Spatiotemporal dynamics and influencing factors of land carbon stock in Chengdu Plain using an integrated model

Abstract

Understanding land carbon stock dynamics is essential for sustainable land use and ecological conservation amid rapid urbanisation. This study investigates how land use changes contribute to carbon sequestration, offering insights to support China’s carbon peaking (2030) and carbon neutrality (2060) goals. Using high-resolution land use data (30 m) from 2000 to 2020 for the Chengdu Plain region, derived via Google Earth Engine and Random Forest classification, the Patch-generating Land Use Simulation (PLUS) model was applied to predict land use changes under four scenarios: natural development scenario (NDS), ecological protection scenario (EPS), cultivated land preservation scenario (CLDS), and economic development scenario (EDS) for 2030 and 2060. Carbon stock dynamics were quantified using the InVEST model, while the Optimised Parameter Geographical Detector (OPQD) model identified key drivers and their interactions. Between 2000 and 2020, cropland decreased by 4.14% while construction land increased by 4.15%, reflecting rapid urban expansion. Scenario simulations predict further cropland loss (2.80%–7.44%) and substantial construction land growth (26.89%–39.95%) by 2060, with forest and grassland recovery only under conservation scenarios. Carbon stock declined by 5.1%–5.5%, with the EPS and CLDS scenarios mitigating losses, while the NDS and EDS scenarios caused significant declines. Anthropogenic factors, such as urbanisation and economic growth, had a greater impact (> 15%) on carbon stock than natural factors (< 4%), with their interactions exhibiting nonlinear enhancement effects.This study underscores the benefits of conservation strategies and provides actionable insights for climate change mitigation, carbon trading, and sustainable urban planning. Further exploration of additional factors and predictive refinements will enhance regional ecological conservation efforts.

Introduction

Amid the escalating challenges of global climate change, reducing greenhouse gas emissions and achieving carbon neutrality have become urgent priorities worldwide. Soil and vegetation carbon stocks are critical components of ecosystem carbon sinks, playing a pivotal role in sustaining regional carbon cycles and ensuring ecological security¹. Recent research highlights that land use and land cover changes (LUCC) significantly influence regional carbon stocks, ranking as the second-largest global source of carbon emissions after fossil fuel combustion. This effect is particularly pronounced in rapidly urbanising and agriculturally expanding regions, where land use transitions substantially impact carbon dynamics^2,3. Given this, investigating land-use carbon stocks is crucial for China in its pursuit of greenhouse gas reductions and carbon neutrality, while also promoting the sustainable development of regional ecosystems ^4,5.

Extensive research has been conducted to understand the role of land-use changes in carbon stock dynamics. Remote sensing imagery and geographic information systems (GIS) are commonly employed to monitor these changes, revealing that urbanisation reduces regional carbon stocks, especially through the conversion of agricultural lands to urban areas⁶. Carbon stocks across different land use types have been evaluated using field sampling and model simulations, with findings consistently indicating that forests and grasslands sequester significantly more carbon than agricultural or urban areas. This underscores the importance of protecting these ecosystems⁷. Additionally, studies using the InVEST model have highlighted urban expansion and agricultural intensification as major contributors to increased carbon emissions, reinforcing the need for more sustainable land-use strategies⁸. Other research integrating ecosystem service frameworks has emphasised the dual impact of land use on both ecosystem services and carbon stocks, advocating for the inclusion of ecosystem service values in land-use planning⁹. Moreover, the Geodetector model has been applied to examine the combined effects of climate change and human activities on carbon stocks, suggesting the need for long-term carbon monitoring to inform policy¹⁰. These studies provide a robust theoretical foundation for understanding land-use carbon stocks in the Chengdu Plain region.

Despite these advancements, several gaps remain. Comprehensive assessments of the diverse impacts of land-use changes on carbon stocks are still limited, potentially leading to inaccurate carbon stock evaluations and undermining the scientific basis for effective policymaking⁹. Furthermore, many studies fail to incorporate long-term monitoring, which is essential for understanding dynamic changes and identifying underlying drivers, especially in rapidly urbanising regions like Chengdu⁸. Previous research has predominantly focused on ecological factors, neglecting the significant role of socioeconomic drivers such as population growth, policy shifts, and economic activities, all of which shape land-use patterns and influence carbon stocks. This omission introduces potential biases into carbon stock assessments¹¹. Furthermore, studies that examine the interactions between natural and anthropogenic factors on carbon stock dynamics are still relatively rare. The Chengdu Plain region, characterised by its complex ecosystem and significant regional variability in topography, soil types, and climate, requires more focused studies to accurately estimate its carbon stocks¹².

The Chengdu Plain region, an agricultural and densely populated area in Sichuan Province, is one of the most rapidly urbanising regions in Southwest China. Frequent land-use changes in this region have profound effects on its carbon stocks¹³.Thus, investigating the spatiotemporal dynamics of carbon stocks in the Chengdu Plain region is essential for assessing the region’s carbon sink capacity, addressing climate change, and achieving national carbon peaking and carbon neutrality goals¹⁴.

Recent advancements in geospatial technologies have revolutionized the study of land-use and land-cover changes. Among these, Google Earth Engine (GEE) stands out for its extensive remote sensing imagery and robust machine learning capabilities. By utilizing the random forest algorithm with Landsat 5/8 imagery, GEE enables precise land use classification and provides detailed spatiotemporal analyses of carbon stock dynamics¹¹. This platform facilitates the integration of large-scale geospatial datasets with powerful analytical tools, making it an ideal solution for monitoring carbon dynamics across expansive landscapes¹⁵.

The Patch-generating Land Use Simulation (PLUS) model, which integrates both geographic and socioeconomic factors, allows for simulating future land-use scenarios. It offers valuable insights into how these scenarios may influence carbon stock changes⁵.

The Patch-generating Land Use Simulation (PLUS) model, which incorporates both geographic and socioeconomic factors, allows for simulating future land-use scenarios, offering valuable insights into how these changes may influence carbon stock dynamics⁵. By considering the impacts of population growth, policy shifts, and economic development, the PLUS model provides realistic land-use trajectories, laying the groundwork for assessing potential future carbon sequestration outcomes¹⁶. This model is particularly useful for understanding the long-term implications of urban expansion, agricultural intensification, and ecological conservation on carbon stocks¹⁷.

The Integrated Valuation of Ecosystem Services and Trade-offs (InVEST) model further enhances our understanding by estimating carbon stocks across various land-use types and quantifying spatial variations in ecosystem carbon sequestration¹⁸. By calculating the carbon storage potential within different land-use categories, InVEST aids in assessing the role of ecosystem services in climate change mitigation. It also highlights areas where conservation efforts can optimize carbon sequestration, providing valuable insights for policymakers aiming to prioritize land-use decisions that promote sustainability while minimizing carbon emissions^19,20.

The Optimal Parameterised Geographical Detector (OPGD) model presents several advantages over traditional statistical and spatial analysis methods. By automating parameter optimisation, OPGD identifies the primary drivers of carbon stock changes more accurately, reducing biases associated with subjective parameter selection. Additionally, it accounts for nonlinear relationships and interactions between factors, making it highly suitable for analysing spatial heterogeneity and carbon stock variability^21–24. In this study, the application of OPGD improves our understanding of the complex interactions between land-use changes, climate variables, and socioeconomic drivers, providing a more nuanced perspective on the factors influencing carbon stock dynamics.

This study presents four future land-use scenarios to evaluate how policy interventions may influence land-use changes, carbon storage, and ecosystem stability in the Chengdu Plain. These scenarios serve as a framework to assess the impact of different policies on ecosystem functions and carbon storage, guiding sustainable land-use management strategies. The study also examines how to balance ecological protection and carbon storage under various policy scenarios. The Natural Development Scenario (NDS) assumes that land-use changes are driven by natural factors such as climate change, soil degradation, and population growth, with no policy intervention. This scenario evaluates the role of environmental factors in shaping land-use trends and carbon storage²⁵. The Cultivated Land Protection Scenario (CLPS) limits the conversion of arable land to ensure food security and ecological stability, exploring the long-term benefits of agricultural land preservation for land-use patterns and ecosystem functions²⁶. The Ecological Protection Scenario (EPS) enforces a strict ecological red line policy to protect sensitive areas and promote ecological restoration, assessing how ecological protection measures impact carbon storage and ecosystem restoration²⁷. Lastly, the Economic Development Scenario (EDS) focuses on urbanization and industrial expansion, analyzing how economic growth pressures land-use changes and exploring how to balance ecological protection with development²⁸. These scenarios provide a scientific basis for future land-use planning, carbon storage management, and ecological protection in the Chengdu Plain, offering valuable insights for policymakers aiming to balance environmental goals with socio-economic development.

