Next Article in Journal
Research on the Impact of Carbon Information Disclosure on Corporate ESG Performance: Evidence from China’s Manufacturing Industry
Previous Article in Journal
Can Financial Development Promote Renewable Energy Transition? An Empirical Research Based on Global Panel Data
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Research on Land Ecological Security Diagnosis and Dynamic Early Warning for China’s Top 100 Counties

1
School of Economics, Zhejiang University of Science and Technology, Hangzhou 310023, China
2
Zhejiang Research Centre of Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era, Hangzhou 310023, China
*
Author to whom correspondence should be addressed.
Sustainability 2025, 17(20), 9271; https://doi.org/10.3390/su17209271
Submission received: 6 September 2025 / Revised: 10 October 2025 / Accepted: 16 October 2025 / Published: 19 October 2025

Abstract

Against the backdrop of global climate change and resource-environmental constraints, land ecological security is paramount to regional sustainable development. This study innovatively integrates the DPSIRM system framework with a CNN-LSTM hybrid neural network model to establish a land ecological security early warning system for China’s top 100 counties, enabling scientific diagnosis and dynamic early warning of security incidents. Findings indicate: (1) From 2010 to 2023, land ecological security conditions across counties showed continuous improvement, with the proportion of counties classified as ‘relatively safe’ or higher rising from 2% in 2010 to 68% in 2023. (2) The comprehensive early warning index exhibited a ‘stepwise leap’ trend, progressing through four stages from ‘relatively unsafe’ to ‘relatively safe’. (3) The six subsystems exhibited markedly divergent evolutionary trajectories, characterised by dual-core leadership from ‘driving-management’, fluctuating improvements in ‘pressure-impact’, and low-amplitude oscillations in ‘state-response’. (4) Over the next five years, the comprehensive early warning index will exhibit a ‘gradual stabilisation and upward trend’, yet subsystems will display a polarised pattern of ‘three rising, two stagnant, and one declining’. The early warning system developed in this study provides local decision-makers with critical leading indicators, supporting differentiated management and source-level interventions. These findings hold significant implications for refining county-level ecological governance and optimising territorial spatial patterns.

