Integrating Grain–Carbon Synergy and Ecological Risk Assessment for Sustainable Land Use in Mountainous High-Risk Areas

Abstract

Amid climate change and land-use transformation, the scientific identification of high-quality arable land reserves is critical for safeguarding both cropland quantity and quality. Conventional approaches, largely based on spatial autocorrelation and heterogeneity theories, inadequately capture the multi-scale integration of ecological functions and carbon cycling, particularly in ecologically high-risk areas where systematic identification and mechanism analysis are lacking. To address these challenges, this study introduces a geographically similar “grain-carbon” synergistic framework, paired with a “bidirectional optimization” strategy (negative elimination + positive selection), to overcome the shortcomings of traditional methods and mitigate grain–carbon trade-offs in high-risk areas. Using land-use data from Yunnan’s mountainous areas (2000–2020), integrated with InVEST-PLUS model outputs, multi-source remote sensing, and carbon pool datasets, we developed a dynamic land-use–carbon storage simulation framework under four policy scenarios: natural development, urban expansion, arable land protection, and ecological conservation. High-quality arable lands were identified through a geographic similarity analysis with the Geo detector, incorporating ecological vulnerability and landscape risk indices to delineate priority high-risk zones. Carbon storage degradation trends and land-use pressures were further considered to identify optimal areas for cropland-to-forest conversion, facilitating the implementation of the bidirectional optimization strategy. Multi-scenario simulations revealed an increase of 454.33 km² in high-quality arable land, with the optimized scenario achieving a maximum carbon storage gain of 23.54 × 10⁶ t, reversing carbon loss trends and enhancing both farmland protection and carbon sequestration. These findings validate the framework’s effectiveness, overcoming limitations of traditional methods and providing a robust strategy for coordinated optimization of carbon storage and arable land conservation in ecologically high-risk regions, with implications for regional carbon neutrality and food security.

1. Introduction

Climate change and land use/land cover change (LUCC), as paramount environmental challenges of this century, are profoundly reshaping ecosystem structure and function, thereby diminishing terrestrial carbon stocks and ecosystem services [1,2]. Terrestrial ecosystems annually sequester approximately 30% of anthropogenic CO₂ emissions, serving as a critical component of the carbon cycle and underpinning the carbon neutrality targets outlined in the Paris Agreement [3,4]. Globally, arable land is increasingly compressed by urbanization and ecological degradation; over the past three decades, per capita arable land has declined by more than 25%, compounded by the rising frequency of extreme climatic events which exacerbate productivity fluctuations [5,6]. Arable land, vital for food security, urban–rural integration, resource allocation, and ecological stability, faces increasing conservation pressures amid the ongoing expansion of urban and industrial land use [7,8].

This tension is particularly acute in ecologically high-risk mountainous regions. Taking Yunnan as a representative case, steep slope farmland erosion and forest degradation are intertwined, forming a negative feedback loop of “carbon sink attenuation–food yield reduction,” significantly weakening regional carbon storage and arable land productivity [9,10,11]. Under the strict constraints of the arable land “red line” policy in the ‘National Territorial Spatial Planning Outline (2021–2035)’, these regions face a dilemma: ensuring food security requires maintaining or even expanding arable land, which may exacerbate ecological degradation and carbon loss; conversely, enhancing carbon stocks necessitates afforestation through farmland retirement, potentially compressing food production space. Therefore, breaking free from the reliance on traditional spatial autocorrelation and heterogeneity frameworks to achieve synergistic optimization of arable land protection and carbon stock enhancement in ecologically high-risk areas has become a critical scientific challenge for sustainable land management [12].

Existing studies on high-quality arable land reserve identification mainly rely on three types of methods: (1) Analytic Hierarchy Process (AHP), which constructs an indicator system combined with expert-assigned weights to rank suitability, yet is highly subjective and struggles to handle multi-source heterogeneous and dynamic data [13,14]; (2) model-driven comprehensive evaluation approaches, integrating land suitability models and geographically weighted regression (GWR) for multidimensional assessments, but these are parameter-sensitive and inadequately account for ecosystem services, requiring optimization under the dual-carbon context [15,16]; (3) intelligent identification techniques combining machine learning with optimization algorithms, which provide dynamic identification capabilities but offer limited analysis of arable land ecological processes and carbon stock mechanisms [17,18,19]. In summary, although each method has advantages, their underlying logic largely relies on traditional spatial autocorrelation and heterogeneity theories, resulting in identification outcomes often concentrated in areas adjacent to existing high-quality arable lands, lacking systematic recognition of regions with both high production potential and elevated ecological risk, thereby constraining the scientific rigor and applicability of arable land identification strategies [20,21].

Meanwhile, research on nature-based solutions (NbS) addressing the synergy between soil carbon sequestration and food production within agricultural ecosystems remains insufficient, limiting robust support for coordinated management of arable land conservation and carbon sinks [22,23]. Notably, in addition to structural NbS, field-scale organic amendments have proved effective in enhancing SOC stabilization and reshaping soil bacterial communities [24], suggesting that ecological high-risk cropland can also be upgraded through soil-quality-centered measures. However, studies on the regional-scale synergy between carbon stocks and food production remain limited, and models capable of systematically elucidating the intrinsic mechanisms linking arable land ecological functions and carbon cycling are lacking.

To address these challenges, this study aims to transcend reliance on traditional spatial autocorrelation and heterogeneity theories and methodologies, focusing on the critical issue of reconciling carbon stock enhancement with food security in ecologically high-risk areas. Using Yunnan Province as a case study, we integrate nearly two decades of remote sensing imagery, soil attribute data, and policy information to construct a reserve arable land and priority reforestation zoning identification system based on the principle of geographic similarity. This framework incorporates environmental factors such as slope and proximity to water, alongside relevant policy conditions. Through evolutionary analysis and multi-scenario simulations, high-quality arable land reserves and priority reforestation areas are identified, and the synergistic potential between carbon sequestration and food security is quantitatively assessed. Building on this foundation, we innovatively develop a geography similarity-driven “grain–carbon” synergy gain framework and propose a bidirectional optimization strategy: on one hand, employing an ecological risk-centered “negative exclusion” mechanism to promote reforestation and enhance carbon stocks; on the other hand, optimizing the spatial configuration of arable land to achieve coordinated improvement of food production capacity and ecological functions. This research not only systematically reveals the carbon stock-arable land synergy mechanisms in ecologically high-risk regions, providing theoretical support for further research, but also expands the regional application of NbS in agricultural ecosystems and offers theoretical guidance and decision-making support for sustainable land management under the dual-carbon framework [25,26].

2. Theoretical Frameworks

2.1. Theoretical Foundations of Geographic Similarity

To address the challenge of simultaneously enhancing carbon stocks and ensuring food security in ecologically high-risk areas, this study develops a “grain–carbon” synergy gain framework grounded in geographic similarity. This framework is inspired by Zhu et al.’s “Third Law of Geography,” which posits that regions separated by considerable distances but exhibiting highly similar geographic environments may demonstrate analogous geographic processes and functional responses [27]. Building on this premise, the theory of Geographic Similarity (GS)—which emphasizes environmentally driven spatial attribute homogeneity—is proposed as the core theoretical foundation [28].

Compared to traditional GIS–MCDA, CA–Markov, and InVEST frameworks that rely on proximity assumptions and struggle to handle regions with strong environmental heterogeneity, this approach overlooks the potential of non-adjacent but environmentally similar areas, the GS theory quantifies multidimensional environmental factor similarity indices to identify environmentally homogeneous yet spatially distant regions, thereby enabling prediction of geographic process evolution trends. This approach transcends the limitations of traditional spatial predictions that rely heavily on spatial proximity and auto correlation. GS theory has demonstrated superior performance compared to mainstream spatial interpolation and machine learning models in domains such as soil classification [29], landslide susceptibility assessment [30], and sample bias correction [31], especially in sample-sparse and ecologically complex regions. Consequently, GS theory serves as a pivotal spatial cognition tool within complex ecosystems, facilitating precise delineation of suitability patterns in highly heterogeneous landscapes and uncovering latent ecological and productivity values. It thus provides a critical theoretical basis for the identification and management strategies of arable land in ecologically high-risk zones.

2.2. Ecological Restoration and Grain–Carbon Synergy Potential in Ecologically High-Risk Areas

Ecologically high-risk areas commonly confront challenges such as land degradation, declining ecological functions, and heightened agricultural vulnerability, making them critical zones for ecological governance and sustainable agricultural development amid industrialization and urban expansion [32,33]. These regions are internally heterogeneous: certain parcels with gentle slopes, fertile soils, and favorable water availability retain substantial agricultural potential despite overall ecological fragility. To capture this variability, we develop a spatially explicit, geographic similarity–based framework that integrates topography, hydrology, soil properties, and policy constraints to identify cropland parcels with controllable ecological risks and high production potential. This approach establishes a rigorous basis for graded risk management and precision-oriented land allocation.

Research indicates that despite significant ecological pressures, these regions possess considerable potential for synergistic ecological restoration and agricultural productivity enhancement. For instance, Cumming et al. Highlighted that ecological restoration, land conservation, and social-ecological feedback mechanisms can improve soil fertility and resource regulation [34]. Kennedy et al. Emphasized the integration of ecosystem services and land-use planning at the landscape scale to simultaneously enhance ecological connectivity and production capacity [35]. Schneider et al. Demonstrated that combining high-resolution land cover data with agricultural suitability models can guide precision land use, thereby mitigating degradation risks [23]. Furthermore, practices such as soil and water conservation, ecological engineering, and fiscal incentives have been proven effective in boosting ecological benefits and food security in high-risk arable lands [36,37].

