.SchouJ. (2022). Spatiotemporal change detection of carbon storage and sequestration in an arid ecosystem by integrating google Earth Engine and InVEST (the Jiroft plain, Iran). Int. J. Environ. Sci. Technol.19 (7), 5929–5944. 10.1007/s13762-021-03676-6AnY.TanX.RenH.LiY.ZhouZ. (2024). Historical changes and multi-scenario prediction of land use and terrestrial ecosystem carbon storage in China. Chin. Geogr. Sci.34 (3), 487–503. 10.1007/s11769-024-1424-yAroraN. K.MishraI. (2024). Life on land: progress, outcomes and future directions to achieve the targets of SDG 15. Environ. Sustain.7 (4), 369–375. 10.1007/s42398-024-00330-yBabbarD.AreendranG.SahanaM.SarmaK.RajK.SivadasA. (2021). Assessment and prediction of carbon sequestration using Markov chain and InVEST model in Sariska Tiger Reserve, India. J. Clean. Prod.278, 123333. 10.1016/j.jclepro.2020.123333BaiY.WongC. P.JiangB.HughesA. C.WangM.WangQ. (2018). Developing China’s ecological redline policy using ecosystem services assessments for land use planning. Nat. Communications9 (1), 3034. 10.1038/s41467-018-05306-130072771Benez-SecanhoF. J.DwivediP. (2021). Analyzing the impacts of land use policies on selected ecosystem services in the upper chattahoochee Watershed, Georgia, United States. Environ. Res. Commun.3 (11), 115001. 10.1088/2515-7620/ac310cBiF.WuZ.LiangH.DuZ.LeiT.SunX. (2024). Spatio-temporal changes and driving factors of carbon storage in the middle reaches of the Yellow River based on PluS-InVEST-geodetector model. Environ. Sci., 1–20. 10.13227/j.hjkx.202406176CaiJ.ChiH.LuN.BianJ.ChenH.YuJ. (2024). Analysis of spatiotemporal predictions and drivers of carbon storage in the Pearl River Delta urban agglomeration via the PLUS-InVEST-GeoDetector model. Energies17 (20), 5093. 10.3390/en17205093ChenJ.ChenJ.LiaoA.CaoX.ChenL.ChenX. (2015). Global land cover mapping at 30 m resolution: a POK-based operational approach. ISPRS J. Photogrammetry Remote Sens.103, 7–27. 10.1016/j.isprsjprs.2014.09.002ChenS.LuN.FuB.WangS.DengL.WangL. (2022). Current and future carbon stocks of natural forests in China. For. Ecol. Manag.511, 120137. 10.1016/j.foreco.2022.120137ChenG.PengQ.FanQ.LinW.SuK. (2024). Spatial-temporal variation and driving forces of carbon storage at the county scale in China based on a gray multi-objective optimization-patch-level land use simulation-integrated valuation of ecosystem services and tradeoffs-optimal parameter-based geographical detector model: taking the daiyun mountain’s rim as an example. Land14 (1), 14. 10.3390/land14010014ChiY.QianT.ShengC.XiC.WangJ. (2021). Analysis of differences in the spatial distribution among terrestrial mammals using Geodetector—A case study of China. ISPRS Int. J. Geo-Information10 (1), 21. 10.3390/ijgi10010021ChoiJ.KimG. (2022). History of Seoul’s parks and green space policies: focusing on policy changes in urban development. Land11 (4), 474. 10.3390/land11040474DangR.-Y.WangJ.-F.ZhouZ.-Y.XuJ.XuJ.-L.AnL.-Z. (2025). Simulation of multi-scenario land use change and carbon storage in protected areas of Beijing. Environ. Sci.46 (12), 7747–7757. 10.13227/j.hjkx.20241118541413996DengH.ChenY. (2025). Remote‐sensing‐based monitoring of spatiotemporal variations and driving forces of terrestrial vegetation NPP in a subtropical urbanizing Region. Environ. Monit. Assess.197 (9), 1028. 10.1007/s10661-025-14481-w40830278DixonR. K.SolomonA.BrownS.HoughtonR.TrexierM.WisniewskiJ. (1994). Carbon pools and flux of global forest ecosystems. Science263 (5144), 185–190. 