1. Introduction

Amidst the dual challenges of intensifying global climate change and tightening resource and environmental constraints, scientifically diagnosing the state of land ecological security and establishing effective dynamic early-warning mechanisms have become leading focal points in international resource management and sustainability science [1,2]. Land ecological security constitutes a core element sustaining human wellbeing and the Earth’s sustainable development, with the challenges it faces exhibiting global prevalence. Land degradation currently impacts approximately 60% of global ecosystem services, directly threatening the livelihoods and health of billions of people [3,4]. This grave situation renders sustainable land management pivotal to achieving the United Nations 2030 Sustainable Development Goals (SDGs), particularly Goal 15 (Life on Land) [5]. Scholars worldwide have explored this theme through multiple dimensions, such as the DPSIR model framework extensively applied in European Union environmental policy assessments [6], and studies examining the relationship between urban expansion and ecological pattern evolution across North America, Europe, and Asia [7,8,9]. These investigations have laid a solid foundation for understanding the complexity of human–land interactions. Focusing on China, against the backdrop of synergistically advancing new urbanisation and rural revitalisation strategies, the contradiction between land resource development and ecological conservation has become increasingly prominent. Ecological security has been incorporated into China’s national security outlook [10]. Conducting scientific diagnostics and dynamic predictions of land ecological security has become an urgent requirement for implementing ecological civilisation construction and achieving high-quality development [11]. Current research on land ecological security exhibits multidimensional development trends. At the methodological level, the integrated application of principal component analysis, ecological footprint modelling, and landscape pattern indices [12,13,14] has propelled a shift in research paradigms from single-factor assessments towards systematic, comprehensive diagnostics [15]. Regarding research subjects, a dual-track research framework has emerged, encompassing administrative units like provinces, cities, and counties [16,17], urban clusters [18], and economic zones [19], alongside natural units such as mountainous regions [20] and river basins [21]. Regarding technical approaches, the innovative application of tools, including grey prediction models [22], system dynamics simulation [23], and deep learning algorithms [24], has provided technological support for dynamic early warning systems.
However, existing research still holds room for improvement. Firstly, current indicator systems are predominantly constructed based on frameworks such as the ‘pressure-state-response’ (PSR) or ‘state-danger-immunity’ (SDI) models [25,26], or organised across distinct dimensions like natural, economic, and social spheres [27,28]. This approach lacks coordinated management across causal chains, resulting in suboptimal systemic integration. The Driving–Pressure–State–Impact–Response–Management (DPSIRM) model, which incorporates a ‘Management’ (M) subsystem into the DPSIR framework, better captures the interactive relationship between human activities and land ecosystems [29]. Secondly, existing predictive models predominantly employ traditional mathematical modelling approaches or shallow machine learning algorithms, lacking effective integration of deep learning models. This results in significant bottlenecks when handling complex nonlinear relationships and fusing multi-source heterogeneous data, rendering them ill-suited to meet medium-to-long-term ecological decision-making requirements. The CNN-LSTM hybrid neural network model developed herein combines convolutional neural networks’ spatial feature extraction capabilities with the time series analysis capabilities of long short-term memory neural networks. It demonstrates outstanding innovation and applicability in the dynamic prediction of land ecological security, providing a basis for formulating differentiated, forward-looking regulatory schemes at the county level [30]. Thirdly, the existing international literature has paid insufficient attention to micro-regions undergoing rapid industrialisation and urbanisation, particularly typical micro-governance units like China’s counties, which bear intense development pressures and possess high-efficiency governance potential [17]. Research into the evolution mechanisms and early warning systems for land ecological security in these areas remains a field awaiting deeper exploration. China’s Top 100 Counties by Economic Performance (hereafter referred to as the ‘Top 100 Counties’) generate over 10% of the nation’s GDP using approximately 2% of its land, serving as exemplary models for regional development. However, as the vanguard of China’s county-level economic advancement, these counties face dual pressures of land resource over-exploitation and ecological degradation amid rapid industrialisation and urbanisation [31]. Therefore, establishing an early warning system tailored to the developmental characteristics of the Top 100 Counties for scientific diagnosis and dynamic forecasting is an intrinsic requirement for resolving the contradiction between land resource utilisation and ecological conservation and a practical necessity for providing replicable experience for China’s county-level green transition.
This study, grounded in national ecological security strategy requirements and using the Top 100 Counties as a representative sample, constructs a land ecological security early warning indicator system based on the DPSIRM framework. It emphasises the importance of proactive human management and regulation. Applying a vertical and horizontal pull-off method and a CNN-LSTM hybrid neural network achieves scientific diagnosis and dynamic early warning of land ecological security levels, thereby proposing differentiated regulatory pathways. The findings will provide scientific grounds for refining county-level land ecological security management systems and optimising the national territory’s spatial development and protection patterns.

2. Materials and Methods

2.1. Study Area

This study selected the Top 100 Counties from the ‘2024 Research on High-Quality Development of China’s County-Level Economies’ published by the China Academy of Information and Communications Technology as its research subjects. Spatially (Figure 1), these counties exhibit marked unevenness across China’s four major economic zones: 66 are in the East, 18 in the Central region, 13 in the West, and merely 3 in the Northeast. Provincially, Jiangsu, Zhejiang and Shandong stand out with 25, 15 and 12 counties, respectively. This distribution pattern reflects the gradient differences in regional development levels. The eastern region, characterised by economic advancement and high development intensity, faces ecological degradation and environmental pressures. The central and western regions exhibit lower development intensity but possess fragile ecological foundations with weak recovery capacity. Meanwhile, the northeastern region grapples with resource depletion and ecological restoration pressures during its economic transition. Focusing on the Top 100 Counties reveals the conflicting characteristics of land use and ecological conservation across regions, providing a basis for formulating differentiated ecological governance strategies.