Nature-based solutions, such as reforestation through the conversion of farmland to forest, provide a systematic pathway for coordinating ecological restoration and food security. These measures not only mitigate soil erosion and restore ecological functions but also significantly enhance regional carbon stocks and biodiversity [38,39]. Through vegetation recovery and soil carbon accumulation, this approach strengthens terrestrial carbon sinks, thereby supporting carbon neutrality strategies [40]. Meanwhile, integrating NbS with ecological risk assessment and spatial optimization techniques establishes a multi-scale, multi-scenario dynamic evaluation and planning framework for coordinated management of arable land protection and carbon stock enhancement. In ecologically high-risk areas, reforestation facilitates the optimization of land-use structure, balancing the dual objectives of ecological restoration and food security. Therefore, deepening the synergistic integration of NbS principles and spatial decision-support tools provides critical support for constructing a “grain–carbon” synergy gain framework, ultimately enabling multi-objective integrated regulation at the regional scale.

3. Materials and Methods

3.1. Overview of the Study Area

Yunnan Province is located at the core of the Yunnan–Guizhou Plateau (97°31′–106°11′ E, 21°8′–29°15′ N), characterized predominantly by mountainous and plateau terrain, with pronounced vertical ecological differentiation and strong spatial heterogeneity (Figure 1). The region spans multiple climatic zones, ranging from the cold temperate to tropical belts and is recognized as a global “Indo-Burma biodiversity hot spot” [41]. Serving as a national ecological security barrier and a significant carbon sink, Yunnan plays a vital role in regional climate regulation and food security assurance. The province boasts a forest coverage rate of 68.67%, with its forest and wetland ecosystems not only supporting carbon sequestration but also contributing to agricultural ecological regulation. Consequently, Yunnan’s unique ecological heterogeneity and carbon stock regulation functions—as a global biodiversity hot spot and a key component of national ecological security—make it a representative and critical study area for this research.

3.2. Data Sources and Processing

3.2.1. Land Use and Fundamental Geographic Data

This study utilizes the three-phase land cover dataset (CLCD) of Yunnan Province from 2000, 2010 and 2020, developed by the research team of Yang Jie and Huang Xin at Wuhan University. The dataset exhibits higher classification accuracy than similar products such as GLC_FCS30 and Global3 [42]. To ensure compatibility with the InVEST model, the original data were reclassified into seven categories: Cropland, Forest, Shrub, Grassland, Water, Construction land, and Unused land. In addition, socioeconomic data, meteorological data, soil data, and other necessary auxiliary datasets were incorporated into the analysis, with detailed information presented in

Table 1. Data Sources.

Datatype	Data Name	Data Source
Land use data	2000, 2010, 2020	Wuhan University (https://zenodo.org/records/4417810 (accessed on 11 November 2025))
Socioeconomic data	Population	Resource and Environmental Science Data Plat from (https://www.resdc.cn)
	GDP
	Distance to tertiary Roads	National Catalogue Service For Geographic Information (https://www.webmap.cn/main.do?method=index (accessed on 11 November 2025))
	Distance to Railways
	Distance to Highways
	Distance to Settlements
	Distance to Water
	Distance to Water Systems
Natural environmental factors	Temperature	Eco-Meteorological Cloud Service Platform (https://em.cams.cma.cn/#/dashhome (accessed on 11 November 2025))
	Potential Evapotranspiration
	Sunshine Duration
	Surface Moisture
	Precipitation
	Soil type	Resource and Environmental Science Data Plat from (https://www.resdc.cn)
	Silt Content
	Elevation	Geospatial Data Cloud (https://www.gscloud.cn)
	Slope	Extracted using ArcGIS 10.8
	Aspect

. Among these, data for Population, GDP, Temperature, Potential Evapotranspiration, Sunshine Duration, Surface Moisture, Precipitation, Soil Type, and Silt Content require resampling to a 30 m resolution, while the remaining datasets do not necessitate resampling.

3.2.2. Construction of Carbon Density Database

This study, integrating the characteristics of Yunnan and drawing upon relevant existing research, established a scientifically robust and regionally adapted carbon density database to support the assessment and dynamic simulation of carbon storage in Yunnan [43,44]. For the diverse land cover types in Yunnan, above ground carbon density for cropland, Water, and unused land was estimated following Chen Lijun et al. [45] based on biomass assessments, while Grassland carbon density was derived from field measurements reported by Mo Jinxi [46]. Due to the lack of observed data for below ground carbon density in Water and built-up areas, these values were estimated using a root-to-shoot ratio coefficient of 0.2 [47]. Below ground carbon densities for cropland and forest were obtained from the averaged multi-point samples compiled by Xu Li et al. [48]. Except for built-up areas, soil carbon density values were referenced from the empirical measurements and modeling results by Li Kerang et al. [49]. Dead organic carbon density was calibrated according to the model developed by Paruk et al. [50].

To address data gaps, carbon density values from representative sites in Sichuan [51], Guizhou [52], and Hubei [53] were incorporated using spatial distance-weighted adjustments. Non-local data were localized through national-scale correction models [11,54], integrating calibration factors and regional parameters to establish a high-precision carbon density database applicable to multiple land cover types and carbon pools in Yunnan (detailed in

Table 2. Carbon Density of Land Use types in Yunnan Province (t/hm²).

Land Usetypes	C_Above	C_Below	C_Soil	C_Dead
Cropland	19.24	12.26	127.24	2.11
Forest	38.57	18.13	167.78	2.78
Shrub	9.54	4.46	80.11	0.92
Grassland	4.79	14.83	125.39	4.71
Water	1.00	0.20	0.00	2.80
Construction land	1.80	0.36	73.00	0.00
Unused land	2.21	0.67	0.00	0.96

). This database provides a fundamental data foundation for carbon stock assessment and dynamic simulation.

$$C_{S\mathrm{p}}=3.3968\times MAP+3996.1$$

$$C_{B\mathrm{p}}=6.798\times {\mathrm{e}}^{(0.0054\times MAP)}$$

$$C_{B\mathrm{t}}=28\times MAT+398$$

$$K_{B\mathrm{p}}={\displaystyle \frac{C_{B\mathrm{p}}^1}{C_{B\mathrm{p}}^2}}$$

$$K_{B\mathrm{t}}={\displaystyle \frac{C_{B\mathrm{t}}^1}{C_{B\mathrm{t}}^2}}$$

$$K_B=K_{B\mathrm{p}}\times K_{B\mathrm{t}}$$

$$K_S={\displaystyle \frac{C_{S\mathrm{p}}^1}{C_{S\mathrm{p}}^2}}$$

$$Y_y=0.41\times {\mathrm{y}}_g+0.19\times y_h+0.4\times y_s$$

In the formula, MAP and MAT represent the annual mean precipitation (mm) and temperature (°C), respectively. From 2000 to 2020, the national averages were 657.37 mm and 10.04 °C, while the corresponding values for Yunnan, Sichuan, Guizhou, and Hubei were 976.12/17.34, 968.26/15.71, 1160.94/15.18, and 1181.65/16.99, respectively. These data were sourced from provincial statistical bureaus and water resource bulletins. C_Sp, C_Bp, and C_Bt denote the corrected soil, biomass, and temperature-adjusted carbon density indices. Correction factors include K_Bp (precipitation), K_Bt (temperature), K_B (above ground biomass), and K_S (soil). Y_y represents the final corrected carbon density value for Yunnan, while yi corresponds to carbon density estimates based on the three provinces.

To assess the reliability of the carbon density database, this study validated it using 12 independent sampling sites. The results show that the corrected estimated soil carbon density (167.78 t/hm²) closely matches the observed value (146.57 t/hm²), with an estimation accuracy of 85.5%, indicating that the database can provide a reliable basis for the assessment and dynamic simulation of regional carbon storage.

3.3. Research Methodology

The technical road map of this study is illustrated in Figure 2. It encompasses the acquisition of land use and environmental data, carbon stock evolution assessment, scenario simulation, vulnerability diagnosis, ecological risk evaluation, and identification of synergistic gain zones. This systematic approach aims to analyze ecosystem response mechanisms, promote coordinated management of carbon storage and farmland, and fill theoretical and methodological gaps in high-risk areas.

3.3.1. Spatiotemporal Dynamic Assessment of Ecological Carbon Storage Driven by Multi-Carbon Pool Synergy

The InVEST model, renowned for its user-friendly operation, modular design, and spatial representation capabilities, addresses the limitations of traditional carbon stock estimation methods in terms of accuracy and adaptability, thus gaining widespread application in carbon stock assessment and land-use optimization [55,56]. This study utilizes the InVEST carbon storage module to integrate above ground biomass, below ground biomass, dead organic matter, and soil carbon pools, systematically analyzing the spatial patterns and temporal dynamics of regional carbon storage, thereby providing a scientific foundation for ecological management and carbon neutrality policies [57].

The total regional carbon stock is calculated as the weighted sum of carbon stocks across different land cover types, using the following formula:

$$C_{T\mathrm{o}tal\_z}={\displaystyle \sum _{\mathrm{i}=1}^n(C_{T\mathrm{o}tal\_i}}\times A_i)$$

where C_{Total_z} is the total regional carbon stock (t); C_{Total_i} is the total carbon density of the i-th land use type (t/hm²); A_i is the area of the i-th land use type within the region (hm²).