10.1126/science.263.5144.18517839174DongK.WangH.LuoK.YanX.YiS.HuangX. (2024). The use of an optimized grey multi-objective programming-PLUS model for multi-scenario simulation of land use in the Weigan–Kuche River oasis, China. Land13 (6), 802. 10.3390/land13060802FangJ.GuoZ.PiaoS.ChenA. (2007). Terrestrial vegetation carbon sinks in China, 1981–2000. Sci. China Ser. D Earth Sci.50 (9), 1341–1350. 10.1007/s11430-007-0049-1FarrT. G.RosenP. A.CaroE.CrippenR.DurenR.HensleyS. (2007). The shuttle radar topography mission. Rev. Geophysics45 (2). 10.1029/2005rg000183FilhoW. L.WallT.SalviaA. L.DinisM. A. P.MifsudM. (2023). The central role of climate action in achieving the United Nations’ Sustainable Development Goals. Sci. Rep.13 (1), 20582. 10.1038/s41598-023-47746-w37996539GongW.DuanX.MaoM.HuJ.SunY.WuG. (2022). Assessing the impact of land use and changes in land cover related to carbon storage by linking trajectory analysis and InVEST models in the Nandu River Basin on Hainan Island in China. Front. Environ. Sci.10, 1038752. 10.3389/fenvs.2022.1038752GuH.ChenX.SunH.HuanC.WangB. (2025). How does land urbanization affect carbon emissions in China? Evidence from 209 cities and three heterogeneous regions in the east of the Hu line of China. Land14 (4), 910. 10.3390/land14040910GuoP.WangH.QinF.MiaoC.ZhangF. (2023). Coupled MOP and PLUS-SA model research on land use scenario simulations in Zhengzhou metropolitan area, Central China. Remote Sens.15 (15), 3762. 10.3390/rs15153762HaklayM.WeberP. (2008). Openstreetmap: user-generated street maps. IEEE Pervasive Computing7 (4), 12–18. 10.1109/mprv.2008.80HamadR.BalzterH.KoloK. (2018). Predicting land use/land cover changes using a CA-Markov model under two different scenarios. Sustainability10 (10), 3421. 10.3390/su10103421HamelP.GuerryA.PolaskyS.HanB.DouglassJ.HamannM. (2021). Mapping the benefits of nature in cities with the InVEST software. Npj Urban Sustain.1 (1), 25. 10.1038/s42949-021-00027-9HuY.KongX.ZhengJ.SunJ.WangL.MinM. (2018). Urban expansion and farmland loss in Beijing during 1980–2015. Sustainability10 (11), 3927. 10.3390/su10113927HuY.SunJ.ZhengJ. (2023). Spatiotemporal analysis of urban carbon metabolism and its response to land use change: a case study of Beijing, China. Atmosphere14 (8), 1305. 10.3390/atmos14081305HuangX.LiuX.WangY. (2024). Spatio-temporal variations and drivers of carbon storage in the Tibetan Plateau under SSP-RCP scenarios based on the PLUS-InVEST-GeoDetector model. Sustainability16 (13), 5711. 10.3390/su16135711Ismaili AlaouiH.ChemchaouiA.El AsriB.GhaziS.BrhaddaN.ZiriR. (2023). Modeling predictive changes of carbon storage using invest model in the Beht watershed (Morocco). Model. Earth Syst. Environ.9 (4), 4313–4322. 10.1007/s40808-023-01697-3JiangB.BaiY.ChenJ.AlataloJ. M.XuX.LiuG. (2020). Land management to reconcile ecosystem services supply and demand mismatches—A case study in Shanghai municipality, China. Land Degrad. and Dev.31 (17), 2684–2699. 10.1002/ldr.3614JiaoK.LiuZ.WangW.YuK.McgrathM. J.XuW. (2024). Carbon cycle responses to climate change across China’s terrestrial ecosystem: sensitivity and driving process. Sci. Total Environ.915, 170053. 10.1016/j.scitotenv.2024.17005338224891JinA.LiH.WangX.WangZ. (2025). Spatiotemporal assessment of ecological quality and driving mechanisms in the Beijing metropolitan area. Sci. Rep.15 (1), 13136. 