To achieve these objectives, this study adopts an integrated PLUS-InVEST-OPGD model framework. Specifically, it aims to: (1) classify Landsat imagery using the GEE random forest algorithm to obtain land-use information for the Chengdu Plain region; (2) simulate future land-use changes under four distinct scenarios—natural development, ecological protection, farmland preservation, and economic development—using the PLUS model; (3) quantify carbon stocks and analyze their spatial and temporal distribution using the InVEST model; and (4) identify key drivers of carbon stock changes and explore their interactions using the OPGD model. Through these objectives, this research seeks to provide robust data for carbon stock management, optimize ecosystem services, and inform land-use decision-making, supporting the Chengdu Plain’s progression toward carbon peaking and carbon neutrality.

Study area

The Chengdu Plain region, situated at the western edge of the Sichuan Basin in southwestern China, spans geographic coordinates 103°–105°E and 29.5°–32°N, covering an area of approximately 18,000 km² (Fig. 1). This region encompasses Chengdu and parts of Mianyang, Deyang, Meishan, Leshan, and Ya’an. The landscape is characterized by a variety of topographies, from the fertile plains at the center to the surrounding hills and mountains, which influence both vegetation patterns and carbon stock dynamics²⁹. The climate is classified as humid subtropical monsoon, with annual rainfall ranging between 1,200–1,600 mm. These climatic conditions, alongside fertile alluvial soils, support high agricultural productivity and contribute to the region’s biodiversity³⁰. The terrain slopes gently from northwest to southeast, which affects water drainage, soil composition, and the overall functioning of local ecosystems³¹.

Fig. 1.

Location of the study area: (a) The Chengdu Plain region’s location within China and Sichuan Province. (b) Administrative divisions within the Chengdu Plain region. (c) Digital Elevation Model (DEM) of the Chengdu Plain region .(d) Land cover types and spatial distribution patterns within the Chengdu Plain region.(The figure was generated using ArcGIS 10.3 software, with the URL link: https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview.)

The Chengdu Plain is a major economic engine for Sichuan Province, contributing over 40% of the province’s GDP and housing more than 30% of its population³². Chengdu itself is a vibrant economic hub, known for its advanced industries and rapid urbanization. However, these developmental trends have led to significant land use changes, including the conversion of farmland into urban and industrial areas. Such transitions have placed pressure on local ecological systems, altering both land cover and carbon sequestration capacity³². As part of the Yangtze River Basin’s ecological conservation zone, the Chengdu Plain faces the critical challenge of balancing economic growth with the need for environmental sustainability³³. The intensifying urbanization, coupled with agricultural intensification, has resulted in diminishing carbon stocks and growing instability within ecosystems³³. This study aims to assess the impacts of these land use changes on carbon sequestration within the Chengdu Plain, focusing on identifying strategies to enhance ecological security and promote sustainable regional development³⁴.

Research methodology

Data sources and processing

This study integrates multi-source data, including remote sensing imagery, natural factors, economic data, accessibility measures, constraint layers, and carbon density information (Table 1). The details of these data and their processing are as follows:

• 1) Remote sensing imagery

Table 1.

Data source and processing.

Types	Indicator	Resolution	Data source
Remote sensing imagery	Land use data	30 m	USGS Landsat 5/8 using GEE random forest classification
Natural factors	Elevation	30 m	Resource and Environment Data Centre, Chinese Academy of Sciences (http://www.resdc.cn)
	Slope	30 m	Derived from DEM
	Aspect	30 m	Derived from DEM
	Annual precipitation	1 km	http://www.resdc.cn)
	Annual temperature	1 km	http://www.resdc.cn
Economic factors	Population density	1 km	http://www.resdc.cn
	GDP	1 km	http://www.resdc.cn
Accessibility factors	Distance to water	30 m	Calculated with OSMnx (Python package)
	Distance to highway	30 m	OSMnx (Python package)
	Distance to railway	30 m	OSMnx (Python package)
	Distance to airport	30 m	OSMnx (Python package)
Constraint factors	Ecological redline	30 m	Vectorised from published maps
	Basic farmland	30 m	Digitized from published maps
Carbon density	t C/ha	30 m	Table 4

Land-use data for 2000, 2005, 2010, 2015, and 2020 were derived from Landsat 5/8 imagery. To minimize the impact of cloud cover and frequent rainfall in the Chengdu Plain region, imagery was primarily acquired between March and October. The land-use classification was performed using the random forest method in GEE. The classification process included parameter tuning and validation using ground truth data, achieving an overall accuracy exceeding 85%. The classified maps were used to analyze land-use dynamics and their impact on carbon stocks.

• 2) Natural factors

Natural factors, including digital elevation model (DEM), annual precipitation, and mean annual temperature, were obtained from the Data Centre for Resources and Environmental Sciences, Chinese Academy of Sciences. The annual spatial interpolation dataset of meteorological elements is created using the annual mean meteorological data (such as precipitation and temperature) from over 2,400 meteorological stations across China, with the Anuspl interpolation software. This method utilizes a thin plate spline function for spatial interpolation and allows the inclusion of covariates (typically elevation), which enhances the interpolation accuracy and produces smoother results. Given the strong correlation between the distribution of meteorological elements and elevation, incorporating elevation as a covariate significantly improves accuracy, especially in capturing the zonal distribution patterns of meteorological elements. In the interpolation process, longitude, latitude, and elevation are used as independent variables. Based on the annual mean data from meteorological stations nationwide, a 1 km grid dataset of annual meteorological elements has been generated from 1960 onward³⁵. Slope and aspect were derived from the DEM using GIS tools. Precipitation and temperature data were interpolated to match the spatial resolution of the DEM. These factors were used to evaluate their influence on land-use patterns and carbon density variability across the Chengdu Plain region.

• 3) Economic data

Economic indicators, such as GDP and population density, were sourced from the Resource and Environment Data Centre of the Chinese Academy of Sciences. These data were processed to evaluate the relationship between economic activities and land-use changes, focusing on the impact of urban expansion on carbon stocks.

• 4) Accessibility factors

Road network data were extracted from OpenStreetMap using the Python OSMnx package. Road density and route distances were calculated to assess accessibility across the study area. These measures were incorporated into the analysis of human activities and their impact on land-use transitions.

• 5) Constraint factors

Constraint layers, including ecological protection zones and basic farmland (ecological redlines), were digitized from official maps published by the Sichuan Provincial Government. These layers were georeferenced, vectorized, and processed to ensure compatibility with other spatial datasets. They were used to identify areas where land-use changes are restricted and to assess the impact of policy-driven land-use constraints on carbon stocks.

• 6) Carbon density

Carbon density data were initially sourced from national datasets and adjusted to account for regional variations in temperature and precipitation based on established correction coefficients. These coefficients were derived from studies focusing on the Chengdu Plain region and similar subtropical monsoon climate regions. The adjusted carbon density data were resampled to a spatial resolution of 30 m, consistent with other datasets, and were used to estimate carbon stocks under different land-use scenarios.

• 7) Data projection and processing

All spatial data were projected using the Krasovsky_1940_Albers projection to ensure uniformity. Spatial resolution was standardized to 30 m, and data pre-processing steps, including cloud masking for remote sensing imagery and alignment of multi-source datasets, were applied to ensure data quality and compatibility.

Land use classification standards

The land use classification system used in this study follows the secondary classification of the China National Land Use and Cover Change (CNLUCC) dataset³⁶, rather than globally recognised systems like CORINE or SIOSE, for several important reasons. The CNLUCC system is specifically designed to reflect the unique land use and cover characteristics of China, accounting for the country’s diverse geographical, socio-economic, and environmental conditions. In contrast, global systems may not capture the specific land use patterns or environmental conditions of China as effectively. Additionally, the CNLUCC classification structure aligns closely with China’s land policies and development needs, making it better suited for domestic research and land management³⁷.

The CNLUCC classification system divides land use into six primary categories: cropland, forest, grassland, water bodies, construction land, and unused land. These categories are further subdivided into 23 specific types, allowing for a more detailed and precise description of land use patterns (Table 2).This level of detail is particularly valuable in regions with diverse geographical environments and land use characteristics. In contrast, systems like CORINE, while suitable for Europe, may not adequately address the specific demands and environmental conditions found across different regions of China³⁸. By adopting the CNLUCC classification system, this study ensures consistency with local standards and facilitates better integration with national land resource management, environmental monitoring, and policy-making databases³⁶.

Table 2.

Classification system of land use/land cover remote sensing monitoring data.