3.3.2. Policy-Driven Land Use Spatial Evolution Simulation

Relevant studies indicate that the PLUS model demonstrates significant advantages in mechanism modeling, scenario forecasting, and ecological management, making it a key tool for urban land system simulation [18,58,59]. Therefore, this study adopts the PLUS model as the core method for land use change simulation and optimization analysis. Based on policies and literature, the 2000, 2010, and 2020 data of the 18 key factors listed in Table 1, excluding elevation, were selected as driving factors for predicting land use categories in 2040 (see Figure A1).

The multi-scenario forecasting aims to reveal land use trends under different future development pathways, providing scientific support for land management policies [60,61]. Based on Yunnan Province’s spatial planning and protection policies, this study designs four development scenarios, detailed in

Table A1. Land Use transition Cost Matrix from 2020 to 2040.

	ND								UD
Land types	A	B	C	D	E	F	G	Land types	A	B	C	D	E	F	G
A	1	1	1	1	1	1	1	A	1	1	1	1	1	1	1
B	1	1	1	1	1	1	0	B	1	1	1	1	1	1	0
C	1	1	1	1	1	0	0	C	1	1	1	1	1	1	0
D	1	1	1	1	1	1	1	D	1	1	1	1	1	1	1
E	1	1	0	1	1	1	1	E	0	0	0	0	1	1	1
F	1	0	0	1	1	1	0	F	0	0	0	0	0	1	0
G	1	1	0	1	1	1	1	G	1	1	1	1	1	1	1
	CP								EP
Land types	A	B	C	D	E	F	G	Land types	A	B	C	D	E	F	G
A	1	0	0	0	0	0	0	A	1	1	1	1	0	0	1
B	1	1	1	1	1	1	1	B	1	1	1	1	0	1	1
C	1	1	1	1	1	1	1	C	1	1	1	1	1	1	1
D	1	1	1	1	1	1	1	D	1	1	1	1	1	1	1
E	1	1	0	1	1	1	0	E	1	0	1	1	1	1	0
F	0	0	0	0	0	1	0	F	0	0	1	1	1	1	1
G	1	1	0	1	1	1	1	G	1	1	1	1	1	1	1

Note: 0 indicates no transition. 1 indicates transition. A represents Cropland, B represents Forest, C represents Shrub, D represents Grassland, E represents Water, F represents Construction land, G represents Unused land.

and

Table A2. Neighborhood Weights and Land Use Demands under Different Scenarios in 2040.

Multiple Scenarios		A	B	C	D	E	F	G
ND	S	100,001,735	298,949,558	10,468,210	19,648,109	3,487,951	2,402,507	780,209
	W	0.52	0.92	1.00	0.00	0.65	0.75	0.63
UD	S	101,096,227	298,274,644	10,473,577	19,126,404	3,376,359	2,616,491	775,777
	W	0.494	0.916	1.000	0.000	0.657	0.772	0.634
CP	S	107,427,932	293,549,834	10,589,011	18,136,352	3,207,221	2,051,484	776,444
	W	1.000	0.366	0.771	0.000	0.543	0.581	0.538
EP	S	97,868,162	301,092,975	10,436,607	20,141,494	3,497,637	1,979,882	721,520
	W	0.169	1.000	0.855	0.000	0.535	0.577	0.506

Note: S represents Quantity, W represents Weight, A represents Cropland, B represents Forest, C represents Shrub, D represents Grassland, E represents Water, F represents Construction land, G represents Unused land.

.

(1) Natural Development Scenario (ND): Serving as the baseline scenario, it follows the land use change patterns observed from 2000 to 2020, without incorporating any policy constraints. Land demand for 2040 is predicted using the Markov chain and used as input for the PLUS model.

(2) Urban Development Scenario (UD): Based on the “Yunnan Province Master Plan for Territorial Spatial Development (2021–2035)” approved by the State Council and Yunnan Province, the expansion of urban areas by 2035 is restricted to no more than 1.293 times the current size. Areas restricted from land use conversion are delineated by overlaying historical built-up land and selecting slopes less than 2°. Markov transition probabilities are adjusted as follows: conversion from cropland to built-up land increases by 40%; from Forest, Shrub, and Grassland to built-up land increases by 30%; conversion from built-up land to other land types decreases by 29.3%; and conversion from unused land to built-up land increases by 60%.

(3) Cropland Protection Scenario (CP): Based on the “14th Five-Year Plan for Agricultural and Rural Modernization in Yunnan Province,” which implements the strictest cropland protection policies to strengthen the quantity, quality, and ecological conservation of cropland. Areas with long-term stable cropland are identified as restricted conversion zones, referring to the “Agricultural Land Grading Regulations” and the “Notice on Cropland Quality Grades in Yunnan Province 2019” issued by the Yunnan Provincial Department of Agriculture and Rural Affairs. Markov transition probabilities are adjusted as follows: conversion from cropland to built-up land is reduced by 60%; conversion from cropland to Forest, Grassland, and Water is reduced by 30%, 30%, and 40%, respectively; conversion from unused land to cropland is increased by 50%, aligning with cropland protection policy requirements.

(4) Ecological Protection Scenario (EP): Based on the “14th Five-Year Plan for Ecological Environment Protection in Yunnan Province” and the “Ecological Protection Red Line of Yunnan Province,” land-use transition dynamics are optimized on the basis of natural development. The probability of conversion from built-up land to cropland, Water, Forest, and Grassland is increased by 35%. Conversely, the probabilities of conversion from cropland, Forest, Grassland, and Water to built-up land are reduced by 50%, 30%, 30%, and 20%, respectively. Triple control measures are strengthened on unused land, decreasing its conversion probability to built-up land by 60%, while enhancing its ecological restoration potential by increasing the probability of conversion to Forest, shrubs, and Grassland by 40%. The scenario promotes the return of cropland on steep slopes to forest and Grassland, increasing this transition probability by 20%. Coupled with ecological red line constraints, this ensures the integrity of ecological spaces and carbon sink functions, highlighting the synergistic effects of Yunnan’s “Three Screens and two Belts” ecological barrier and restoration efforts, thereby achieving sustainable protection goals for the mountain-watershed-border complex system.

3.3.3. Method for Identifying High-Quality Cropland Based on Geographical Similarity

In highly heterogeneous mountainous areas, traditional proximity-based principles struggle to fully identify the potential productivity and ecological functions of cropland. The principle of geographic similarity, as a complement to the first law of geography (spatial auto correlation) and the second law of geography (spatial heterogeneity), indicates that spatial units with highly similar environmental variables tend to exhibit convergent geographic processes and functional attributes [62]. This principle overcomes the limitations of proximity, revealing functional equivalence among non-adjacent plots, and provides theoretical support for the structural identification and attribute inference of high-quality reserve farmland.

The geographic similarity calculation framework used in this study originates from the application by Zhu Axing et al. in predictive soil mapping (PSM) [31]. Based on the multidimensional environmental vectors of each grid cell, the geographic similarity between any unknown cell i and high-quality sample cell j can be calculated as follows:

$$S_{i,j}=P(E(e_{1i},e_{1j}),E(e_{2i},e_{2j}),\dots ,E(e_{vi},e_{vj}),\dots ,E(e_{mi},e_{mj}))$$

Here, S_i,j represents the geographic similarity between unknown point i and sample point j; e_vi and e_vj are the values of the v-th environmental variable at the two points. E(·) and P(·) are functions used to calculate the environmental similarity for each geographic variable and sample, respectively.

Finally, by calculating the weighted sum of similarity scores across all variables, the overall geographical environmental similarity between the unknown point and the sample point can be obtained.

$${\mathrm{S}}_{\mathrm{z}}={\displaystyle \sum _{v=1}^m(w_v\times S_{v(i,j)})}$$

Here, S_z represents the overall similarity degree, W_v is the weight of the v-th variable, and S_v(i,j) denotes the geographic similarity between the v-th variable and the target for points i and j.

In summary, within this theoretical framework, this study develops a methodological system for identifying high-quality reserve farmland based on geographic similarity, consisting of the following four steps:

(1)

Selection and Standardization of Key Geographic Environmental Variables

Based on the formation mechanisms of farmland quality, key environmental variables were selected to systematically characterize the region’s ecological and climatic features. Additionally, constraint to exclude areas unsuitable for development. All variables were uniformly rasterized and standardized to establish a unified environmental data platform, providing a solid foundation for geographic similarity calculations.

(2)

Quantification of Factor Weights Driven by Geographic Detector

To accurately assess the contribution of environmental factors to farmland quality, this study employs the Geographic Detector method proposed by Wang Jinfeng et al. [63,64] to quantify variable importance. this method outperforms traditional approaches by effectively revealing nonlinear spatial heterogeneity and driving mechanisms, and has been widely applied in environmental and land use studies [65,66,67]. The core metric, the q-statistic, reflects the explanatory power of a single factor on the spatial differentiation of farmland quality. In this study, its values are normalized and converted into weights (W) representing the relative importance of the variables.

(3)

A Novel GS Framework for Multidimensional Similarity Modeling

Based on the standardization and weighting of environmental variables, a similarity model is constructed using GS theory to quantify the multidimensional feature differences between candidate grid cells and high-quality farmland, producing a comprehensive similarity matrix.

(4)

Credibility Grading Mapping of Potential Farmland for Sustainable Agriculture

Combining the comprehensive similarity matrix, the “Yunnan Province High-Standard Farmland Construction Plan (2021–2030),” and practical constraints, standards are set to select land types with irrigation and cultivation potential. Priority gradients are delineated using quantile or standard deviation methods, ultimately constructing a multi-level premium farmland reserve map to provide scientific and refined spatial decision support for the optimized allocation of regional farmland resources and high-standard farmland construction.