10.1038/s41598-025-97156-340240405KangJ.ZhangL.MengQ.WuH.HouJ.PanJ. (2025). Land use and carbon storage evolution under multiple scenarios: a spatiotemporal analysis of Beijing using the PLUS-InVEST model. Sustainability17 (4), 1589. 10.3390/su17041589KeenanR. J.ReamsG. A.AchardF.de FreitasJ. V.GraingerA.LindquistE. (2015). Dynamics of global forest area: results from the FAO global forest resources assessment 2015. For. Ecology Management352, 9–20. 10.1016/j.foreco.2015.06.014KumarS.RadhakrishnanN.MathewS. (2014). Land use change modelling using a Markov model and remote sensing. Geomatics, Nat. Hazards Risk5 (2), 145–156. 10.1080/19475705.2013.795502LeiT.WangD.YuX.MaS.ZhaoW.CuiC. (2023). Global iron and steel plant CO2 emissions and carbon-neutrality pathways. Nature622 (7983), 514–520. 10.1038/s41586-023-06486-737731002LeiJ.ZhangL.ChenZ.WuT.ChenX.LiY. (2024). The impact of land use change on carbon storage and multi-scenario prediction in Hainan island using InVEST and CA-Markov models. Front. For. Glob. Change7, 1349057. 10.3389/ffgc.2024.1349057LiP.ChenJ.LiY.WuW. (2023). Using the InVEST-PLUS model to predict and analyze the pattern of ecosystem carbon storage in Liaoning Province, China. Remote Sens.15 (16), 4050. 10.3390/rs15164050LiJ.HuJ.KangJ.ShuW. (2025a). Spatio-temporal variation and prediction of land use and carbon storage based on PLUS-InVEST model in Shanxi province, China. Landsc. Ecol. Eng.21 (1), 107–119. 10.1007/s11355-024-00627-4LiY.WangC.SunM.ZhangH. (2025b). Dynamic simulation of land use change and assessment of ecosystem services under climate change scenarios: a case study of Shanghai, China. Land14 (9), 1791. 10.3390/land14091791LiangY.XuC. (2023). Knowledge diffusion of geodetector: a perspective of the literature review and Geotree. Heliyon9 (9), e19651. 10.1016/j.heliyon.2023.e1965137810142LiangG.FangF.LinY.ZhangZ. (2024). Spatiotemporal characteristics of typical ecosystem services and their spatial responses to driving factors in ecologically fragile areas in Upper Yellow River, China. Chin. Geogr. Sci.34 (4), 674–688. 10.1007/s11769-024-1445-6LinT.WuD.YangM.MaP.LiuY.LiuF. (2022). Evolution and simulation of terrestrial ecosystem carbon storage and sustainability assessment in karst areas: a case study of Guizhou Province. Int. J. Environ. Res. Public Health19 (23), 16219. 10.3390/ijerph19231621936498292LiuX.LiangX.LiX.XuX.OuJ.ChenY. (2017). A future land use simulation model (FLUS) for simulating multiple land use scenarios by coupling human and natural effects. Landsc. Urban Planning168, 94–116. 10.1016/j.landurbplan.2017.09.019LiuY.DuM.YangL.CuiQ.LiuY.LiX. (2025). Mitigation policies interactions delay the achievement of carbon neutrality in China. Nat. Clim. Change15 (2), 147–152. 10.1038/s41558-024-02237-2LutongW.XiaodanL.HanqingG.WanlaiX.KexuanL.LeiL. (2025). Spatio-temporal succession and prediction of carbon storage in ecological conservation area of western Beijing based on InVEST-PLUS model. Geol. Bull. China44 (07), 1272–1284.LyuK.LiZ. (2025). Scenario based assessment of carbon storage and habitat quality under land use change in Shandong Province China. Sci. Rep.15 (1), 38098. 10.1038/s41598-025-25097-y41174022LyuR.MiL.ZhangJ.XuM.LiJ. (2019). Modeling the effects of urban expansion on regional carbon storage by coupling SLEUTH‐3r model and InVEST model. Ecol. Res.34 (3), 380–393. 