Primary Classification	Code	Name	Secondary code & name	Description
1. Cropland	1	Cultivated land	11 Paddy Field, 12 Dryland	Land used for growing crops
2. Forest	2	Forestland	21 Woodland, 22 Shrubland, 23 Sparse forest, 24 Other forestland	Land with trees, shrubs, bamboo, coastal mangroves, etc
3. Grassland	3	Grassland	31 High vegetation coverage, 32 Medium vegetation voverage, 33 Low vegetation coverage	Land covered by herbaceous plants, grassland coverage > 5%
4. Water Bodies	4	Water area	41 Rivers/Canals, 42 Lakes, 43 Reservoirs/Ponds, 44 Permanent ice/snow, 45 Mudflats, 46 Beaches	Natural water bodies and water management facilities
5. Construction land	5	Construction Land	51 Urban land, 52 Rural residential area, 53 Other construction land	Land for urban and rural settlements, industrial and transportation use
6. Unused Land	6	Unused land	61 Sandy land, 62 Gobi desert, 63 Saline-Alkali land, 64 Bare land, 65 Bare rocky land, 66 Other

GEE random forest classification method

To extract land use information for the Chengdu Plain region in 2000, 2005, 2010, 2015, and 2020, this study employs the Random Forest classification method integrated with the GEE platform. Random Forest, an ensemble learning algorithm, is widely recognized for its robustness in handling high-dimensional datasets and complex nonlinear relationships. By constructing multiple independent decision trees and aggregating their outputs through majority voting, Random Forest effectively mitigates overfitting, enhancing both classification accuracy and model stability^39,40.

In the context of remote sensing, Random Forest is particularly advantageous for processing large-scale datasets and high-dimensional features, such as multiple spectral bands and vegetation indices. The GEE platform further strengthens this capability by providing cloud-based computational power and extensive geospatial datasets, enabling efficient processing and analysis of multi-temporal Landsat imagery^41,42.

For this study, Landsat 5 and Landsat 8 imagery covering March to October were selected to minimize the impact of cloud cover. The Random Forest model was trained using a combination of ground truth samples and ancillary data, including spectral bands, vegetation indices (e.g., NDVI, EVI), and texture features. Key parameters, such as the number of decision trees and maximum tree depth, were optimized through cross-validation to ensure the model’s reliability and generalization.

The classification workflow included the following steps: (1)training sample selection: representative training samples for land use classes were manually delineated based on field survey data and high-resolution imagery. (2) feature selection: spectral bands, vegetation indices, and topographic features (e.g., slope, aspect) were selected to capture the region’s diverse land cover characteristics^40,43. (3)model training and validation: the random forest model was trained and validated using stratified random sampling, achieving an overall accuracy above 85% and a kappa coefficient exceeding 0.8^39,43.(4)post-classification refinement: misclassified pixels were corrected using spatial filters and ancillary data to improve classification accuracy.

The integration of Random Forest with GEE significantly reduced computational costs and improved operational efficiency, enabling the rapid processing of large-scale and multi-temporal datasets. This approach was particularly effective in addressing the Chengdu Plain region’s complex land use patterns, characterized by a mix of urban, agricultural, and natural landscapes. The resulting land use maps provided a robust foundation for analyzing spatiotemporal dynamics and their ecological impacts^41,42.

Patch-generating land use simulation (PLUS)

The Patch-generating Land Use Simulation (PLUS) model, developed by Liang Xun et al. at the High-Performance Spatial Computational Intelligence Laboratory (China University of Geosciences, Wuhan), combines a Land Expansion Analysis Strategy (LEAS) and a Cellular Automata model with Random Seeds (CARS). This integration enables the model to effectively identify key drivers of land expansion and landscape transformation. Compared to traditional models, PLUS not only achieves higher simulation accuracy but also generates more realistic landscape patterns, which has led to its widespread application in land use change prediction^44–47.

The PLUS model consists of two core modules: the Transfer Probability Module and the Patch Generation Module. The Transfer Probability Module employs LEAS to quantify the probability of land use transitions by incorporating both natural and anthropogenic driving factors, offering a robust mechanism for identifying transition rules. Meanwhile, the Patch Generation Module utilizes CARS to simulate the spatial distribution of new land use patches, ensuring that the resulting patterns maintain spatial contiguity and ecological realism⁴⁸.

By integrating these two modules, PLUS bridges the gap between statistical analysis and spatial simulation, making it an advanced tool for understanding and predicting complex land use dynamics. Future studies could explore its potential for coupling with climate models or assessing ecosystem services to further enhance its application scope.

Future multi-scenario setup

To simulate future land use changes in the Chengdu Plain region more scientifically and systematically, this study integrates two key policy frameworks: the Ecological Protection Red Line Plan of Sichuan Province and the Farmland and Permanent Basic Farmland Protection Plan (2021–2035). These plans guide adjustments to the probabilities, intensity, and spatial direction of land conversions, ensuring that the simulation outcomes align with policy objectives, ecological protection priorities, and actual land use demands in Chengdu, Mianyang, Deyang, Meishan, Leshan, and Ya’an. Four future land use scenarios are constructed for 2030 and 2060 based on varying degrees of policy intervention:

• (1) Natural development scenario(NDS): This baseline scenario assumes no policy intervention, enabling an assessment of land use changes driven solely by natural conditions and inherent trends.

• (2) Cultivated land protection scenario (CLPS): In this scenario, permanent basic farmland protection zones are established to strictly regulate the conversion of farmland to other land types. This reflects the impact of farmland protection policies on ensuring food security and maintaining regional ecological stability⁴⁹.

• (3) Ecological protection scenario (EPS): This scenario incorporates ecological red line areas as land use restriction zones. By prioritizing green space preservation and ecological restoration, EPS promotes sustainable land use and biodiversity conservation⁵⁰.

• (4) Economic development scenario (EDS): Focused on urbanization and industrial expansion, this scenario evaluates the impact of economic growth and urban development demands on land use dynamics. It highlights the potential trade-offs between rapid development and ecological sustainability⁵¹.

The parameters for these scenarios, including land use transfer probabilities and intensity adjustments, are derived from policy guidelines, land use demand forecasts, and regional ecological assessments. Table 3 presents the land use transfer matrices for the Chengdu Plain region under the multi-scenario simulations for 2030 and 2060.

Table 3.

Land use transfer matrix in Chengdu Plain region under multi-scenario simulation in 2030 and 2060.

Scenarios simulate	Land use types	Cultivated land	Forest land	Grassland	Water area	Construction land	Unused land
NDS	Cultivated land	1	1	1	1	1	1
	Forest land	1	1	1	0	1	0
	Grassland	1	1	1	1	1	1
	Water area	1	0	1	1	1	1
	Construction land	1	0	1	1	1	1
	Unused land	1	0	1	1	1	1
EPS	Cultivated land	1	1	1	1	1	1
	Forest land	0	1	0	0	0	0
	Grassland	0	1	1	1	0	0
	Water area	1	1	1	1	0	0
	Construction land	1	0	1	1	1	0
	Unused land	1	0	1	1	1	1
CLPS	Cultivated land	1	0	0	0	0	0
	Forest land	1	1	1	0	1	0
	Grassland	1	1	1	1	1	1
	Water area	1	0	1	1	1	1
	Construction land	1	0	1	1	1	1
	Unused land	1	0	1	1	1	1
EDS	Cultivated land	1	1	1	1	1	1
	Forest land	1	1	1	0	1	0
	Grassland	1	1	1	1	1	1
	Water area	1	0	1	1	1	1
	Construction land	0	0	0	0	1	0
	Unused land	1	0	1	1	1	1

In Table 3, a value of “1” signifies that the land use type can transition to the specified other type, whereas “0” indicates it cannot. Under the NDS, all land use types demonstrate high conversion flexibility, reflecting a natural trend without policy constraints. In the EPS, the conversion of forest and grassland to other types (particularly construction land) is strictly limited to prioritise ecological land protection. The CLPS strictly restricts the conversion of cultivated land to protect food security and maintain ecological stability. In the EDS, the expansion of construction land is prioritised, permitting the conversion of cultivated land and other types into construction land, while limiting the conversion of construction land back to ecological land.

By comparing the outcomes across these scenarios, this study provides valuable insights into the interplay between economic development and ecological protection. The results aim to inform policymakers by offering evidence-based recommendations for balancing urban growth with sustainable land management in the Chengdu Plain region.

Estimation of carbon stock using the InVEST model

The Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model is a widely used spatial evaluation tool that quantifies ecosystem services in terms of both volume and economic value, offering critical insights for policymakers seeking to balance human activities with environmental sustainability⁵². Among its modules, the carbon stock module is particularly suited for assessing regional carbon dynamics due to its integration of spatially explicit land use data and detailed carbon pool calculations.

In this study, the InVEST model was selected for its ability to provide a comprehensive evaluation of carbon stocks by incorporating four primary carbon pools: aboveground biomass, belowground biomass, dead organic matter, and soil carbon. Using land use types as the basic measurement units, the model estimates carbon stocks by multiplying the area of each land type by its corresponding average carbon density. This approach enables the quantification of total carbon storage, changes in carbon stocks, and net carbon sequestration over the study period, providing a scientific basis for evaluating ecosystem service functions and supporting land use optimisation⁵³.