3.3.4. Three-Dimensional Diagnosis of Ecosystem Vulnerability Under Land Use Intensity Gradients

The IPCC defines vulnerability as a function of exposure, sensitivity, and adaptive capacity, laying the foundation for systematic analysis [68]. Vulnerability, as a dynamic social process, varies significantly among different systems in their response to external pressures [69]. Therefore, conducting baseline vulnerability assessments that incorporate land use change can broaden the scope and dimensions of evaluation [70].

Drawing on the vulnerability assessment framework and Potential Impact Index (PI) constructed by the aforementioned methods [71,72], this study focuses on the Yunnan ecosystem and establishes a dimensionless indicator system to quantify ecological vulnerability driven by land use change. The calculation formula is as follows:

$$L=100\times {\displaystyle \sum _{i=1}^n(D_i\times P_i)}$$

$$PI={\displaystyle \frac{C_{\mathrm{x}}-C_y}{C_x}}÷{\displaystyle \frac{L_x-L_y}{L_x}}={\displaystyle \frac{L_x\times (C_y-C_x)}{C_x\times (L_y-L_x)}}$$

Here, C represents carbon storage, L denotes land use intensity, and x and y correspond to the starting and ending years, respectively; D_i is the land use intensity value at level i, P_i is the proportion of the area occupied by the land use type i, and n is the total number of land use intensity levels. Referring to the study by Zhuang Dafang et al. [73], land use intensity in Yunnan is classified into four levels: Level 1: Unused land; Level 2: Forest, Shrub, Grassland, and Water; Level 3: Cropland; Level 4: Construction land.

3.3.5. Construction of a Multi-Dimensional Landscape Ecological Risk Index and Identification of Spatial Differentiation Characteristics

The Ecological Landscape Loss Index is a comprehensive indicator used to assess the degree of ecological degradation of various landscape types under natural and human disturbances. It systematically reflects the degradation of landscape structure, function, and services driven by factors such as urban expansion and agricultural development. This index is constructed based on two dimensions: landscape disturbance and landscape vulnerability [74].

$$E_{\mathrm{i}}=a{C}_{\mathrm{i}}+b{N}_{\mathrm{i}}+c{F}_{\mathrm{i}}$$

$$R_{\mathrm{i}}=E_i\times S_i$$

Here, E_i is the ecological landscape disturbance index, which includes fragmentation (C_i) [75], separation (N_i) [76], and fractal dimension (F_i) [77], and S_i is the vulnerability index. The vulnerability weights are assigned based on the normalized land use intensity classification in Yunnan: Cropland is 0.1875; forest, Shrub, Grassland, and Water are 0.125; Construction land is 0.25; and unused land is 0.0625. The weighting coefficients a, b and c satisfy a + b + c = 1, and based on the characteristics of the study area and references [78,79], they are set to 0.5, 0.3, and 0.2, respectively.

On this basis, the Ecological Landscape Risk Index (ERI) is constructed to quantitatively assess the ecological risk level of each risk unit. Its expression is:

$$ER{I}_i={\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}(\frac{A_{li}}{A_i}\times R_i)}$$

Here, R_i is the ecological landscape loss index of the i-th landscape type; A_i is the area of the i-th ecological landscape type; A_li is the area of the i-th ecological landscape type within the l-th risk unit.

3.3.6. Ecological Risk–Carbon Sink Synergy-Oriented Bidirectional Optimization Strategy

To implement the “mountain-water-forest-farmland-lake-grass-sand” integrated governance strategy, this study, based on the core methods outlined in Section 3.3.1, Section 3.3.2, Section 3.3.3, Section 3.3.4 and Section 3.3.5, constructs a geography similarity–driven “grain–carbon” synergy framework and proposes the following “bidirectional optimization” strategy:

① Positive Screening: Identify high-quality reserve farmland within ecological high-risk areas.

② Negative Withdrawal: Designate priority areas for returning farmland to forest based on the principle of maximizing carbon sequestration gains.

This framework mainly includes the following four steps:

(1)

Spatial Delineation of Ecological High-Risk Areas

Based on the Ecosystem Vulnerability Index (PI) developed in Section 3.3.4 and the Ecological Landscape Risk Index (ERI) from Section 3.3.5, ecological high-risk areas in Yunnan Province are identified, with a focus on delineating spatial units where natural vulnerability and human disturbances significantly overlap.

(2)

Identification of High-Quality Reserve Farmland

Based on the geographic similarity method described in Section 3.3.3, the comprehensive similarity (Sz) between candidate sites in high-risk areas and high-quality farmland samples is calculated to select GS high-potential farmland plots, forming a multi-level farmland reserve map layer. Relying on the principles of “concentrated contiguous areas” and “ecological friendliness” from the Yunnan High-Standard Farmland Construction Plan (2021–2030), the layout of high-quality farmland is coordinated to establish a “protection-promoted utilization” reserve system.

(3)

Diagnosis of Low-Quality Farmland and Prioritization of Farmland Conversion

Based on Cultivated land quality grade (GB/T33469-2016) [80], a three-dimensional identification system integrating ecological vulnerability, ecological landscape risk exposure, and terrain constraints is developed to accurately delineate priority patches for farmland-to-forest conversion.

(4)

Quantification of Carbon Sequestration Gains from Farmland-to-Forest Conversion

Based on the PLUS model in Section 3.3.2, a multi-scenario simulation is constructed, comprising baseline scenarios (ND, UD, CP, and EP) and optimized scenarios (a coordinated regulation pathway centered on the preservation of high-quality farmland and the withdrawal of low-quality farmland), combined with the InVEST carbon storage module in Section 3.3.1 to obtain changes in carbon storage and net carbon gain (ΔC) under each scenario.

4. Results

4.1. Multi-Scenario Land Use and Carbon Storage Evolution Process and Prediction Results

4.1.1. Land Use Prediction Accuracy Validation

To verify the predictive accuracy and temporal robustness of the PLUS model, land use data from 2000, 2010, and 2020 were selected for two-period validation. First, the 2020 land use pattern was predicted using 2010 as the base year and compared in four typical regions (Figure 3). The results showed a Kappa value of 0.82 and an overall accuracy of 92.3%. Furthermore, using the 2000 data to predict the 2010 land use pattern yielded a Kappa value of 0.80 and an overall accuracy of 90.6% (Figure A2). According to relevant studies, when Kappa ≥ 0.75, the predicted results are highly consistent with actual data, indicating that the PLUS model has high applicability and temporal stability for land use prediction in Yunnan Province in 2040 [81,82].

4.1.2. Land Use Long-Term Evolution and Prediction Results

From 2000 to 2020, land use in Yunnan Province was primarily composed of forest land, arable land, and Grassland, collectively accounting for over 96%, presenting a pattern of “more forests, less arable land, and fewer Grasslands”. Among these, forest land had the largest coverage, maintaining a stable proportion of around 68%, while arable land accounted for about 21%, and Grassland remained between 5% and 7%. The overall pattern showed significant spatial stability, reflecting the solid structure of land use and the relative balance of regional ecosystems.

Based on this, the PLUS model prediction, starting from the year 2020, shows that between 2020 and 2040, different scenarios (ND, UD, CP, EP) will significantly impact land spatial patterns and ecological pressure (

Table A3. Land Use Changes under Different Scenarios in Yunnan Province/km².

Land Use Types	2020	2040 ND		2040 UD		2040 CP		2040 EP
	Area	Area	Change	Area	Change	Area	Change	Area	Change
Cropland	85,670.70	89,997.06	4326.36	89,806.43	4135.73	96,680.58	11,009.88	88,076.84	2406.14
Forest	269,329.21	269,296.53	−32.68	269,331.01	1.80	264,235.54	−5093.67	271,492.98	2163.77
Shrub	9941.99	9420.58	−521.41	9414.85	−527.14	9529.30	−412.69	9392.14	−549.85
Grassland	22,741.09	18,092.99	−4648.10	18,130.75	−4610.34	16,537.78	−6203.31	18,344.84	−4396.25
Water	2530.42	2918.08	387.66	2940.59	410.17	2749.97	219.55	2511.64	−18.78
Construction land	1397.83	1765.83	368.00	1868.66	470.83	1761.30	363.47	1725.46	327.63
Unused land	581.40	701.41	120.01	700.19	118.79	698.02	116.62	648.59	67.19

).

In the natural development scenario (ND), the area of Cropland in Yunnan will increase from 85,670.7 km² to 89,997.06 km², with a net increase of 4326.36 km². The area of built-up land will expand by 26.33%, reaching 1765.83 km². During the same period, the areas of forest land (−32.68 km²), shrub land (−521.41 km²), and Grassland (−4648.10 km²) will decrease significantly, reflecting the continuous pressure on the ecosystem caused by construction expansion in the absence of intervention, leading to an intensified risk of ecological degradation.

In the urban development scenario (UD), the area of built-up land will expand to 1868.66 km², with a net increase of 470.83 km², reflecting the acceleration of urbanization. Forest land will slightly increase by 1.80 km², showing some ecological compensation capacity. However, Grassland and shrub land will decrease by 4610.34 km² and 527.14 km², respectively, indicating that the ecological environment degradation risk remains significant.

In the cropland protection scenario (CP), the area of Cropland will significantly increase to 96,680.58 km², with a net increase of 11,009.88 km² (an increase of 12.85%), confirming the effectiveness of Cropland protection. During the same period, forest land will sharply decrease by 5093.67 km², reflecting clear competition and trade-offs between land use functions.