2.2. Indicator System

2.2.1. Theoretical Framework Construction

This paper employs the DPSIRM model to construct a theoretical framework for land ecological security systems (Figure 2), thereby scientifically elucidating the complex interactions between human activities and land ecosystems at the county level. Compared to the traditional DPSIR model, the introduction of the ‘Management’ (M) subsystem enhances consideration of human proactive management behaviours, categorising human responses into two dimensions: ‘passive compensation’ and ‘active control’. Compared to studies employing traditional PSIR models or DPSIR models lacking the ‘Management (M)’ dimension [32], the DPSIRM framework more clearly captures the pivotal role of deliberate human intervention in system evolution. Within Figure 2, the long dashed arrows denote the intrinsic mechanisms of land ecological security, the thick solid arrows represent the operational processes of the DPSIRM model, and the dotted arrows indicate the implementation processes of management measures. This ‘risk-feedback-governance’ framework provides a systematic and comprehensive quantitative assessment tool for land ecological security. It lays a robust theoretical foundation for subsequent dynamic early warning and differentiated regulatory strategy formulation [33]. The framework comprises six major subsystems, which are detailed below.
Driving (D) denotes the latent catalysts inducing land ecological issues, functioning as a kinetic engine driven by economic scale expansion, policy radiation disparities, and rapid population concentration. These forces determine the initial intensity of land resource demand at the county level. Regional GDP reflects economic scale, providing financial support for ecological restoration and green infrastructure; proximity to higher-level administrative centres correlates with greater urban radiation intensity and heightened land development pressure; larger populations entail greater demands on land, resources, and the environment, alongside heightened potential risks; higher per capita disposable income correlates with stronger resident willingness to consume ecological goods and pay for environmental protection.
Pressure (P) denotes countries’ negative polluting forces on land and associated resources during industrialisation and urbanisation, directly triggering ecological and environmental changes. Higher population density increases land carrying capacity and ecological pressure; a greater industrial share intensifies emissions and land occupation intensity; higher industrial exhaust emissions exacerbate air and soil pollution; and elevated concentrations of fine particulate matter in surface air indicate poorer ecosystem health.
State (S) denotes the land ecosystem’s actual structural and functional level under the combined influence of drivers and pressures, encompassing measurable characteristics such as soil quality, habitat integrity, and resource carrying capacity. Adequate precipitation facilitates vegetation and wetland restoration alongside soil conservation; sunlight enhances ecosystem primary productivity; increased water area proportion elevates ecosystem regulation and ecological service functions; Expanding per capita arable land area comes at the cost of sacrificing natural ecological space.
Impact (I) denotes the positive and negative feedback effects arising from altered land ecological conditions on county-level economic development, resident welfare, and social stability. Higher night-time light intensity reflects greater ecological space compression due to built-up land expansion; ample savings among urban and rural residents enhance their capacity to invest in environmental protection and ecological quality improvement; consumption upgrades stimulate demand for green products, indirectly promoting eco-friendly industries; an increased share of the service sector reduces pollution intensity per unit of land.
Response (R) denotes passive remedial measures adopted by governments, enterprises, and the public to mitigate or reverse land ecological degradation, such as environmental investments, industrial phase-outs, and ecological restoration projects. Higher fiscal expenditure ensures more adequate funding for ecological restoration and pollution control; increased agricultural labour implies greater engagement in land ecological restoration and conservation activities; a lower proportion of impervious surfaces moderates surface runoff and the urban heat island effect; a reduction in the total industrial output value of enterprises above designated size is typically accompanied by decreased energy consumption, emissions, and land use pressure.
Management (M) denotes proactive interventions that decision-makers implement to ensure the long-term security of land ecosystems. This encompasses institutional design, spatial control, and green governance, representing a pivotal transition for counties from ‘end-of-pipe treatment’ to ‘source prevention’ [34]. Specifically, higher levels of fixed-asset investment across society can be channelled into green infrastructure and ecological projects; stronger public cultural service capabilities correlate with heightened environmental awareness and participation among residents; and improved medical facilities enhance regional living environment quality and ecological habitability.

2.2.2. Indicator System Design

The Top 100 Counties exhibit characteristics of dense ecological economies, high land development intensity, and abundant environmental governance resources. This study comprehensively evaluates the integrated disturbances to land ecosystems caused by economic expansion, industrial transformation, population concentration, and ecological governance in these counties. Drawing upon indicator frameworks from multiple relevant literature sources [12,32,35] and incorporating expert consultation, this study established a county-level land ecological security early warning indicator system within the DPSIRM framework. This system encompasses six subsystems—Driving, Pressure, State, Impact, Response, and Management—and comprises 23 indicators, selected based on principles of data representativeness, accessibility, and validity (Table 1).