Compared to other carbon assessment models, the InVEST model offers significant advantages, including its modular design, ease of integration with diverse datasets, and capacity to link carbon stock estimations with broader land use planning goals. In the context of the Chengdu Plain region, the model enables an in-depth analysis of carbon dynamics under different land use scenarios, aligning with policy objectives to mitigate carbon emissions and enhance ecosystem resilience⁵⁴.

The application of the InVEST carbon stock module in this study serves to bridge the gap between scientific analysis and practical land management, offering valuable guidance for balancing ecological protection with economic development.

1

2

where, i denotes the land use type,C_i represents the carbon density of land use type i,

C_i−above, C_i−below, C_i−soil, and C_i−dead represent the carbon densities of aboveground vegetation, belowground vegetation, soil, and dead organic matter, respectively (units: t/hm²), C_t indicates the total carbon stock of the ecosystem (units: t/hm²), A_i is the area of land use type iii (units: hm²),

n represents the total number of land use types.

Adjustment of land use carbon density

This study selected temperature and precipitation as key factors for carbon density adjustment. The data were sourced from the China Meteorological Administration, covering the period from 2000 to 2020, with a spatial resolution of 1 km × 1 km. To ensure data quality, missing values were imputed using mean substitution, and noise was removed using denoising algorithms.

Land use types, vegetation cover, and soil types were excluded from the model, as these factors are often indirectly reflected through climatic variables. This decision aligns with the theoretical basis established in previous studies⁵⁵. Future research will explore the potential impacts of these factors on carbon density.The correction coefficients for carbon density in the Chengdu Plain region were calculated using the following equations:

3

4

5

where C_sp, C_bp, and C_bt represent the soil and biomass carbon density calculated based on annual precipitation and mean annual temperature (units: kg/m²). Map and Mat denote mean annual precipitation (mm) and mean annual temperature (°C), respectively. By substituting the temperature and precipitation data for the Chengdu Plain region (16.1 °C, 1000 mm) and the national averages into the equations, their ratios were calculated to derive correction coefficients.

The adjusted carbon density is expressed as:

6

7

8

9

where, K_bp represents the precipitation correction coefficient for biomass carbon density, K_bt denotes the temperature correction coefficient for biomass carbon density, K_b is the overall biomass carbon density correction coefficient, and K_s is the soil carbon density correction coefficient. C_c and C_z refer to the carbon densities of the Chengdu Plain region and China, respectively, both calculated based on mean annual temperature and precipitation values.

The carbon density of the Chengdu Plain region is then derived by multiplying the national carbon density value by the corresponding carbon density correction coefficient (Table 4).

Table 4.

Carbon density for Chengdu Plain region (t C/ha).

Land use type	C_above	C_below	C_soil	C_dead
Cultivated land	8.92	13.07	22.85	0.19
Forest land	12	14.87	32.81	7.413
Grassland	9.79	12.06	24.49	1.101
Water area	0	0	9.41	1
Construction land	3.4	0	8.56	0
Unused land	0	0	0	0

OPGD model

The Optimal Parameters-based Geographical Detector (OPGD) model is a refined spatial statistical approach designed to explore relationships among variables and their driving mechanisms within spatial data. Compared to the traditional Geographical Detector (GD) model, OPGD improves analytical precision by selecting optimal parameter combinations, enhancing its ability to quantify the interactive effects of natural factors and human activities^56,57.

In factor detection, OPGD refines discretisation methods and interval combinations to calculate the Q-value for continuous variables, which measures a factor’s contribution to spatial heterogeneity. By evaluating various classification methods and interval combinations, the model identifies parameter sets that maximise the Q-value, thereby increasing the spatial stratification capability and explanatory power of variables influencing spatial phenomena^21,22,57,58. Furthermore, OPGD facilitates interaction analysis between natural and anthropogenic factors across different spatial scales, effectively determining the relative importance of each factor in spatial differentiation and identifying dominant variables. The factor explanatory power (Q) is calculated as follows:

The calculation for the factor explanatory power (Q) is as follows:

10

where, Q represents the explanatory power of the factor on land carbon stock, with a range of [0, 1]; higher values indicate stronger explanatory power of the factor on land carbon stock. h = 1,…,L denotes the stratification of land carbon stock and influencing factors.N_h is the number of units in stratum h. N represents the total number of units in the region. σh² and σ² denote the variance of land carbon stock within stratum h and across the entire region, respectively. SSW is the sum of within-stratum variances. SST represents the total variance across the entire region.

The variance of the regional Y-value is calculated as follows:

11

In this formula, Y_j and

12

where, Y and are respectively the values and mean values of sample i in layer h.

In this study, land carbon stock is designated as the dependent variable (Y), with the selected independent factors (X) presented in Table 5. Factors X1, X2, X3, and X6 were obtained from the Resource and Environment Science and Data Centre of the Chinese Academy of Sciences. Factors X10, X11, X12, X13, X14, and X15 were sourced from county and city government websites, the Statistical Bulletin of National Economic and Social Development, and the Sichuan Statistical Yearbook. Spatialisation of these variables was conducted using GIS tools to assess their impacts on carbon stock changes.

Table 5.

Index of factors in study area.

Factors types	Code	Index	Unit	Resolution
Natural factors	X1	Vegetation type	type	1 km
	X2	NPP		500 m
	X3	NDVI	-	500 m
	X4	Slope	°	100 m
	X5	Aspect	°	100 m
	X6	Soil type	type	1 km
	X7	Annual temperature	°C	1 km
	X8	Annual precipitation	mm	1 km
	X9	Elevation	m	100 m
Humanistic factors	X10	Non-agricultural population	%	30 m
	X11	Total population	104person	30 m
	X12	Gross industrial output	104yuan	30 m
	X13	Gross agricultural output	104yuan	30 m
	X14	Per capita GDP	Yuan/person	30 m
	X15	Gross service output	104yuan	30 m

Results and analysis

The random forest classification results based on the GEE platform show that the Kappa coefficients for the years 2000, 2005, 2010, 2015, and 2020 were 0.84, 0.82, 0.81, 0.85, and 0.83, respectively. The overall classification accuracies were 0.87, 0.86, 0.84, 0.92, and 0.91. Both the Kappa coefficients and classification accuracies remained at a high level, and the trends in accuracy changes were consistent, indicating that the classification results were stable and largely aligned with the actual conditions.

Spatiotemporal characteristics of land use change from 2000 to 2020

Temporal variation characteristics

From 2000 to 2020, the region underwent significant land cover changes, primarily characterized by a substantial decline in arable land, a rapid expansion of construction land, and relatively stable forest coverage (Fig. 2). Arable land decreased by 1,549.36 km², representing a 6.7% reduction, while construction land expanded by 1,552.78 km², marking an 84.47% increase. Forest area experienced a slight decline of 55.69 km² (0.59%), and grassland decreased by 145.27 km² (5.94%). Meanwhile, water bodies increased by 69.63 km² (11.15%), and unused land saw a dramatic rise of 128.17 km², a growth rate of 873.98% (Fig. 2).

Fig. 2.

Temporal variation of land use in Chengdu Plain region from 2000 to 2020.

From 2000 to 2005, the most notable changes were a decrease in arable land and a corresponding increase in construction land, driven by rapid urbanization during this period. Between 2005 and 2010, both arable and grassland areas declined further, while construction land continued its upward trajectory, reflecting the increasing demand for urban and industrial development (Fig. 2). The reasons for the sharp increase in unused land are as follows: Firstly, with urbanization and the expansion of construction land, some land is reserved for future development, although it has not yet been developed, it is classified as unused land⁵⁹. Secondly, some agricultural land has been converted to unused land due to labor migration and abandonment of farming⁶⁰. Additionally, changes in land classification standards and the implementation of environmental protection policies may have led to some areas being reclassified as unused land. During the period from 2015 to 2020, the decrease in arable land persisted, but the areas of grassland and water bodies remained relatively stable. Construction land, however, maintained its rapid growth, underscoring the ongoing urban expansion in the region (Fig. 2).

Spatial patterns of land cover from 2000 to 2020

Cultivated land, represented by light yellow areas in Fig. 3, is predominantly distributed across flat terrain with fertile soils and abundant water resources, mainly around Chengdu’s urban core and surrounding districts such as Wenjiang, Pidu, and Shuangliu. Adjacent areas, including Chongzhou, Xinjin, Pengzhou, and Dujiangyan, also exhibit extensive arable land coverage. Forest areas (dark green) are primarily located along hilly and mountainous regions, such as the Longquan Mountains in the east, Longmen Mountains in the northeast, Qingcheng Mountains in the northwest, and Qionglai Mountains in the south. Scattered forest patches are also present in plain and hilly areas, often along riverbanks, in farmland shelterbelts, and within wetland parks as artificial or ecological forests.

Fig. 3.

Spatial pattern of land use in Chengdu Plain region from 2000 to 2020 (The figure was generated using ArcGIS 10.3 software, with the URL link: https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview).