In the ecological protection scenario (EP), priority will be given to maintaining forest land, Cropland, and built-up land, with efforts to limit forest loss. As a result, Grassland and shrub land will decrease by 4396.25 km² and 549.85 km², respectively. This reflects a trend of ecological space reconstruction, where land use pressure shifts from forest land to Grassland and Shrub.

Figure 4 visually presents the spatial patterns and evolution characteristics of land use under different scenarios, helping to analyze the dynamic coupling relationship between land functions in greater depth.

From 2000 to 2020, the Cropland area in Yunnan initially decreased but then rebounded to 85,670.7 km². Under the ND scenario forecast for 2040, it is projected to reach 89,997.06 km². Forest area remained stable, fluctuating between 267,971.23 km² and 269,296.53 km². Shrub and Grassland continued to decrease, with their areas shrinking to 9420.58 km² and 18,092.99 km², respectively. The area of Water showed slight fluctuations and increased by a net 783.12 km². The area of unused land first decreased and then increased, accumulating a net increase of 195.74 km². Built-up land continued to expand, with the fastest growth occurring between 2000 and 2010, and the expansion rate slowing in later years, reflecting a staged adjustment of urbanization.

4.1.3. Long-Term Evolution and Prediction of Carbon Storage

Based on regional-scale model calculations, the carbon storage of forest vegetation in Yunnan Province was 6.121 × 10⁹ t. Compared with the 5.211 × 10⁹ t reported in the “Yunnan Forestry and Grassland Work Conference,” the accuracy reached 82.5%, indicating that the adopted data fusion and climate correction methods are reliable at both plot and regional scales. Using this approach to estimate carbon storage from 2000 to 2020, a slight declining trend was observed, with carbon storage of 7949.68 × 10⁶ t, 7946.20 × 10⁶ t, and 7945.49 × 10⁶ t at the three time points (2000, 2010, and 2020), amounting to a cumulative decrease of approximately 4.19 × 10⁶ t (

Table 3. Carbon Storage by Land type from 2000 to 2040 (×10⁶ t).

Land Use Types	2000	2010	2020	2040
				ND	UD	CP	EP
Cropland	1318.26	1310.59	1378.01	1447.60	1444.54	1555.11	1416.72
Forest	6089.92	6114.86	6120.78	6120.03	6120.82	6005.02	6169.95
Shrub	84.91	98.39	94.48	89.52	89.47	90.56	89.25
Grassland	451.58	414.02	340.48	270.89	271.45	247.60	274.66
Water	0.85	0.88	1.01	1.17	1.18	1.10	1.00
Construction land	3.97	7.28	10.51	13.27	14.04	13.24	12.97
Unused land	0.19	0.18	0.22	0.27	0.27	0.27	0.25
Sum	7949.68	7946.20	7945.49	7942.75	7941.77	7912.90	7964.80

). In terms of spatial distribution, high-carbon storage areas were concentrated in the western regions dominated by forests and grasslands, including Pu’er, Xishuangbanna, Nujiang, and Dehong; whereas central areas such as Kunming, Qujing, and Yuxi, dominated by cropland and construction land, exhibited lower carbon storage and weaker carbon sequestration capacity.

The 2040 multi-scenario predictions show that: Under the ND scenario, the carbon storage in cropland increases by 69.59 × 10⁶ t, while the carbon storage in Grassland decreases by the same amount. Under the UD scenario, the carbon storage in cropland increases by 66.53 × 10⁶ t, while the carbon storage in Grassland decreases by 69.03 × 10⁶ t. Under the CP scenario, the increase in carbon storage in cropland is the largest, reaching 177.10 × 10⁶ t, while the carbon storage in forest land decreases by the largest amount, 115.76 × 10⁶ t. Under the EP scenario, the carbon storage in forest land increases by 49.17 × 10⁶ t, while the carbon storage in Grassland decreases by 65.82 × 10⁶ t. Overall, the EP scenario shows the highest total carbon storage, reflecting that ecological protection policies significantly enhance the land’s carbon sequestration capacity.

To reveal the spatiotemporal variation of carbon storage, it was classified into three categories: “almost unchanged,” “increased,” and “decreased.” Four regions (a, b, c, and d) were selected for visual analysis (see Figure 5). From 2000 to 2020, the total carbon storage showed a slight decline, with the western region remaining stable, the eastern region slightly increasing, and the central region fluctuating minimally. From 2020 to 2040, the differences across scenarios were significant: under the ND scenario, carbon storage continued to increase in the eastern region, following the historical trend; under the UD scenario, the carbon storage in southeastern farmland and forest areas increased significantly; under the CP scenario, carbon storage in forests and shrubs remained stable amid urban expansion, while Grassland decreased notably; under the EP scenario, carbon storage generally increased slightly, except for shrubs, Grasslands, and water areas.

4.2. Geographic Similarity-Based Identification Results of High-Quality Farmland

To systematically identify high-quality farmland reserves in Yunnan Province and overcome the traditional reliance on spatial proximity and sample density, this study constructs a high-dimensional environmental variable space based on the Geographic Similarity theory. By analyzing the key variable weights, it reveals the spatial pattern of similarity, conducts a graded assessment, and clarifies the spatial boundaries and development priority sequence of high-quality farmland resources.

4.2.1. Contribution of Geographic Environmental Variable Weights and Dominant Mechanisms

Based on the quantified factor importance results from the geographic detector (

Table A5. Analysis of Geographic Environmental Variables’ Weight and Explanatory Power.

Variable	Weight (W)	Geodetector q-Value	Significance Level
Elevation	0.0840	0.08889	p < 0.01
Silt Content	0.0916	0.0969	p < 0.01
Aspect	0.0034	0.0036	p < 0.01
Sunshine Duration	0.0983	0.1040	p < 0.01
Solar Radiation	0.0898	0.0950	p < 0.01
Surface Moisture	0.0713	0.0755	p < 0.01
Precipitation	0.2137	0.2262	p < 0.01
Temperature	0.0733	0.0776	p < 0.01
Potential Evapotranspiration	0.1106	0.1170	p < 0.01
Distance to Water Systems	0.0716	0.0758	p < 0.01
Soil type	0.0924	0.0978	p < 0.01

), climate factors (precipitation and potential evapotranspiration) contribute the most to the distribution of high-quality farmland, with a combined weight of 0.3243. Although temperature has a significant effect (q = 0.077, p < 0.01), its explanatory power is lower (0.0733), and thus it is not included in the core framework. Light conditions (sunshine duration and solar radiation) are the next most important, with a combined weight of 0.1881. Soil properties (sand content and soil type) are one of the key factors determining farmland quality, with a combined weight of 0.1840. Sandy soil influences soil structure and water-nutrient retention. Hydro logical conditions (distance to water systems and surface moisture) have a combined weight of 0.1429, making them another core driving factor group. Among them, the distance to water systems is a critical restrictive factor, with a significant contribution (0.0716), verifying that the condition of ≤6 km from a water system effectively ensures irrigation accessibility. Terrain factors (elevation and slope aspect) contribute relatively less, with a combined weight of only 0.0874. However, the slope constraint of ≤6° effectively excludes areas unsuitable for mechanized farming, highlighting the role of terrain selection. In conclusion, soil properties (material basis), terrain conditions (spatial carrier), and hydro logical factors (water guarantee) form the “soil-terrain-hydrology” triangular driving mechanism for the formation of high-quality farmland in Yunnan Province, serving as the foundational framework for the spatial differentiation of high-quality farmland resources.

The weight system reveals that the quality of farmland in Yunnan is driven by a “soil-topography-hydrology” triangular mechanism: with soil properties and moisture conditions (such as rainfall, evapotranspiration, distance to water systems, and surface humidity) at the core, and topographic constraints (slope ≤ 6°) ensuring the feasibility of cultivation, though their contribution is indirect. This mechanism provides theoretical support for the GS quantification, significantly enhancing the precision of identifying high-quality farmland and determining the priority sequence for development.

4.2.2. High-Quality Farmland Spatial Pattern and Credibility Grading Assessment

(1) Geographical Similarity Spatial Differentiation Patterns

Based on the “soil-terrain-hydrology” triangular driving mechanism weight system established by the geographical detector, after standardizing and assigning weights to 11 key environmental variables, the GS model was constructed to quantify the multidimensional feature differences between candidate grids and high-quality farmland. The comprehensive similarity was then outputted. Using ArcGIS spatial analysis, the similarity was visualized with a color gradient (Figure 6a), revealing the geographical similarity and spatial differentiation characteristics of high-quality farmland reserves in Yunnan Province.

(1)

High Similarity Core Areas: The Yuxi–Honghe and central–southern Yunnan region constitute the core high-similarity zone (similarity 0.89), dominated by mountain valleys and gentle sloping terraces at 1000–2000 m elevation. The soil, mainly red soil and lateritic red soil, features balanced organic matter and good water retention and fertility. The climate is monsoon-controlled, with an average annual temperature of 18–22 °C and precipitation of 800–1200 mm, distinct dry and wet seasons, and well-matched temperature and precipitation. Irrigation water sources are evenly distributed, forming the most geographically synergistic core for high-quality farmland reserves in the province.

(2)

Transition Zone: The buffer zone surrounding the high-similarity core area exhibits a gradient decrease (yellow to yellow-green), encompassing regions such as Dali, Kunming, and Pu’er. Farmland in Kunming is more fragmented than in the Yuxi core area due to differences in terrace reclamation and irrigation. Dali and Lijiang, located further north, have slightly lower accumulated temperatures and delayed precipitation periods, with soils transitioning from red soil to purple soil and arid red soil. This buffer zone reflects the moderating effect of geographic factors on farmland similarity.