2.3. Data Sources

Research data primarily originates from the China County Statistical Yearbook (County and City Edition), the China Urban and Rural Construction Statistical Yearbook, county-level statistical yearbooks, statistical bulletins of counties (cities, districts), and official publicly available data from provincial and municipal departments of ecology and environment, departments of natural resources, among others. It also draws upon data resource sharing platforms such as the National Earth System Science Data Centre and the Resource and Environmental Science Data Platform of the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences. Partially missing values were imputed using linear interpolation with data from adjacent years.

2.4. Research Methods

2.4.1. Vertical and Horizontal Pull-Off Method

In conducting ecological security diagnostics for the Top 100 Counties, this study employs a vertical and horizontal pull-off method to overcome the limitations of traditional static evaluation methods in handling the spatiotemporal coupling characteristics of panel data [36]. Constructing a three-dimensional data matrix reveals the overall developmental patterns of each evaluation subject from both temporal and spatial perspectives [37]. Traditional evaluation methods predominantly employ principal component analysis and entropy weighting. The former focuses on extracting principal components from data, being well-suited for cross-sectional data at a single point in time [38], while the latter emphasises weight allocation based on data variability, lacking consideration for dynamic temporal changes [39]. The vertical and horizontal pull-off method first applies dynamic standardisation to the indicator layer, then calculates the symmetric matrix of standardised data. The normalised eigenvector corresponding to the matrix’s maximum eigenvalue is determined as the indicator weight (Table 1). Weighted multiplication of standardised data with these weights yields the land ecological security early warning index [40].

2.4.2. Cluster Analysis

To further categorise the land ecological security early warning levels for the Top 100 Counties, this study employs a combined classification strategy of the Elbow Method and Natural Break Method. The Elbow Method analyses inflexion points in the sum of squared errors (SSE) curves corresponding to different cluster numbers K [41]. At K = 5, the rate of decrease in SSE exhibits a significant change, forming an ‘elbow’ that demonstrates favourable intra-cluster compactness and inter-cluster separation-determining K = 5 as the optimal classification number for the land ecological security early warning index. The Jenks Natural Breaks method was then applied to grade the comprehensive evaluation index under spatial autocorrelation constraints [42], forming a five-tier early warning system (Table 2). This approach ensures the spatial clustering characteristics of the grading results while enabling the visual representation of alert information.

2.4.3. CNN-LSTM Hybrid Neural Network Model

To forecast the dynamic evolution trends of land ecological security levels across the Top 100 Counties, this study employs a hybrid CNN-LSTM neural network prediction model. This approach combines Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) networks, leveraging the former’s capability to extract features from high-dimensional data and the latter’s proficiency in processing time series data to capture temporal patterns [43]. The model’s operational workflow involves pre-processing raw data within the input layer. Processed data then enters the hidden processing layer, where the CNN module extracts spatial features and converts them into vectors, while the LSTM module extracts temporal features. Finally, after comprehensive feature integration via the fully connected layer, the output layer generates the prediction results [44].
This study selected three representative benchmark models for comparative experiments to validate the superiority of the CNN-LSTM hybrid neural network model in land ecological security early warning. The first is the Autoregressive Integrated Moving Average (ARIMA) model, a classical time series forecasting model used to benchmark the performance of traditional statistical methods. The second is Support Vector Regression (SVR), a mature machine learning model employed to evaluate the performance of shallow machine learning algorithms. The third model employed was the Long Short-Term Memory (LSTM) neural network, utilised to isolate and evaluate the contribution of the Convolutional Neural Network (CNN) feature extraction module within the model. All comparison models were trained and evaluated using identical training and testing datasets, with Mean Absolute Error (MAE) and Coefficient of Determination (R2) serving as unified performance evaluation metrics.