10.1111/1440-1703.1278MareyA.WangL.GoubranS.GaurA.LuH.LeroyerS. (2024). Forecasting urban land use dynamics through patch-generating land use simulation and Markov chain integration: a multi-scenario predictive framework. Sustainability16 (23), 10255. 10.3390/su162310255MutaleB.QiangF. (2024). Modeling future land use and land cover under different scenarios using patch-generating land use simulation model. A case study of Ndola district. Front. Environ. Sci.12, 1362666. 10.3389/fenvs.2024.1362666NayakR. K.PatelN.DadhwalV. (2010). Estimation and analysis of terrestrial net primary productivity over India by remote-sensing-driven terrestrial biosphere model. Environ. Monitoring Assessment170 (1), 195–213. 10.1007/s10661-009-1226-919908154NiC.MaoF.DuH.LiX.XuY.HuangZ. (2025). Spatiotemporal evolution and influencing factors of Forest carbon storage based on BIOME-BGC model and geographical detector in eight basins of Zhejiang Province in China. For. (19994907)16 (2), 316. 10.3390/f16020316NieQ.WuG.LiL.ManW.MaJ.BaoZ. (2024). Exploring scaling differences and spatial heterogeneity in drivers of carbon storage changes: a comprehensive geographic analysis framework. Ecol. Indic.165, 112193. 10.1016/j.ecolind.2024.112193NowakD. J.CraneD. E. (2002). Carbon storage and sequestration by urban trees in the USA. Environ. Pollution116 (3), 381–389. 10.1016/s0269-7491(01)00214-711822716NowakD. J.GreenfieldE. J.HoehnR. E.LapointE. (2013). Carbon storage and sequestration by trees in urban and community areas of the United States. Environ. Pollution178, 229–236. 10.1016/j.envpol.2013.03.01923583943PanT.KuangW.PanR.NiuZ.DouY. (2022). Hierarchical urban land mappings and their distribution with physical medium environments using time series of land resource images in Beijing, China (1981–2021). Remote Sens.14 (3), 580. 10.3390/rs14030580PanY.BirdseyR. A.PhillipsO. L.HoughtonR. A.FangJ.KauppiP. E. (2024). The enduring world forest carbon sink. Nature631 (8021), 563–569. 10.1038/s41586-024-07602-x39020035PenmanJ.GytarskyM.HiraishiT.KrugT.KrugerD.PipattiR. (2003). Good practice guidance for land use, land-use change and forestry. Japan: Institute for Global Environmental Strategies (IGES) for the IPCC.PiaoS.FangJ.CiaisP.PeylinP.HuangY.SitchS. (2009). The carbon balance of terrestrial ecosystems in China. Nature458 (7241), 1009–1013. 10.1038/nature0794419396142PickardB. R.Van BerkelD.PetrasovaA.MeentemeyerR. K. (2017). Forecasts of urbanization scenarios reveal trade-offs between landscape change and ecosystem services. Landsc. Ecol.32 (3), 617–634. 10.1007/s10980-016-0465-8SallustioL.QuatriniV.GenelettiD.CoronaP.MarchettiM. (2015). Assessing land take by urban development and its impact on carbon storage: findings from two case studies in Italy. Environ. Impact Assess. Rev.54, 80–90. 10.1016/j.eiar.2015.05.006SarathchandraC.Alemu AbebeY.WorthyF. R.Lakmali WijerathneI.MaH.YingfengB. (2021). Impact of land use and land cover changes on carbon storage in rubber dominated tropical Xishuangbanna, South West China. Ecosyst. Health Sustain.7 (1), 1915183. 10.1080/20964129.2021.1915183SetoK. C.GüneralpB.HutyraL. R. (2012). Global forecasts of urban expansion to 2030 and direct impacts on biodiversity and carbon pools. Proc. Natl. Acad. Sci. U. S. A.109 (40), 16083–16088. 10.1073/pnas.121165810922988086SharmaR.PradhanL.KumariM.BhattacharyaP.MishraV. N.KumarD. (2024). Spatio-temporal assessment of urban carbon storage and its dynamics using InVEST model. Land13 (9), 1387. 