(3)

Low Similarity Edge Zone: The northeastern Yunnan (Zhaotong), southeastern Yunnan (Wenshan), and western Yunnan (Dehong–Nujiang–Diqing) constitute low-similarity zones (0.24, green), characterized by highly fragmented farmland, complex terrain, severe soil erosion, and a wide climatic spectrum. Zhaotong lies in the Yunnan–Guizhou Plateau–Sichuan Basin transition belt, with widespread sloped farmland and soil erosion 1.3–4.6 times higher than the core area; Wenshan features prominent karst land forms and marked vertical precipitation differentiation; Dehong–Nujiang comprises deep valleys of the Hengduan Mountains with farmland elevation differences exceeding 3000 m; Diqing has high-cold alpine meadow and dark brown soil farmland. Extreme geographic differentiation results in low connectivity between edge-area high-quality farmland and the core area, making them typical low-similarity zones.

(2) Graded Potential Farmland Reserves and Development Priorities

Based on the comprehensive similarity matrix generated by the GS model, this study overlays the three environmental constraints of slope, proximity to water systems, and land use to exclude undevelopable plots. For the remaining candidate plots, a combination of percentiles and standard deviation methods is used to scientifically determine the similarity threshold (Sz). The farmland reserves in Yunnan Province are then classified in to three levels of credibility: priority, alternative, and to-be-assessed (as shown in Figure 6b and

Table A4. Area Statistics of High-Quality Farmland Reserves in Yunnan Province by Classification.

Classification Attribute	Similarity Threshold(Sz)	Actual Area (km2)	Area Proportion
Priority Level	Sz ≥ 0.80	510.14	0.13%
Alternative Level	0.60 ≤ Sz < 0.80	6665.59	1.70%
Pending Assessment Level	Sz < 0.60	385,016.91	98.17%

). This classification reveals hierarchical differences in geographic similarity and development potential, enabling precise identification and differentiated management of potential farmland.

Priority plots (Sz ≥ 0.80) are highly consistent with high-quality farmland samples across multiple environmental variables, exhibiting significant functional homogeneity. The total area of these plots is 510.14 km², accounting for about 0.13% of the total land area, primarily concentrated in the core region of Yuxi-Honghe. This area has suitable terrain, fertile soil, stable climate, and sufficient water resources. The high similarity ensures low development risks and high benefits, making it ideal for high-standard farmland development and intensive farming practices.

Alternative plots (0.60 ≤ Sz < 0.80) cover an area of approximately 6665.59 km², accounting for about 1.70% of the total land area, and are distributed in central Yunnan and parts of southern Yunnan. These plots differ from core high-quality farmland in terms of their geographic environment, influenced by local variations in terrain, climate, and hydrology. As a result, their functional homogeneity is lower than that of priority plots. Such areas are suitable for inclusion in medium-to long-term reserves, with a focus on improving irrigation systems and soil quality to enhance farmland potential.

The “to-be-assessed” plots (Sz < 0.60) account for 98.17% of the total area, primarily located in the eastern and western edges of Yunnan. These areas are characterized by significant environmental heterogeneity and marked differences from high-quality farmland, with low suitability or ecological sensitivity constraints. Development in these regions should be carefully planned, taking into account field surveys and socio-ecological factors. A differentiated strategy must be developed to avoid blindly applying core area models, ensuring the sustainable development of these regions.

To validate the stability and transferability of the GS model under sparse sample conditions, this study randomly swapped high-quality samples between Weixin County and Hongta District to assess the identification accuracy. The experimental results show that after the sample exchange, the identification accuracy for Hongta District and Weixin County reached 92.7% and 91.5%, respectively (Figure 7). These results demonstrate that the multi-dimensional geographic similarity assessment system based on the GS model can effectively integrate soil, terrain, and hydrological factors, enabling precise identification of high-quality farmland reserves and three spatial zones in Yunnan Province. This method provides a theoretical foundation and technical support for the scientific management of farmland and the optimal allocation of land resources. It has significant practical value and offers strong decision-making support for promoting sustainable agricultural development.

4.3. Bi-Directional Optimization Strategy for High-Risk Areas Driven by Ecological Risk and Carbon Storage Vulnerability

4.3.1. Vulnerability Assessment and Spatial Response of Ecosystem Carbon Storage Services

To systematically assess the impact of land use changes on carbon storage services in Yunnan, a multi-scale evaluation framework is constructed based on provincial and county-level spatial scales. This framework focuses on land use intensity and the PI to conduct vulnerability analysis of carbon storage services. The 2016–2020 land use data is selected with a two-year interval to smooth short-term fluctuations and reflect a more stable trend of changes.

First, from the temporal dimension, between 2016 and 2020, the land use intensity index of Yunnan Province increased from 221.754 to 222.41, showing a steady rise (+0.66), indicating a strengthening of human intervention. Meanwhile, the PI of carbon storage services’ negative effects increased from −0.57 to 0.27, suggesting a significant improvement in ecosystem adaptability and a reduction in negative impacts. This trend reflects the enhanced ecological resilience driven by ecological protection policies and green transformation efforts.

Further, from a spatial perspective, the PI values at the county level show significant heterogeneity from 2016 to 2020. Negative impact areas are concentrated in the peripheral mountainous regions and central urban belt, reflecting the uneven spatial distribution of carbon storage services caused by land use structural differences. Between 2016 and 2018, carbon storage in areas like Suijiang (−37.44), Gongshan (−24.81), and Shuangbai (−15.48) was severely impacted due to the combined pressures of forest degradation and agricultural land expansion. In contrast, areas like Eryuan (+47.04), Chuxiong (+25.49), and Funing (+17.13) showed positive impacts, indicating the significant effectiveness of ecological compensation and the “Return Farmland to Forest” policy.

Due to the lack of a unified standard for carbon storage vulnerability, this study adopts the natural breaks method to categorize the PI values from 2016 to 2018 into four risk levels: Extremely High (PI < 0), Medium-High (0 ≤ PI < 0.25), Potentially High (0.25 ≤ PI < 0.5), and Ecological Safety (PI ≥ 0.5). The same classification is applied for 2018–2020 to ensure consistent temporal comparison. The spatial distribution of landscape ecological risk in Yunnan Province is shown in Figure 8.

Based on the natural breakpoint method, county-level Potential Impact (PI) values are divided into four ecological risk levels: extremely high, medium-high, potential high, and ecological security, aiming to identify regional ecological risks and guide differentiated management. The analysis indicates that extremely high-risk areas exhibit significant declines in carbon storage services and require strict protection and restoration; medium-high-risk areas have limited resilience and should be managed with ecological compensation and land management interventions; potential high-risk areas have fragile foundations but show early signs of recovery, necessitating sustained policy support; and ecological security areas demonstrate strong carbon storage resilience, warranting consolidation of achievements and dissemination of successful experiences to enhance overall performance.

4.3.2. Ecological Landscape Risk Evolution and Its Spatiotemporal Response Analysis

This study identifies the ecological landscape risk in Yunnan Province using six indices: fragmentation, separation, fractal dimension, disturbance, vulnerability, and loss. The Ecological Landscape Risk Index (ERI) for each risk area is calculated. Kriging interpolation and normalization are performed with the help of ArcGIS 10.8, and the natural break method is used to classify the 2000 ERI risk into five levels: low (0 < ERI ≤ 0.279), medium (0.279 < ERI ≤ 0.4), medium-high (0.4 < ERI ≤ 0.596), high (0.596 < ERI ≤ 0.812), and very high (0.812 < ERI ≤ 1). The risk levels for 2010 and 2020 are uniformly based on the 2000 classification standard. The spatial distribution is detailed in

Table 4. Area Changes in Ecological Landscape Risk Levels in Yunnan Province (km²).

Risk Level	2000		2010		2020
	Area	Proportion	Area	Proportion	Area	Proportion
Low Risk	31,052.50	7.92	24,820.67	6.33	68,391.91	17.44
Moderate Risk	96,244.05	24.54	96,618.70	24.64	94,874.75	24.19
Moderate to High Risk	163,217.64	41.62	178,146.07	45.42	176,410.02	44.98
High Risk	100,532.08	25.63	91,430.81	23.31	51,855.06	13.22
Very High Risk	1146.60	0.29	1176.62	0.30	661.13	0.17

and Figure 9.

From atemporal perspective, between 2000 and 2020, the ecological landscape of Yunnan predominantly consisted of medium-high risk areas, which continuously accounted for over 40% of the total area. Medium-risk areas accounted for approximately 24%, with high-risk areas following, fluctuating between 13.22% and 25.63%. Low-risk areas exhibited at rend of decline followed by an increase, reaching 68,391.91 km² by 2020, accounting for 17.44%. High-risk areas expanded between 2000 and 2010 but significantly decreased between 2010 and 2020. To facilitate a better understanding of these changes, Table 4 presents the area changes in the respective ecological landscape risk levels from 2000 to 2020.

During the study period, the area of low-risk and medium-high risk zones increased by 37,339.41 km² and 13,192.38 km², respectively. Other risk levels generally decreased, with the high-risk zone experiencing the largest reduction of 48,677.02 km². This improvement in the ecological landscape mainly resulted from policies such as returning farmland to forest, ecological restoration, and land optimization. These measures alleviated ecological pressure in high-risk areas and promoted the expansion of low-risk zones. Significant progress was made in forest, Grassland, and soil conservation restoration, while ecological compensation and land-use structure adjustments also contributed to the transformation of high-risk areas into low-risk zones.