3. Results and Analysis

3.1. Analysis of Temporal Patterns in Land Ecological Security Across the Top 100 Counties

3.1.1. Analysis of Alert Situation in County Areas

From 2010 to 2023, the proportion of counties within the Top 100 Counties classified under various land ecological safety alert levels underwent significant changes (Figure 3a). The proportion of counties reaching ‘Relatively Safe (Low Alert)’ or higher levels increased from 2% in 2010 to 68% in 2023. The ‘Critically Safe (Medium Alert)’ level became a transitional zone, rising from 6% to 40% before stabilising around 30%. Conversely, the proportion of counties classified as ‘Relatively Unsafe (High Alert)’ or below rapidly declined from 92% to 7%. The 2010–2023 warning level transition string diagram (Figure 3b) provides a more intuitive observation of shifts between warning levels among the Top 100 Counties. Transitions from ‘Relatively Unsafe (High Alert)’ to ‘Relatively Safe (Low Alert)’, ‘Unsafe (Giant alert)’ to ‘Relatively Safe (Low Alert)’, and ‘Relatively Unsafe (High Alert)’ to ‘Safe (No Alert)’ reached 21%, 19%, and 17%, respectively. This indicates that most counties successfully escaped high-risk land ecological security conditions, achieving cross-level improvements.
Based on the characteristics of safety level improvements, the Top 100 Counties were categorised into three types. The first type comprises ‘Significant Progress’ counties where the land ecological safety alert level improved by at least three tiers, accounting for 32% of the total. Examples include Jiaozhou City, Feidong County, and Tianmen City, where the land ecological safety alert level progressively rose from ‘Unsafe (Giant Alert)’ in 2010 to ‘Safe (No Alert)’ by 2023. Specifically, Jiaozhou City has vigorously developed strategic emerging industries, with its high-tech enterprises increasing from 49 in 2012 to 888 in 2022—an expansion of 18.12 times. Fei Dong County and Tianmen City have invested substantially in ecological restoration and environmental infrastructure. Fei Dong County allocated ¥221 million to remediate the geological environment of mines around Lake Chaohu, restoring 3703.5 mu of farmland and forest land. Tianmen City invested ¥2.366 billion in the comprehensive management project for the Tianmen section of the Hanjiang River basin, effectively improving the ecological condition of the land. The second category comprises the ‘Consistently Leading’ counties, whose alert levels reached at least ‘Critically Safe (Medium Alert)’ and have largely remained at ‘Safe (No Alert)’ in recent years, accounting for 14% of counties. Examples include Yiwu City and Xichang City, which maintained ‘Relatively Safe (Low Alert)’ or higher throughout the study period. This achievement stems from Yiwu’s green transformation of its small commodities supply chain, where exports of new energy products such as solar cells grew by approximately 30.0% year-on-year in the first quarter of 2023. It is also attributable to Xichang’s implementation of the ‘Three Retreats, Three Returns’ policy for the Qionghai Wetland, which restored 1067 hectares of lake wetlands and riparian zones. These measures enable a sustained equilibrium between economic development and ecological conservation. The third category comprises ‘Relatively Lagging’ counties whose warning levels never exceeded ‘Critically Safe (Medium Alert)’, accounting for 22% of the total. Counties such as Gongyi, Yongcheng, Wu’an, Renqiu, Shehong, Guangrao, and Etuokeqi consistently registered land ecological safety warnings below ‘Critically Safe (Medium Alert)’. These counties exhibit high resource dependency yet lack ecological compensation mechanisms, facing significant ecological pressures and risks. Challenges such as obstructed industrial transformation and relatively weak ecological restoration capacity constrain improvements in land ecological security levels.