Construction land (red) is concentrated in Chengdu’s urban districts, Tianfu New Area, and Shuangliu District, as well as in suburban zones and industrial clusters across county-level cities (Fig. 3). The rapid urbanization of Chengdu and the development of industrial zones have driven significant land conversion, primarily from arable land and forest to urban and industrial areas.

Grassland (light green) is distributed along the Longquan Mountains, Dujiangyan, and the banks of the Min and Tuo Rivers, as well as around the Qionglai Mountains. Despite a slight decline due to urbanization, grassland has been preserved and restored within ecological protection zones and wetland areas, reflecting effective conservation efforts (Fig. 3).

Water bodies (blue) remain relatively stable, with significant concentrations along the Min and Tuo River basins, the Dujiangyan irrigation network, and sites such as Longquan Lake and Baihetan Wetland (Fig. 3). These water resources play a vital role in supporting agricultural irrigation, urban water supply, climate regulation, wetland ecosystem maintenance, and biodiversity conservation.

Unused land (purple), although limited in area, has shown significant growth. It is mainly distributed along the Longquan Mountains, low-lying areas of the Min and Tuo River valleys, and the southwestern edges of hilly zones (Fig. 3). The substantial increase in unused land can be attributed to terrain constraints, soil degradation, and flood risks, making these areas less suitable for development. Despite these limitations, unused lands contribute to ecological functions such as water conservation, flood control, and biodiversity protection.

Land use transition from 2000 to 2020

Using GIS spatial analysis, a land use transition matrix and spatial transition maps were generated for the Chengdu Plain region from 2000 to 2020 (Fig. 4). These results highlight substantial land use dynamics over the two decades, primarily driven by urban expansion and ecological restoration initiatives.

Fig. 4.

Land use transition matrix (a, b, c) and transition maps (a, b, c) from 2000 to 2020.

Overall transition characteristics

The land use transition matrix indicates significant transformations between cultivated land, construction land, and forested land (Fig. 4). From 2000 to 2020, cultivated land underwent the largest total transition (1,304.21 km²), with 45.3% converted to construction land and 33.8% to forested areas (Fig. 4c). This shift underscores the dual drivers of urbanization and ecological restoration. Forested land exhibited a total transition of 1,078.88 km², primarily converted to cultivated land (42.3%) and grassland (28.9%), reflecting adjustments in agricultural and land management practices. Grassland recorded a relatively smaller transition (376.6 km²), largely towards cultivated land (49.2%) and forested areas (32.5%), indicating an interplay between reforestation and agricultural expansion (Fig. 4c).

According to Fig. 4c, Water bodies experienced limited transitions, totalling 169.79 km², primarily converted to cultivated and construction land, which could be attributed to water resource reallocation or urban sprawl. Construction land exhibited an interesting trend, with a notable reversion to cultivated land (1,766.55 km²) due to policy-driven land reclamation projects. Unused land, though minimal in extent, saw a significant proportion transition to forested and grassland areas (69.2%) (Fig. 4c), suggesting further land resource optimization and ecological development.

Temporal trends

Between 2000 and 2010, cultivated land saw a transition of 1,269.82 km², mainly converted to construction land (56.7%) and forested areas (31.2%) (Fig. 4a). This period coincided with rapid urbanization and the implementation of ecological restoration programs in the Chengdu Plain region. Concurrently, forested land transitioned to cultivated land (46.5%) and grassland (24.8%), while grassland experienced shifts towards cultivated (53.9%) and forested areas (29.1%). Notably, construction land exhibited limited reversion to cultivated and forested land (67.05 km²) (Fig. 4a).

From 2010 to 2020, cultivated land transitions intensified, with 1,518.49 km² primarily converted to construction (48.9%) and forested land (35.4%) (Fig. 4b). Forested land showed a total transition of 629.56 km², largely reverting to cultivated land (39.8%) and shifting to grassland (26.9%) (Fig. 4b). Grassland transitions remained consistent, with 311.36 km² primarily converted to cultivated and forested land. Water bodies showed minor but noteworthy conversions (79.28 km²), largely to cultivated and construction land. Construction land, while continuing its expansion, also reverted significantly to cultivated land (288.29 km²), reflecting ongoing land reclamation initiatives (Fig. 4b).

Spatial patterns of transition

The transition maps reveal spatially distinct patterns of land use change. Cultivated land was predominantly converted to construction land in urban expansion zones, such as Chengdu’s central districts and suburban regions (Fig. 4a,b,c). According to Fig. 4a,b,c, forested land transitions were more dispersed, often located along the Longquan Mountains and other ecological restoration zones. Grassland transitions were concentrated near river valleys and hilly terrain, where reforestation and agricultural development are prevalent.

Construction land expansions were concentrated in urban cores and industrial clusters, such as the Tianfu New Area. The limited transitions of water bodies were primarily observed in river basins undergoing urbanization (Fig. 4a,b,c). Unused land conversions were localized along the peripheries of ecological zones and marginal lands.

Implications of land use change

The land use transitions observed in the Chengdu Plain region reflect the dual pressures of urbanization and ecological sustainability. The conversion of cultivated land to construction land aligns with economic and population growth, while the transitions to forested and grassland areas suggest increasing policy attention to ecological restoration. However, these changes also raise concerns regarding the loss of agricultural land and potential impacts on regional food security and carbon sequestration.

Future land use multi-scenario simulation

Using the PLUS model, this study conducted multi-scenario simulations of land use types in the Chengdu Plain region for 2030 and 2060, incorporating factors such as elevation, slope, population, and proximity to roads. To validate simulation accuracy, the 2010 land use status of the Chengdu Plain region was used as the initial state to simulate the 2020 land use pattern. The simulated 2020 land use was then compared to the actual 2020 land use, with accuracy evaluated using the Kappa coefficient and overall accuracy. A Kappa coefficient of 0.75 is generally regarded as indicative of a reliable simulation. In this study, the PLUS model achieved a Kappa coefficient of 0.86 and an overall accuracy of 0.91, confirming the robustness of the simulation results.

Multi-scenario simulation for 2030

The multi-scenario simulation results indicate that, by 2030, land use in the Chengdu Plain region will remain dominated by cultivated land and forested areas, accounting for more than 55% and 25%, respectively, compared to 2020.

• (1) NDS

Under the NDS, urban expansion significantly influences land use structure. Cultivated land decreases by 595.91 km² (1.59%), likely driven by urban development and infrastructure needs (Fig. 5a). Forested land decreases modestly by 54.77 km² (0.15%), while grassland also declines slightly by 5.77 km² (0.02%) (Fig. 5a). Water bodies expand by 33.67 km² (0.09%), potentially reflecting improved wetland management. Construction land exhibits the largest change, increasing by 626.63 km² (1.67%), highlighting rapid urbanisation (Fig. 5a). Unused land decreases minimally by 3.84 km² (0.01%), suggesting limited development of marginal lands (Fig. 5a).

• (2) EPS

Fig. 5.

Scenario-based land use simulations for 2030 using the PLUS model: NDS (a), EPS (b), CLDS (c), and EDS (d). (The figure was generated using ArcGIS 10.3 software, with the URL link: https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview).

Under the EPS, ecological restoration policies drive notable land use changes. Cultivated land decreases by 962.74 km² (2.57%), primarily converted to forested land (299.17 km², 0.80%) and grassland (63.38 km², 0.17%) (Fig. 5b). These changes reflect enhanced afforestation and grassland protection initiatives. Water bodies expand slightly by 15.14 km² (0.04%), linked to wetland conservation. Construction land increases by 598.57 km² (1.60%), but growth appears more regulated compared to the NDS. Unused land decreases by 13.66 km² (0.04%), indicating marginal land repurposing for ecological or construction uses.

• (3) CLDS

The CLDS prioritises cultivated land protection, resulting in an increase of 349.38 km² (0.93%) (Fig. 5c). However, this leads to a reduction in forested land by 359.74 km² (0.96%), suggesting trade-offs between agriculture and forest conservation. Grassland decreases by 64.92 km² (0.17%), while water bodies expand by 29.38 km² (0.08%), reflecting targeted ecological restoration. Construction land grows modestly by 50.72 km² (0.14%), indicating controlled urban development. Unused land decreases slightly by 4.96 km² (0.01%).

• (4) EDS

Under the EDS, rapid urbanisation dominates land use changes. Cultivated land decreases by 782.25 km² (2.09%), while construction land increases significantly by 827.97 km² (2.21%) (Fig. 5d). Forested land decreases by 67.14 km² (0.18%), and grassland declines minimally by 7.88 km² (0.02%). Water bodies expand by 32.91 km² (0.09%), suggesting ongoing wetland conservation. Unused land decreases slightly by 3.74 km² (0.01%), indicating limited land development.