10.3390/land13091387ShiX.XiaH.MachimuraT.MatsuiT.HagaC.WangQ. (2024). Scenario-based land use simulation and integrated analysis of karst ecosystem service bundles. Glob. Ecol. Conservation54, e03096. 10.1016/j.gecco.2024.e03096van VlietJ.BregtA. K.Hagen-ZankerA. (2011). Revisiting Kappa to account for change in the accuracy assessment of land-use change models. Ecol. Modelling222 (8), 1367–1375. 10.1016/j.ecolmodel.2011.01.017WangS. Q.ZhengX.ZangX. (2012). Accuracy assessments of land use change simulation based on Markov-cellular automata model. Procedia Environ. Sci.13, 1238–1245. 10.1016/j.proenv.2012.01.117WangR.DerdouriA.MurayamaY. (2018). Spatiotemporal simulation of future land use/cover change scenarios in the Tokyo metropolitan area. Sustainability10 (6), 2056. 10.3390/su10062056WangB.LiaoJ.ZhuW.QiuQ.WangL.TangL. (2019). The weight of neighborhood setting of the FLUS model based on a historical scenario: a case study of land use simulation of urban agglomeration of the Golden triangle of Southern Fujian in 2030. Acta Ecol. Sin.39 (12), 4284–4298.WangZ.GuoM.ZhangD.ChenR.XiC.YangH. (2023). Coupling the calibrated GlobalLand30 data and modified PLUS model for multi-scenario land use simulation and landscape ecological risk assessment. Remote Sens.15 (21), 5186. 10.3390/rs15215186WangP.LiuC.DaiL. (2024a). Spatiotemporal variation and prediction of carbon storage in terrestrial ecosystems at multiple development stages in Beijing city based on the plus and integrated valuation of ecosystem services and tradeoffs models. Land13 (9), 1544. 10.3390/land13091544WangX.ZhangW.ZhaoX.WangD.LiY. (2024b). Sustaining carbon storage: an analysis of land use and conservation strategies in China’s Huang-Huai-Hai Plain. Sustainability17 (1), 139. 10.3390/su17010139XuQ.LiK. (2024). Multi-Scenario simulation of land use and assessment of carbon stocks in terrestrial ecosystems based on SD-PLUS-InVEST coupled modeling in Nanjing city. Forests15 (10), 1824. 10.3390/f15101824YeY.LaiM.DongM.LiZ.YuanJ.LyuJ. (2025). Coupling the PLUS-InVEST model for multi-scenario land use simulation and carbon storage assessment in Northern Anhui, China. Sustainability17 (9), 4185. 10.3390/su17094185YuanL.YulingP.HaonaP.WeiyingC. (2025a). Spatiotemporal evolution of carbon storage and multi-scenario prediction in the Yunnan-Guizhou Plateau based on the InVSET-Ridge Regression-PLUS model. Environ. Sci., 1–16. 10.13227/j.hjkx.202501105YuanX.BaiX.ZhouZ.LuoG.LiJ.RanC. (2025b). Global impacts of land use on terrestrial carbon emissions since 1850. Sci. Total Environ.963, 178358. 10.1016/j.scitotenv.2024.17835839818149YuxinZ.PeihongJ.JiaxuanC.XiangtingC. (2024). Simulation and analysis of land use and multi-scenario carbon sink in the upper Yangtze river. China Land Sci.38 (11), 129–140.ZhangJ.ZhangY. (2024). Quantitative assessment of the impact of the three-North shelter forest program on vegetation net primary productivity over the past two decades and its environmental benefits in China. Sustainability16 (9), 3656. 10.3390/su16093656ZhangZ.LiN.WangX.LiuF.YangL. (2016). A comparative study of urban expansion in Beijing, Tianjin and Tangshan from the 1970s to 2013. Remote Sens.8 (6), 496. 10.3390/rs8060496ZhangX.WangF.LuW.LiS.ZhengJ. (2022). Gridded reconstruction of the population in the traditional cultivated region of China from 1776 to 1953. Sci. China Earth Sci.65 (2), 365–378. 