3.1.2. Analysis of Comprehensive Alert Situation

From 2010 to 2023, the comprehensive early warning index for land ecological security in China’s top 100 counties exhibited a ‘step-like surge’ (Figure 4), rising from 43.52 in 2010 to 55.00 in 2023—an increase of 26.31%. The land ecological security rating has moved from ‘Relatively Unsafe’ to ‘Relatively Safe’, with the alert level downgraded from ‘High Alert’ to ‘Low Alert’. This indicates that under the impetus of policies such as ‘ecological civilisation construction’, the ‘dual carbon strategy’, and ‘high-quality county-level development’, the Top 100 Counties have significantly optimised their regional ecological security framework through greening industrial structures, intensifying resource utilisation, and institutionalising ecological governance.
Specifically, the comprehensive evolution of land ecological security in the Top 100 Counties can be divided into four periods: 2010–2013 was a period of fluctuation and adjustment, with the warning index oscillating between 43.53 and 46.51, and the warning level consistently remaining at ‘Relatively Unsafe (High Alert)’. This period coincided with the transition from the late 11th Five-Year Plan to the early 12th Five-Year Plan in China. County-level economic growth relied on traditional resource-intensive models, while ecological governance policies were still exploratory. The initial implementation of the 2013 Air Pollution Prevention and Control Action Plan triggered short-term disruptions in industrial restructuring, causing index volatility. The years 2014–2016 constituted a period of rapid advancement, with the early warning index achieving an average annual growth rate of 3.8%. In 2015, it first crossed into the ‘Critically Safe (Medium Alert)’ level, with the green indicator signifying that ecological safety had entered a controllable range. This progress stemmed from strengthened ecological constraints in the latter phase of the 12th Five-Year Plan and the 2015 release of the Overall Plan for Ecological Civilisation System Reform. This spurred county-level elimination of outdated production capacity and unleashed policy dividends. The period from 2017 to 2020 constituted a consolidation phase, during which the early warning index fluctuated within a narrow range of 50.31 to 53.35. By 2019, it had decisively advanced to the ‘Relatively Safe (Low Alert)’ level. This period coincided with the deepening of the ‘Pollution Prevention and Control Campaign’ outlined at the 19th CPC National Congress. Counties faced profound challenges such as greening traditional industries and addressing shortcomings in environmental infrastructure. Despite the impact of the 2020 pandemic, the warning index remained at the ‘Relatively Safe (Low Alert)’ level, reflecting the effectiveness of ecological resilience-building efforts. The years 2021–2023 constituted a period of steady optimisation, with the early warning index growing at an average annual rate of 1.07%. It remained consistently within the ‘Low Alert’ range while steadily approaching the ‘Safe (No Alert)’ threshold. During this time, the Carbon Peak Action Plan was implemented, and the pilot scheme for realising the value of ecological products was expanded. These measures propelled the county’s development model towards a profound transformation towards ‘low consumption and high efficiency’, placing land ecological security on a path of stable improvement.

3.1.3. Analysis of Subsystem Alert Situation

The comprehensive early warning index for land ecological security across the Top 100 Counties shows an upward trend, indicating a gradual improvement in ecological security levels. However, the evolutionary trajectories of the subsystem early warning indices exhibit significant divergence (Figure 5), presenting distinct characteristics: dual-core leadership from ‘driving-management’, fluctuating improvements in ‘pressures-impacts’, and low-amplitude oscillations in ‘status-responses’. During land ecological governance across different counties, variations in governance strategies, resource allocation, industrial structures and other factors have led to distinct characteristics and outcomes in each subsystem’s response to ecological security challenges. This reflects structural disparities in county-level land ecological governance.
Dual-Core Leadership of ‘driving-management’. The driving and management subsystems’ early warning indices leapt from the ‘Unsafe (Giant Alert)’ level in 2010 to the ‘Safe (No Alert)’ level by 2023. Both surpassed the upper threshold of ‘Critically Safe (Medium Alert)’ in 2019, reaching the ‘Relatively Safe (Low Alert)’ level, with average annual growth rates of 5.06% and 9.18%, respectively, demonstrating their prominent role in elevating land ecological safety levels. The continuously strengthening economic strength of the Top 100 Counties and significant improvement in social undertakings and public service levels collectively provide robust economic support and effective management safeguards for land ecological security construction.
Pressure-impact fluctuations show positive trends. The pressure and impact subsystem early warning indices rose from the ‘Relatively Unsafe (High Alert)’ level in 2010 to the ‘Relatively Safe (Low Alert)’ level by 2023, with annual growth rates of 1.61% and 1.47%, respectively. The indices exhibited repeated fluctuations around the upper threshold of the ‘Critically Safe (Medium Alert)’ level. On one hand, while energy conservation and emission reduction measures have alleviated some ecological pressures, root causes such as high population density and a large share of secondary industry remain prominent. On the other hand, the positive impact of ecosystems on the economy has begun to manifest, though the ecological effects of industrial restructuring remain complex. There is an urgent need to optimise industrial layout further and promote a virtuous cycle between ecology and the economy.
The ‘State-Response’ subsystem exhibits low-amplitude oscillations. The State and Response subsystems’ early warning indices show annual growth rates below 1%, demonstrating low-amplitude oscillations within the ‘Critically Safe (Medium Alert)’ level. The State subsystem’s index shifted from 49.73 in 2010 to 48.25 in 2023, reflecting overall stability with a slight decline. Indicators significantly influenced by human activities fluctuate, indicating ecosystem complexity. The early warning index for the response subsystem evolved from 50.06 in 2010 to 50.59 in 2023, remaining largely stable with a slight increase. The capacity to address land ecological security issues has gradually strengthened, with fiscal expenditure supporting ecological governance. However, response mechanisms require refinement to address risks more effectively.
The pronounced divergence in the evolutionary pathways of the land ecological security subsystem underscores the long-term and arduous nature of land ecological security development. Sustained efforts are required to enhance land ecological security levels steadily. Moving forward, the Top 100 Counties should continue strengthening the dual-core leadership of ‘driving and management’, persistently optimising pressure source control, promoting industrial upgrading and structural optimisation, accelerating improvements in the ecological status of the state and response subsystems, enhancing their stability and coordination, and achieving comprehensive enhancement and sustainable development of land ecological security.