From the spatial distribution perspective, significant changes in ecological landscape risk occurred in Yunnan Province from 2000 to 2010 (see Figure 9). The ultra-high-risk areas slightly expanded, with some high-risk areas upgrading to ultra-high risk. Most high-risk areas were downgraded to medium-high risk, with a general west-to-east compression. Meanwhile, the medium-high-risk areas expanded from east to west. From 2010 to 2020, ultra-high-risk areas contracted, with some areas downgraded to high risk. High-risk areas experienced a rapid shrinkage, while medium-high-risk areas expanded inward. Low-risk areas notably expanded in the northwest and southwest. These changes reflect the combined effects of natural conditions, policy implementation, and land-use adjustments. Notably, the expansion of low-risk areas in the southwest and northwest reflects the effectiveness of ecological compensation and land-use structure optimization.

Global spatial autocorrelation analysis shows that the Moran’s I value of the ecological risk index in Yunnan Province is 0.9054, with a Z value of 29.65 (p < 0.001), indicating a significant positive spatial autocorrelation and a high degree of spatial clustering (as shown in Figure 10a). Local spatial autocorrelation analysis reveals that ecological risk hotspots are significantly clustered in central, western, and southeastern Yunnan, covering about 5.45% of the province’s area. These areas are characterized by complex terrain, high population density, and frequent agricultural activities; cold spots account for 0% (as shown in Figure 10b). Overall, the ecological risk in Yunnan Province exhibits a spatial pattern of “high values concentrated, low values dispersed”.

The ecological landscape risk in Yunnan Province is significantly negatively correlated with carbon storage. The total carbon storage has decreased, with the main losses concentrated in high-risk areas. Urbanization and agricultural expansion have led to reductions in forests and grasslands, weakening carbon sequestration functions. The disappearance of high-carbon-storage ecosystems has driven the decrease in carbon storage. Although ecological restoration measures like returning farmland to forests have been effective in some areas, the overall carbon storage is negatively correlated with ecological landscape risk. Carbon storage is higher in the southwest and northwest regions, while it is lower in the urbanized and agricultural expansion areas in the central and eastern regions.

4.3.3. Ecological High-Risk Area Bidirectional Optimization in the Grain–Carbon Synergy Mechanism

Based on the PI and ERI indicators to identify high-ecological-risk areas in Yunnan, this study integrates the GS model and a three-dimensional diagnostic system to delineate high-quality farmland and priority areas to return farmland to forests. The PLUS and InVEST models are used to simulate carbon storage changes under different land-use scenarios.

(1)

Spatial Distribution Characteristics of High-Ecological-Risk Areas

This study constructs an ecological risk identification system based on the fusion of the 2019–2020 PI and the 2020 ERI, clearly defining the spatial distribution of high ecological risk areas in Yunnan. The ecological vulnerability areas are identified with PI < 0.5, and the landscape high-risk areas are defined with ERI > 0.4. By overlaying both indices, the ecological high-risk regions are accurately identified (see Figure 11).

The results show that several contiguous high ecological risk areas are formed in the central, western, and southeastern regions of Yunnan, with a total area of approximately 168,101.38 km², accounting for 42.86% of the entire province. These areas exhibit characteristics of high vulnerability, high risk exposure, and uneven spatial distribution. The PI and ERI fusion system scientifically reveals the spatial risk pattern, providing theoretical support for ecological management and land optimization.

(2)

Identification of High-Quality Farmland Reserves

This study, based on the GS model, calculates the similarity (Sz) between candidate farmlands and typical high-quality samples within the ecological high-risk areas of Yunnan. Based on the results, the farmland is classified into priority areas (Sz ≥ 0.80) and secondary areas (0.60 ≤ Sz < 0.80). The priority areas, covering 269.91 km², are concentrated in Kunming and Hong he, with favorable terrain and strong connectivity, making them core areas for priority protection and development. The secondary areas, covering 3077.57 km², are widely distributed in Dali, Chuxiong, Baoshan, and other regions, exhibiting a spatial pattern of “dispersed distribution with local aggregation” (Figure 11a). These two categories together form a high-quality farmland reserve area of 3347.48 km² within the ecological high-risk zones. In conjunction with the provincial protection plan, a high-quality farmland spatial reserve system of 3587.71 km² is constructed under the “protection-promoted utilization” approach, laying a solid foundation for future farmland resource reserves.

(3)

Identification of Low-Quality Farmland and Zoning for Cropland-to-Forest Conversion

A three-dimensional diagnostic framework was developed in this study by integrating ecological vulnerability, landscape risk exposure, and topographic constraints, to systematically identify priority zones for cropland-to-forest conversion in Yunnan Province. In accordance with the Regulations on Returning Farmland to Forest and Cultivated land quality grade (GB/T 33469–2016) [80], and guided by regional ecological function optimization goals, farmland with a slope greater than 26° was used as the primary criterion for identification. As a result, a total of 3133.38 km² of low-quality arable land was delineated, accounting for approximately 5.98% of the ecologically high-risk Cropland. The spatial distribution of this land exhibits a pattern of “contiguous patches with broad geographic coverage” (Figure 11b), with concentrations primarily located in Zhaotong, Dali, and Hong he, among which Zhaotong shows the highest density.

The identified regions are characterized by fragmented terrain, severe soil erosion, high ecological sensitivity, and low agricultural potential. These areas exhibit low ecosystem resilience to extreme climatic events and anthropogenic disturbances, highlighting the critical need for cropland-to-forest conversion. The proposed three-dimensional diagnostic framework aligns closely with national standards, showing a 36.83% overlap with the core zones of ecological conservation red lines, significantly surpassing the expected overlap under random spatial distribution. This result not only validates the ecological relevance of the framework but also confirms its diagnostic effectiveness.

(4)

Carbon Sink Simulation and the Effectiveness of the “Bidirectional Optimization” Strategy

Based on the PLUS multi-scenario simulation framework, four baseline scenarios—ND, UD, CP, and EP—were established to construct a “bidirectional optimization” strategy that simultaneously retains high-quality arable land and withdraws low-quality farmland. Corresponding optimized scenarios were designed for each baseline, and four representative sub regions (a, b, c, d) were selected for spatial visualization and analysis (Figure 12). Carbon storage dynamics under each scenario were assessed using the InVEST Carbon Storage and Sequestration model. Results indicate a general increase in net carbon gain (ΔC) under the optimized scenarios (

Table 5. Comparison of Carbon Storage and Net Carbon Gain under Different Land Use Scenarios in 2040 (×10⁶ t).

Scenarios	Baseline Scenario in 2040	Optimized Scenario in 2040	Net Carbon Gain (ΔC)
ND	7942.75	7946.95	4.2
UD	7941.77	7942.27	0.5
CP	7912.9	7918.08	5.18
EP	7964.8	7969.03	4.23

). Specifically, net carbon gains for the ND, UD, CP, and EP scenarios reached 4.2 × 10⁶ t, 0.5 × 10⁶ t, 5.18 × 10⁶ t, and 4.23 × 10⁶ t, respectively. Using the 2020 total carbon stock (7945.49 × 10⁶ t) as a baseline, all baseline scenarios—except EP—exhibited declines in carbon storage, with the CP scenario showing the largest reduction (–32.59 × 10⁶ t). In contrast, under the optimized scenarios, the EP scenario achieved the greatest improvement, with a net carbon gain of 23.54 × 10⁶ t. Although carbon storage in the UD and CP scenarios still declined, the rates of loss were significantly mitigated, while ND shifted to a positive growth trend. The enhancement of carbon sinks is primarily attributed to the synergistic effect of retiring inefficient farmland and intensifying the use of high-quality cropland. This reflects a triple benefit of “ecological restoration–productivity safeguarding–carbon sequestration enhancement,” reinforcing the region’s carbon sink potential. Among all scenarios, the EP scenario demonstrated the most favorable performance, achieving the highest net carbon gain of 23.54 × 10⁶ t.

To validate the actual impact of the ERI on spatial decision-making and quantify the correlation between carbon sink changes and land use optimization results based on GS, this study selected two land use types: cropland and forest, and compared them with the original baseline scenario. Based on the PI and ERI fusion identification system, combined with geographic similarity, high-quality cropland within ecological high-risk areas was selected, and four scenarios were constructed. Meanwhile, an optimization scenario centered on preserving high-quality cropland and withdrawing low-quality cropland was selected according to this identification system. By comparing changes in land use pixel categories, area changes, and their relationships with carbon storage under different scenarios (Figure 13), the effectiveness of the optimization strategy was further evaluated.

The results indicate that the PI and ERI fusion identification system developed in this study can accurately delineate the ecological high-risk areas in Yunnan Province, which account for 42.64% of the total land area (Figure 11). By incorporating a three-dimensional threshold system that includes slope > 26°, vulnerability, and risk exposure, the proposed “bidirectional optimization” strategy forms a “protection-promoted utilization” farmland reserve system, comprising 510.14 km² of priority areas and 3077.57 km² of alternative areas, along with a cropland-to-forest conversion list totaling 3133.38 km². This has led to a significant expansion of high-quality arable land by 454.33 km². Multi-scenario simulations validated that this optimization path significantly enhances carbon sequestration (with the EP scenario achieving a carbon gain of 23.54 × 10⁶ t), showing a positive correlation between carbon storage changes and land-use optimization results (Figure 13 and Table 5). At the same time, the intensive use of farmland is ensured. The results highlight the effectiveness of the PI and ERI fusion identification system, combined with geographic similarity, in selecting high-quality farmland and cropland-to-forest conversion areas. This not only optimizes land use but also effectively improves carbon storage, providing solid theoretical support for the integrated management of the “mountain-water-forest-farmland-lake-grass-desert” system and land optimization in global high-risk areas.