Multi-scenario simulation for 2060

Under the NDS for 2060, cultivated land decreases by 1,205.27 km², a reduction of 3.22% (Fig. 6a). This marks the largest decline among all land types, primarily driven by urban expansion or other land use changes. Forested land increases by 179.87 km² (0.48%), likely due to natural ecological restoration or conservation measures. Grassland experiences a minor decrease of 17.17 km² (0.05%), indicating limited conversion to other uses. Water bodies show a significant decrease of 284.59 km² (0.76%), potentially due to natural resource depletion or wetland conversion. Construction land expands by 1,335.64km² (3.57%), reflecting substantial urbanisation or infrastructure development. This expansion is likely a primary driver of reductions in other land types, including cropland and water bodies. Unused land decreases modestly by 8.54 km² (0.02%), suggesting minimal development activity in these areas.

Fig. 6.

Scenario-based land use simulations for 2060 using the PLUS model: NDS (a), EPS (b), CLDS (c), and EDS (d). (The figure was generated using ArcGIS 10.3 software, with the URL link: https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview.)

Under the EPS, despite the implementation of ecological protection policies, cultivated land, forested land, grassland, and water bodies all experience reductions (Fig. 6b). Notably, water bodies decrease by 236.48 km² (0.63%), possibly due to the artificial transformation of wetlands and other water resources (Fig. 6b). Cultivated land decreases by 577.97 km² (1.54%), suggesting that urban expansion continues to exert pressure on cropland despite policy efforts. Forested land decreases by 174.90 km² (0.47%), indicating that ecological protection measures may not fully counterbalance development pressures. Grassland decreases by 81.29 km² (0.22%), likely reflecting minor conversions to construction land. Construction land increases significantly by 1,069.86 km² (2.86%), highlighting rapid urbanisation. Unused land exhibits a marginal increase of 0.86 km², suggesting minimal impact from development activities.

Under the CLDS, construction land continues to expand significantly, resulting in a cultivated land decrease of 1,631.70 km² (4.36%). This reflects substantial challenges in balancing urbanisation with land protection (Fig. 6c). In contrast, forested land increases by 483.80 km² (1.29%), indicating some success in ecological restoration and forest protection initiatives. Grassland increases modestly by 44.86 km² (0.12%), suggesting limited restoration efforts. However, water bodies decrease significantly by 263.91 km² (0.70%), potentially due to poor water resource management. Construction land expands by 1,374.97 km² (3.67%), indicating continued urban growth. Unused land decreases marginally by 7.94 km² (0.02%).

Under the EDS, cultivated land decreases by 1,190.58 km² (3.18%), reflecting urban expansion’s persistent impact on cropland (Fig. 6d). Forested land increases moderately by 182.64 km² (0.49%), likely due to greening policies. Grassland decreases slightly by 18.31 km² (0.05%), indicating limited impacts of urban development. Water bodies experience a significant reduction of 300.64 km² (0.80%), underscoring the need for improved water resource management. Construction land expands substantially by 1,336.84 km² (3.57%), driven by rapid urbanisation and infrastructure development. Unused land decreases marginally by 9.87 km² (0.03%), indicating minimal changes.

Land use transition

• (1) Under the NDS (2030–2060): Approximately 1,226.23 km² of cropland is projected to transition between 2030 and 2060 (Figure upper 7a), with the majority shifting to construction land (599.22 km²), indicating the impact of urbanisation. A substantial portion (435.02 km²) is converted to forested areas, likely due to ecological restoration efforts. A smaller portion shifts to grassland (108.51 km²) and water bodies (82.64 km²). Forest land transitions predominantly into cropland (859.12 km²), but also to grassland (108.23 km²) and construction land (32.32 km²). Grassland remains relatively stable, with 111.57 km² converted to cropland and 84.65 km² to forest. Water bodies see a slight reduction, with the majority transitioning to forest land (91.96 km²) and cropland (51.85 km²). Construction land undergoes substantial transitions, with a total area of 1,971.75 km² reverting to other uses, predominantly cropland (1,407.27 km²), forest (224.49 km²), and water bodies (335.65 km²).

• (2) Under the EPS (2030–2060): A total of 1,366.92 km² of cropland transitions between 2030 and 2060 (Figure upper 7b), with significant conversion to construction land (1,167.88 km²), highlighting continued urbanisation. However, 632.15 km² of cropland transitions to forested areas, reflecting conservation efforts. A small proportion shifts to grassland (85.43 km²) and water bodies (57.87 km²). Forest land transitions mainly to cropland (488.35 km²), but also to grassland (89.25 km²), construction land (262.67 km²), and water bodies (98.5 km²). Construction land sees considerable reversion to cropland (640.32 km²) and forest (22.2 km²), signaling partial ecological restoration. Unused land transitions minimally, with the majority going to forest (9.9 km²) and grassland (2.88 km²).

• (3) Under the CLDS (2030–2060): Land use transitions are less pronounced, with 1,232.99 km² of cropland converted (Figure upper 7c), predominantly to construction land (850.85 km²). A smaller portion transitions to forest (273.34 km²) and grassland (65.17 km²). Forested areas decrease by 404.49 km², with the primary conversion to cropland (359.74 km²). Grassland transitions slightly to cropland (49.11 km²) and forest (19.74 km²). Construction land shifts primarily to cropland (648.23 km²), forest (215.92 km²), and water bodies (64.62 km²).

• (4) Under the EDS (2030–2060): The most significant transition occurs in construction land, with 2,057.64 km² reallocated (Figure upper 7d), primarily to cropland (1,874.55 km²) and forest (166.83 km²). Cropland transitions primarily to construction land (931.98 km²) and forest (788.73 km²), while forested areas transition to cropland (380.61 km²) and construction land (203.57 km²).

• (5) Spatial transfer patterns (2030–2060): The subplots in Figure lower 7 (a. NDS, b. EPS, c. CLDS, d. EDS) illustrate the spatial distribution characteristics and clustering trends of land use transitions under different simulation scenarios. For instance, the expansion of construction land is predominantly concentrated around urban areas, reflecting the significant impact of rapid urbanisation on land use changes. In contrast, transitions involving forested land and grassland are relatively dispersed, and in certain regions, these areas may be replaced by cropland or construction land. Peripheral regions are typically subject to minimal human interference (e.g., minor changes in unused land), whereas central areas, particularly those surrounding cities, emerge as hotspots of intense land use change.

Fig. 7.

Upper: land use transition matrix (a, b, c, d) and lower: land use transition maps (a, b, c, d) from 2030 to 2060.

Temporal and spatial changes in land carbon stock

• (1) Temporal changes in total carbon stock

Between 2000 and 2020, the total carbon stock in the Chengdu Plain region decreased from 181.56 Mt to 175.45 Mt, representing a reduction of 6.11 Mt or 3.36%. This decline primarily occurred in croplands, where carbon stock dropped from 104.12 Mt in 2000 to 97.14 Mt in 2020. Notable decreases occurred in 2010, 2015, and 2020, which coincide with periods of accelerated urbanisation and significant land use changes. The carbon stock in forest land showed minor fluctuations, decreasing slightly from 62.99 Mt in 2000 to 62.61 Mt in 2020, suggesting stable ecological conditions in forested areas. Grassland carbon stock followed a fluctuating trend, declining from 11.61 Mt in 2000 to 10.92 Mt in 2020. A sharp increase was observed in 2010, potentially linked to climatic variations or ecological restoration policies during that period.

Conversely, carbon stock in construction land rose substantially from 2.20 Mt in 2000 to 4.06 Mt in 2020, reflecting urbanisation and infrastructure expansion. Water bodies and unused land exhibited steady increases in carbon stock, from 0.65 Mt and 2.20 Mt in 2000 to 0.72 Mt and 4.06 Mt in 2020, respectively.

• (2) Spatial distribution and variability

The spatial distribution of carbon stock across different land types in the Chengdu Plain region is highly uneven (Fig. 8). Cropland, concentrated in the core agricultural areas, holds the highest carbon stock due to extensive cultivation and fertile soils. However, urban expansion around Chengdu’s central area has replaced significant cropland with construction land, leading to a notable decline in carbon stock in these regions. Forested areas, primarily located in the hilly and mountainous regions to the east, west, and south, serve as the primary carbon sink, maintaining relatively stable carbon stock levels. Grassland areas, sparsely distributed at the interface of agricultural and pastoral zones, show modest contributions to the overall carbon stock, with trends influenced by local land use changes.

Fig. 8.

Spatial pattern of land carbon stock in Chengdu Plain region from 2000 to 2020 (The figure was generated using ArcGIS 10.3 software, with the URL link: https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview.)

Construction land, concentrated in Chengdu’s urban core, exhibits a low per-unit carbon stock due to impervious surface coverage, despite its overall increase in total carbon stock. The carbon stock in water bodies, although minor, has gradually increased, particularly along major river systems such as the Minjiang, Dujiangyan, and Jinjiang rivers.