10.1007/s11430-020-9866-2ZhangH.LuoJ.WuJ.DongH. (2024a). Dynamic response of carbon storage to future land use/land cover changes motivated by policy effects and core driving factors. J. Plant Ecol.17 (5), rtae042. 10.1093/jpe/rtae042ZhangY.LinJ.HuangY.ChenZ.ZhuC.YuanH. (2024b). Delineation of urban growth boundary based on FLUS model under the perspective of land use evaluation in hilly mountainous areas. J. Mt. Sci.21 (5), 1647–1662. 10.1007/s11629-023-8499-2ZhangY.NaerkeziN.ZhangY.WangB. (2024c). Multi-scenario land use/cover change and its impact on carbon storage based on the coupled GMOP-PLUS-InVEST model in the Hexi Corridor, China. Sustainability16 (4), 1402. 10.3390/su16041402ZhangJ.CaoP.RoosliR. (2025a). Assessing land use and carbon storage changes using PLUS and InVEST models: a multi-scenario simulation in Hohhot. Environ. Sustain. Indic.26, 100655. 10.1016/j.indic.2025.100655ZhangY.DuS.ZhuL.GuoT.ZhaoX.GuoJ. (2025b). Sub-district level spatiotemporal changes of carbon storage and driving factor analysis: a case study in beijing. Land14 (1), 151. 10.3390/land14010151ZhangY.TangJ.HuX.ChenC.LuoZ.LiQ. (2025c). Understanding the spatiotemporal patterns and drivers of carbon stock in central-southern China’s hilly regions through land use change and scenario simulation. Sustainability17 (12), 5578. 10.3390/su17125578ZhaoH.GaiY.WangF.CuiL.DuY.WangH. (2025). Investigation into the supply-demand relationship of carbon sequestration in the yellow river basin using the optimal parameter geographical detector model. Sci. Rep.15 (1), 29079. 10.1038/s41598-025-15298-w40781535ZhenY.ZhangX.ZhangC.GaoQ.DongJ.ZhangL. (2023). Effects of climate change and land use/cover changes on carbon sequestration in forest ecosystems in the coastal area of China. Front. For. Glob. Change6, 1271239. 10.3389/ffgc.2023.1271239ZhouM.LuL.GuoH.WengQ.CaoS.ZhangS. (2021). Urban sprawl and changes in land-use efficiency in the Beijing–Tianjin–Hebei region, China from 2000 to 2020: a spatiotemporal analysis using earth observation data. Remote Sens.13 (15), 2850. 10.3390/rs13152850ZhouH.TangM.HuangJ.MeiX.ZhaoH. (2024). Driving mechanisms and multi-scenario simulation of land use change based on National Land Survey Data: a case in Jianghan Plain, China. Front. Environ. Sci.12, 1422987. 10.3389/fenvs.2024.1422987ZhouH.TangM.HuangJ.ZhangJ.HuangJ.ZhaoH. (2025a). Instability and uncertainty of carbon storage in karst regions under land use change: a case study in Guiyang, China. Front. Environ. Sci.13, 1551050. 10.3389/fenvs.2025.1551050ZhouJ.JohnsonV. C.ShiJ.TanM. L.ZhangF. (2025b). Multi-scenario land use change simulation and spatial-temporal evolution of carbon storage in the Yangtze River Delta region based on the PLUS-InVEST model. PloS One20 (1), e0316255. 10.1371/journal.pone.031625539854555ZhuJ.HuX.YaoH.LiuG.JiC.FangJ. (2015). A significant carbon sink in temperate forests in Beijing: based on 20-year field measurements in three stands. Sci. China Life Sci.58 (11), 1135–1141. 10.1007/s11427-015-4935-z26501378ZhuL.SongR.SunS.LiY.HuK. (2022). Land use/land cover change and its impact on ecosystem carbon storage in coastal areas of China from 1980 to 2050. Ecol. Indic.142, 109178. 10.1016/j.ecolind.2022.109178
Edited by:Zhiheng Yang, Shandong University of Finance and Economics, China
Reviewed by:Cai Jinghang, Guangdong Ocean University, China
Gao Wei, Capital University of Economics and Business, China