5. Discussion

5.1. Applicability and Advantages of the Spatial Distance-Weighted Carbon Density Correction Method

In long-term studies of carbon storage dynamics, the quality and spatial distribution characteristics of carbon density data across land use types are critical determinants of estimation accuracy and stability [83,84]. Traditional approaches often apply uniform carbon density corrections based on climatic factors at the national scale. While this method is relatively straightforward to implement, it relies heavily on secondary derivation from external regional datasets, which can lead to error accumulation and spatial mismatches—particularly in areas lacking empirical observations—thereby compromising the reliability of carbon storage estimates. Previous research has emphasized that the completeness and representativeness of regional carbon density data are essential for improving the spatial accuracy of carbon storage modeling. Accordingly, there is an urgent need for high-quality correction methods that are spatially adaptive and context-sensitive [85,86].

To address this issue, this study proposes a spatial distance-weighted carbon density correction strategy. This method leverages known carbon density data from regions adjacent to Yunnan and employs spatial interpolation to fill data gaps. Weights are assigned based on the inverse of spatial distance, ensuring a smooth transition of values while preserving local heterogeneity. Unlike climate-driven correction models, which often result in regional mismatches and error propagation due to uniform adjustment, the spatial weighting approach minimizes such discrepancies.

The spatial distance-weighted correction method achieves a carbon density accuracy of 85.5% and a carbon storage prediction accuracy of 82.5%. Empirical results demonstrate that this method outperforms traditional approaches in both spatial accuracy of carbon density reconstruction and local adaptability (see

Table 6. Comparison of Multi-Source Carbon Density Correction Methods in Yunnan Province.

Land Use Types	Based on Hubei Baseline Adjustment				Based on Sichuan Adjustment
	C_Above	C_Below	C_Soil	C_Dead	C_Above	C_Below	C_Soil	C_Dead
Cropland	16.31	10.77	83.06	2.11	4.42	0.83	98.49	2.11
Forest	29.81	18.09	109.71	2.78	24.89	19.84	127.21	2.78
Shrub	0.00	0.00	0.00	0.00	9.54	4.46	83.20	0.92
Grassland	14.13	16.96	95.36	2.42	3.96	12.87	90.76	2.42
Water	1.57	0.00	70.15	1.78	1.75	0.00	64.26	1.78
Construction land	7.53	1.50	37.61	0.00	0.91	0.09	43.87	0.00
Unused land	10.25	2.05	37.71	0.96	0.65	0.70	56.38	0.96
Land Use Types	Based on Guizhou Adjustment				Based on Spatially Weighted Adjustment
	C_Above	C_Below	C_Soil	C_Dead	C_Above	C_Below	C_Soil	C_Dead
Cropland	35.05	6.66	100.44	2.11	19.24	12.26	127.24	2.11
Forest	55.98	18.79	188.29	2.78	38.57	18.13	167.78	2.78
Shrub	0.00	0.00	0.00	0.00	9.54	4.46	80.11	0.92
Grassland	1.27	1.34	146.59	7.28	4.79	14.83	125.39	4.71
Water	0.00	0.00	0.00	3.98	1.00	0.20	0.00	2.80
Construction land	0.00	0.00	117.82	0.00	1.80	0.36	73.00	0.00
Unused land	0.00	0.00	117.82	0.96	2.21	0.67	0.00	0.96

). This advantage is particularly evident in Yunnan Province, where complex topography and ecological diversity pose significant challenges. By incorporating observed data from surrounding regions, the method effectively mitigates the estimation bias caused by data gaps, significantly improving the accuracy and stability of carbon storage assessments. The integration of multi-source datasets effectively suppresses the systematic bias commonly found in conventional models, enhancing the method’s adaptability and robustness in complex land use patterns.

The spatial distance-weighted correction method demonstrates significant advantages in improving carbon density accuracy and local adaptability, particularly in complex ecosystems with strong spatial autocorrelation. This method is similar to the study by Zhou Y et al. [87], which used remote sensing and the GWRF model to enhance carbon density accuracy, but it exhibits stronger adaptability in local precision and spatial recovery accuracy. The weighted interpolation method by Uddin M S et al. [88] also improved data estimation accuracy, while the spatial distance-weighted correction method in this study further optimizes local adaptability and significantly enhances carbon density reconstruction accuracy. Furthermore, Munyati C’s [89] research indicates that the application of inverse distance weighting interpolation in ecosystem integrity assessment shows similar advantages to the spatial distance-weighted method used in this study, particularly in regions with complex topography and strong ecological heterogeneity. This method excels in carbon storage estimation in Yunnan Province and is highly adaptable to diverse ecological contexts and scales, providing valuable insights and technical support for ecological restoration and carbon storage assessments in other regions.

5.2. Contributions, Limitations, and Future Directions

This study has made significant progress in two key areas. First, it addresses the global challenge of reconciling carbon sequestration gains with food security in ecologically high-risk regions. By leveraging geographical similarity, a synergistic “food–carbon” framework was developed that integrates carbon sequestration enhancement, ecological risk regulation, and farmland protection. This framework incorporates ecological vulnerability diagnostics and landscape risk theory, and it pioneers a novel paradigm for farmland identification driven by geographical similarity. In doing so, it overcomes the limitations of traditional spatial auto correlation and spatial heterogeneity theories in identifying high-quality farmland reserves. Through cross-scale scenario simulations and risk-based zoning mechanisms, the framework achieves multi-objective optimization of carbon sink enhancement and food production security.

Second, in response to the challenges of strong ecological heterogeneity and complex spatial coupling in ecologically high-risk areas, this study developed a highly adaptive and scalable technical pathway. This approach focuses on enhancing model adaptability under complex topographic conditions by incorporating the Geodetector method to quantitatively analyze the interactive effects of multiple influencing factors. As a result, the simulation accuracy and robustness of the integrated In VEST-PLUS model were significantly improved. Moreover, the pathway demonstrates strong modular integration capacity and high computational efficiency, supporting multi-source data fusion and large-scale, multi-temporal ecological process simulations. These features collectively highlight its strong potential for cross-regional application and international scalability.

Although this study addresses critical gaps in the coordination between carbon storage and farmland protection in high-risk regions, certain limitations remain. The employed models rely on static parameters, making them insufficient for capturing the nonlinear and dynamic responses of ecosystems—particularly under extreme climatic disturbances. This limitation reduces the sensitivity and accuracy of predictions, thereby introducing a degree of uncertainty into the simulation results. Additionally, the quantification of ecological risk primarily focuses on natural factors, while the integrated impacts of socioeconomic activities are largely overlooked. This constraint limits the contextual adaptability and comprehensiveness of the risk assessment framework.

To address the aforementioned limitations, future research should incorporate process-based ecological simulation models coupled with high spatiotemporal resolution observational data, in order to enhance the realism and robustness of dynamic carbon cycle modeling. Additionally, the development of dynamic human–environment coupling models is essential for integrating ecological processes with socioeconomic drivers. This would involve incorporating variables related to economic activities, policy interventions, and human behavior, thereby improving the scientific foundation of policy-oriented decision-making. The deep integration of process-based mechanisms and empirical observations will provide a solid theoretical and technical basis for the coordinated governance of regional carbon neutrality and food security.

6. Conclusions

This study focuses on the ecological high-risk areas in Yunnan Province and constructs an integrated framework combining multi-carbon pool dynamic simulations, ecological risk assessments, and spatial optimization. Addressing the limitations of traditional methods that rely on spatial autocorrelation and heterogeneity theories and struggle to balance carbon storage gains with food security, this study innovatively proposes a geographically similarity-based “food-carbon” synergistic framework and a bidirectional optimization strategy. Geodetector analysis reveals the coupling driving effects of the “soil-terrain-hydrology” triangular mechanism on the spatial heterogeneity of cropland. Building on this, the PI and ERI fusion identification system accurately delineates ecological high-risk areas (42.64% of the area in 2020) as cropland-to-forest conversion zones while simultaneously selecting potential high-quality farmland with favorable natural conditions, achieving a “negative withdrawal-positive selection” synergy, resulting in a net increase of 454.33 km² of high-quality arable land.

Scenario simulations based on the InVEST-PLUS model show that the optimized scenarios all lead to net increases in carbon storage, with the ecological protection (EP) scenario achieving the largest carbon gain (23.54 × 10⁶ t), significantly reversing the downward trend in carbon storage. Through global and local spatial autocorrelation analysis, this study effectively identifies hotspot and cold spot areas of ecological risk, enhancing the spatial explanatory power of the risk pattern. Additionally, cross-regional sample transfer tests validate the stability (with accuracy > 90%) and strong generalization capability of the geographic similarity (GS) model under sparse sample conditions.

This study breaks through traditional spatial analysis paradigms and systematically reveals the synergistic optimization mechanism of “food-carbon,” providing a quantifiable and replicable decision-making paradigm for achieving the triple goals of “ecological restoration-carbon sequestration-food security.” The framework demonstrates both regional adaptability and scalability, offering scientific support for the delineation of farmland protection red lines and the identification of priority areas for ecological restoration in ecologically fragile regions, facilitating the transformation of regional ecological governance towards precision, systematization, and intelligent management. Future research can further integrate climate change scenarios and multidimensional dynamic data to continuously optimize the long-term mechanism for food–carbon synergy.