• (3) Drivers of change

Rapid urbanisation is a key driver of carbon stock decline in cropland and forested areas, particularly around urban centres. Policies promoting ecological restoration and afforestation in hilly regions have mitigated carbon losses to some extent, stabilising carbon stock in forested areas. However, the continued expansion of construction land poses a significant threat to the region’s overall carbon sequestration capacity.

• (4) Recommendations for future research

Future studies should focus on quantitatively linking urbanisation rates, land use transitions, and policy impacts to carbon stock changes. Incorporating high-resolution spatial data and advanced modelling techniques can enhance the understanding of carbon dynamics in rapidly urbanising regions like the Chengdu Plain region .

Future changes in carbon stock under different scenarios

Carbon stock change by 2030

By 2030, carbon stock is projected to increase across all four land-use scenarios, but at varying rates: NDS by 2.58%, EPS by 3.05%, farmland conservation scenario (CLDS) by 3.26%, and economic development scenario (EDS) by 2.18% (Fig. 9). The EPS, with its focus on expanding forest and grassland areas, shows the largest relative increase, highlighting the importance of natural vegetation in carbon sequestration. In contrast, urbanisation under the EDS results in the smallest increase, reflecting the carbon sink losses associated with land conversion to urban use.

Fig. 9.

Land carbon stock under different scenarios in 2030 and 2060.

Under the NDS, carbon stock decreases in cropland (− 2.68 Mt), forest (− 0.45 Mt), and grassland (− 0.03 Mt), while increases in water bodies (+ 6.85 Mt) and construction land (+ 0.75 Mt) offset these losses (Fig. 9). Similarly, in the EPS, reductions in cropland (− 4.33 Mt) are compensated by increases in forest (+ 2.01 Mt), grassland (+ 0.30 Mt), water bodies (+ 6.66 Mt), and construction land (+ 0.72 Mt).

In the CLDS, cropland increases carbon stock by + 1.57 Mt, but decreases in forest (− 2.41 Mt) and grassland (− 0.31 Mt) partially counteract these gains(Fig. 9) . Meanwhile, water bodies (+ 6.81 Mt) and construction land (+ 0.06 Mt) make additional contributions.

Under the EDS, cropland, forest, and grassland experience losses (− 3.52 Mt, − 0.45 Mt, and − 0.04 Mt, respectively), while water bodies (+ 6.85 Mt) and construction land (+ 0.99 Mt) continue to rise (Fig. 9) .

In summary, the CLDS demonstrates the highest potential for carbon stock increase due to the prioritisation of cropland conservation. However, its overall contribution to carbon sequestration is limited compared to the EPS, which leverages forest and grassland expansion to achieve superior carbon storage. Urbanisation under the EDS results in the smallest net increase, emphasising the trade-offs between economic development and ecological sustainability.

Carbon stock change by 2060

Figure 9 illustrates the projected carbon stock changes across different scenarios by 2060. Overall, the total carbon stock exhibits varying degrees of decline, driven primarily by reductions in water bodies and cropland.

Under the NDS, the total carbon stock decreases by 9.8202 Mt (5.46%). This decline is primarily attributed to substantial losses in water bodies (7.1156 Mt) and cropland (5.4273 Mt) (Fig. 9) . However, forest expansion contributes an additional 1.2068 Mt, and construction land adds 1.5974 Mt, albeit with a limited impact.

EPS shows a relatively smaller decline in total carbon stock, reducing by 8.9496 Mt (4.95%) (Fig. 9) . This is primarily due to a moderate reduction in cropland carbon stock (4.2544 Mt), which is less pronounced compared to the NDS. Forest and construction land contribute 1.2013 Mt and 1.246 Mt, respectively, mitigating some of the losses. However, water bodies still experience a marked decline of 7.0848 Mt.

Under the CLDS, the total carbon stock decreases by 8.1017 Mt (4.47%), representing the smallest decline among all scenarios (Fig. 9). The reduction in cropland carbon stock is limited to 3.0909 Mt, reflecting the scenario’s emphasis on preserving agricultural land. Forest carbon stock increases by 1.1999 Mt, while construction land adds 0.9557 Mt, the smallest increase among the scenarios. Water bodies and grassland continue to experience losses, with reductions of 7.0985 Mt and 0.0678 Mt, respectively.

Conversely, the EDS experiences the greatest total carbon stock loss, decreasing by 10.4483 Mt (5.83%)(Fig. 9) . This significant decline is driven by reductions in cropland (6.2003 Mt) and grassland (0.0969 Mt). Forest and construction land add 1.1424 Mt and 1.8397 Mt, respectively. Notably, urban expansion under this scenario leads to the largest increase in construction land carbon stock, partially offsetting other losses. Water bodies again show the largest reduction, losing 7.1331 Mt.

In summary, water bodies exhibit the most significant and consistent decline in carbon stock across all scenarios, suggesting that future management strategies should prioritize mitigating these losses. The differences in cropland carbon stock reductions highlight the impact of land-use policies, particularly in scenarios emphasizing agricultural preservation. Forest expansion remains a key contributor to mitigating carbon stock losses, while urbanization under the EDS leads to the most pronounced changes in construction land. Future analyses should explore the underlying mechanisms driving water body and grassland carbon stock reductions, as well as potential mitigation strategies.

Spatial variation in simulated carbon stock across multiple scenarios

The spatial distribution of carbon stock in the Chengdu Plain region reveals a pronounced urban–rural divide, particularly as urbanization intensifies between 2030 and 2060 (Fig. 10). Urban expansion in the Chengdu city center and surrounding areas leads to a marked conversion of cropland and forest land into construction land, causing a significant decline in carbon stock. Specifically, carbon stock in the central urban zones declines, while peripheral cropland areas face significant reductions due to expanding urban encroachment. Core agricultural regions, however, maintain relatively high carbon stock, demonstrating resilience against urban expansion.

Fig. 10.

Spatial distribution pattern of carbon stock under different scenarios in 2030 and 2060 (The figure was generated using ArcGIS 10.3 software, with the URL link: https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview.).

In contrast, carbon stock in remote hills, mountainous areas, and ecological protection zones remains relatively stable or shows potential for increase under supportive policies. These areas, located primarily in the eastern, western, and southern hills and mountain zones, are characterized by high forest cover, which serves as a critical carbon sink. Quantitatively, forest-dominated regions contribute over 24.99% of the total carbon stock in the Chengdu Plain region, emphasizing their ecological significance. Although limited in area, grassland carbon stock is concentrated in transitional agro-pastoral zones, where it makes a notable contribution to the overall carbon stock.

Construction land exhibits consistently low carbon stock, particularly in urbanized areas where the expansion of buildings and infrastructure further constrains carbon sequestration capacity. Water bodies, concentrated along major rivers and lakes such as the Min River, Dujiangyan, and Jinjiang, account for 1.27% of the total carbon stock and show relative stability during the study period.

Overall, the spatial distribution of carbon stock highlights a distinct urban–rural divide. Urban centers experience significant declines, primarily driven by land-use changes and infrastructure development, while surrounding ecological zones, protected by policy interventions, are poised to play an increasingly vital role as carbon sinks. To enhance carbon sequestration capacity in the Chengdu Plain region, further integration of urban planning with ecological conservation policies is recommended.

Driving factors for carbon stock change

The factor detector reveals that both human and natural factors contribute to changes in land-use carbon stock, with human factors playing a dominant role. By calculating Q values (Fig. 11a), the influence of individual factors was quantified. The ranking of their contributions is as follows: non-agricultural population (X10) > total agricultural output (X13) > total service sector output (X15) > per capita GDP (X14) > total population (X11) > total industrial output (X12) > vegetation type (X1) > slope (X4) > soil type (X6) > temperature (X7) > NDVI (X3) > precipitation (X8) > elevation (X9) > NPP (X2) > aspect (X5).

Fig. 11.

Exploration of explanatory variables of land carbon stock change in Chengdu Plain region based on OPGD: single factor contribution (a) and interaction detection (b).

The Q value for non-agricultural population (X10) is 0.1529, indicating that urban expansion significantly impacts carbon stock, while the Q values for total agricultural output (X13, 0.1528) and total service sector output (X15, 0.1527) suggest that economic activities are closely linked to land-use changes. Collectively, these results underscore the substantial influence of urbanization and economic development, as these factors demonstrate explanatory powers of approximately 15% each. In comparison, natural factors exhibit weaker effects, with Q values below 4%. For example, the Q value for NDVI (X2) is only 0.0076, while mean annual temperature (X7) reaches 0.0206. This indicates that natural factors such as vegetation productivity and climate conditions have limited contributions to carbon stock changes in the Chengdu Plain region.

Human factors also show significant correlations with one another, with coefficients ranging from 0.153 to 0.171 (Fig. 11b). This strong interrelationship suggests a bilinear enhancement effect, where combinations of factors such as population growth (X10, X11) and economic output (X13, X15) amplify changes in carbon stock. Conversely, natural factors demonstrate weaker correlations, with most coefficients below 0.06, indicating relatively independent influences. For example, interactions between NDVI (X2) and elevation (X9), or slope (X4) and soil type (X6), primarily contribute through nonlinear enhancement effects, reflecting complex dynamics in their combined influence on carbon sequestration.

In summary, human activities, particularly urban expansion and industrial development, dominate changes in carbon stock, whereas natural factors have comparatively minor direct effects. However, their interactions with human factors can amplify their influence under specific scenarios. These findings highlight the importance of managing urbanization and economic activities to mitigate carbon loss while leveraging natural factors through ecological conservation and restoration strategies. The interplay of multiple factors underscores the need for an integrated approach to land-use planning and carbon management in the Chengdu Plain region .

Discussion

Climate-based adjustment of carbon density

This study identified precipitation and temperature as key factors in carbon density adjustment due to their significant influence on biomass and soil carbon dynamics. Consistent with prior research⁶¹, precipitation was confirmed as a dominant driver of soil carbon density, shaping both biomass and soil carbon pools, as also observed by Zhang et al.⁶². In contrast, this study highlights a comparatively minor role of temperature, aligning with findings by⁶³, suggesting that models solely dependent on temperature may oversimplify soil carbon dynamics.

The adjustment formulas, integrating regional climatic characteristics, provide a novel framework to assess the combined impacts of temperature and precipitation. However, their broader applicability is constrained by limited data on hydrological conditions and vegetation composition, factors highlighted as crucial in earlier studies⁶⁴. Incorporating detailed ecological and hydrological data in future adjustments could enhance model robustness and accuracy.

Validation efforts confirmed the accuracy of the adjusted carbon density values through comparison with observed data. Additionally, the correction coefficient demonstrated stability and reliability under varying climatic and regional conditions. Future studies should integrate additional factors, such as soil types and vegetation cover, to refine the analysis of carbon density. Addressing spatial variability will further improve the applicability of correction coefficients and advance regional carbon dynamics modeling.

Advantages of the OPQD Model

The OPQD model significantly advances the analysis of land-use carbon stock drivers by optimising parameter selection and addressing nonlinear interactions, thereby exceeding the explanatory power of traditional models like STWR. This aligns with findings from Li et al.⁶⁵, who highlighted OPQD’s strength in capturing multifactorial interactions. However, while OPQD excels in addressing spatial discontinuities and handling nonlinear data, its application is limited by its higher computational demands, which may restrict its scalability to larger datasets.

The integration of OPQD and STWR could provide a more holistic approach, leveraging OPQD’s interaction-detection capabilities and STWR’s strength in assessing dynamic spatiotemporal trends. Such integration could address existing gaps in the analysis of carbon stock changes, offering actionable insights for regional land-use planning.

Analysis of the causes of land carbon stock decrease

Urbanisation, economic growth, and population expansion emerged as primary drivers of carbon stock reduction in the Chengdu Plain region . Our findings are consistent with Wang et al.⁶⁶, who observed a similar decline in arable land carbon stock due to urbanisation. However, our analysis extends these findings by quantifying the specific contributions of economic and population factors, as indicated by the Q values.

Interestingly, natural factors such as NPP and slope exhibited minimal influence, corroborating Zhang et al.⁶⁷. Yet, the interplay between natural stability and anthropogenic disruption warrants deeper exploration. For example, the relative stability of natural factors may buffer the effects of urbanization, or there may be thresholds beyond which this stability is compromised.

Rational land use planning and ecological protection measures

To address the carbon stock challenges posed by urbanisation, this study underscores the importance of integrating land-use planning with ecological protection measures. Aligning with Zhao et al.⁶⁸, we advocate for prioritising high-carbon stock areas through permanent farmland protection and ecological redlines. Additionally, promoting ecological agriculture and green infrastructure could offset carbon stock losses, as supported by findings from Li et al.⁶⁹.

In rapidly urbanizing regions like Chengdu, implementing ecological protection measures faces challenges such as competing land-use demands, financial constraints, and the difficulty of balancing urban development with ecological conservation⁷⁰. The rapid expansion of infrastructure often conflicts with the preservation of natural habitats, making it challenging to incorporate ecological protection into urban planning⁷¹. To ensure the practical feasibility of these measures, future research should focus on evaluating their cost-effectiveness and scalability, emphasizing land-use optimization, economic incentive mechanisms, and the potential of green infrastructure. Both short-term and long-term financial assessments are essential to maintaining high-carbon stock areas while promoting urban growth⁷².

Additionally, it is crucial to consider the socio-economic context of these measures, enhancing local community participation and support⁷³. Integrating participatory planning, community-led conservation initiatives, and incentive mechanisms can significantly improve the feasibility and acceptance of these measures.

Cross-regional case studies and data-driven approaches should be employed to evaluate their effectiveness, particularly in regions with similar ecological conditions and urbanization pressures⁷⁴. The use of remote sensing data and GIS tools will help identify priority conservation areas and optimize resource allocation.

Moreover, understanding the role of governance structures, policy interventions, and regulatory frameworks is key. Drawing from successful regions and combining scientific analysis with local knowledge will provide valuable insights for adapting and implementing effective ecological protection measures in other rapidly urbanizing areas.

Conclusion

This study employed the GEE platform and the Random Forest classification algorithm to generate high-precision land-use datasets for 2000–2020. The PLUS model was used to simulate land-use changes under four scenarios (NDS, EPS, CLDS, EDS) for 2030 and 2060, while the InVEST and the OPQD models quantified carbon storage dynamics and the influence of driving factors. The findings offer significant insights into balancing ecological conservation and economic development during rapid urbanisation.

• (1) Land-use transformations and future scenarios

Between 2000 and 2020, the Chengdu Plain region witnessed significant land-use changes, with a 4.14% decrease in farmland and a 4.15% increase in built-up land due to urbanisation. Ecological land (forests and grasslands) remained largely stable, highlighting the effectiveness of conservation measures. Future simulations suggest a potential decline in farmland (2.80%–7.44%) and a substantial increase in built-up land (26.89%–39.95%) by 2060, particularly under NDS and EDS, which may exert significant ecological pressure. In contrast, EPS and CLDS demonstrate improved ecological protection, promoting forest and grassland recovery.

• (2) Carbon storage dynamics

Land-use changes led to a notable reduction in regional carbon stocks, with cropland-to-built-up land conversions contributing to a decline of approximately 6.11 Mt between 2000 and 2020. Scenario simulations predict a further decline in carbon stocks (5.1%–5.5%) by 2060, with EPS achieving the highest carbon storage levels due to effective cropland protection policies. Conversely, NDS and EDS highlight intensified ecological stress caused by urban expansion, underscoring the importance of land-use management for maintaining carbon stability.

• (3) Driving factors and interactions

Socio-economic factors, particularly non-agricultural population, GDP components (agriculture, services, and per capita GDP), and total population, emerged as dominant drivers of carbon stock changes, contributing over 15% each. In contrast, natural factors (e.g., vegetation type, slope, and climate variables) showed limited influence, with contributions below 3%. These results highlight the critical role of human activity in shaping land-use carbon stock dynamics and call for targeted socio-economic policies to mitigate ecological impacts.

• (4) Policy implications and future research

This study underscores the necessity of integrating ecological conservation with economic development to achieve sustainable urbanisation. By offering a scientific basis for land-use optimisation and ecological protection policies, the findings provide valuable references for addressing climate change. Future research could focus on refining the parameterisation of simulation models, incorporating more granular socio-economic and ecological data, and expanding the spatial and temporal scope to assess regional differences and long-term trends. Additionally, exploring the synergistic effects of anthropogenic and natural factors at finer scales will further enhance our understanding of land-use carbon dynamics.

Author contributions

Jie Tang:Data collection, processing, computation, initial writing. Wenfu Peng:Methods, modeling, and writing revision.

Funding

National Ministry of Education Humanities and Social Sciences Research Planning Fund Project,China,17YJA850007.

Data availability

We confirm that all datasets are publicly available for research purposes and their accuracy has been verified. The relevant datasets can be accessed through the following links: 1.Remote sensing imagery and land-use classification were processed using the random forest method on the GEE platform (https://developers.google.com/earth-engine/datasets). 2.Administrative boundary data for China and the Chengdu Plain (1:4 million scale), natural factor data (including Digital Elevation Model (DEM), annual precipitation, and mean annual temperature), and human activity data (GDP and population density) are provided by the Resource and Environmental Science Data Center, Chinese Academy of Sciences (RESDC) (http://www.resdc.cn). 3.Road network data were extracted from OpenStreetMap using the Python OSMnx package (https://osmnx.readthedocs.io/en/stable/). 4.Constraint layer data, including ecological protection zones and basic farmland (ecological redlines), were digitised from official maps published by the Sichuan Provincial Government (http://www.sc.gov.cn/). 5.Carbon density data were initially sourced from national datasets (http://www.resdc.cn/).

Declaration

Competing interests

The authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-95756-7.