3.2. Predictive Analysis of Land Ecological Security Alerts in the Top 100 Counties

3.2.1. Early Warning Trends for County Areas

To objectively evaluate the predictive performance of the CNN-LSTM model, this study first conducted a systematic benchmark comparison against ARIMA, SVR, and LSTM models. As shown in Table 3, the CNN-LSTM hybrid model achieved the optimal overall performance on the test set, exhibiting the highest R2 (0.9588) and lowest MAE (0.1347). The LSTM model (R2 = 0.8164, MAE = 0.2940) performed second best, confirming LSTM’s advantage in handling time series dependencies. However, its accuracy fell short of the hybrid model, demonstrating the contribution of the CNN module in extracting spatial features. Conversely, the predictive accuracy of traditional machine learning models, SVR (R2 = 0.6918, MAE = 0.7249) and classical time series model ARIMA (R2 = 0.5096, MAE = 0.5525) was comparatively low. This indicates that the land ecological security system’s complex, nonlinear and spatiotemporal coupling characteristics exceed the optimal processing capabilities of traditional models and shallow machine learning approaches. Consequently, all subsequent analyses were conducted using the CNN-LSTM model.
A CNN-LSTM hybrid neural network model generated dynamic forecasts for the land ecological security early warning indices of the Top 100 Counties from 2024 to 2028. Training model evaluation metrics indicate a high fit with an R2 value of 0.9779 and an average absolute error of 0.2849, demonstrating satisfactory fitting results. A traffic light mapping diagram illustrating land ecological security alerts from 2010 to 2028 was produced from a county-level perspective to visually represent the dynamic shifts in land ecological security across counties. County names are denoted by their Top 100 Counties ranking codes (Figure 6).
The mapping reveals two key trends: firstly, an overall positive trajectory, with the proportion of counties displaying red and orange warning lights steadily decreasing over time, while the share of cyan and blue lights progressively increases. Qian’an City and Yizheng City demonstrate the most significant improvement in land ecological security alerts, advancing from the ‘Critically Safe (Medium Alert)’ level in 2023 to the ‘Safe (No Alert)’ level by 2028. Fifteen counties, including Changsha, Yiwu, Zhuji, and Yueqing, remain relatively stable, maintaining the ‘Safe (No Alert)’ level over the next five years. This demonstrates that countries can translate short-term governance achievements into long-term stability through industrial restructuring, increased ecological investment, and robust governance mechanisms, ensuring sustainable land ecological improvement. Secondly, latent regression risks exist: over the next five years, 16 counties, including Gaitou City, Huian County, and Qingzhou City, show deteriorating alert trends. Sihui City and Tianmen City’s incidents declined from the ‘Safe (No Alert)’ level in 2023 to the ‘Critically Safe (Medium Alert)’ level by 2028. This demonstrates that land ecological security is not a one-way improvement; its fragility and reversibility necessitate establishing high-frequency monitoring and dynamic early-warning systems to prevent governance achievements from being reversed due to industrial fluctuations or policy relaxation.

3.2.2. Comprehensive Early Warning Trends

The fitted results are presented using a CNN-LSTM hybrid neural network prediction model to forecast the comprehensive ecological security early warning index for the Top 100 Counties from 2024 to 2028 (Figure 7). Based on current development trends, as concepts such as ‘green development’, ‘lucid waters and lush mountains are invaluable assets’, and ‘ecological priority’ gain widespread acceptance, alongside the deepening implementation of strategies like ‘sustainable development’, the ‘ecological compensation system’, and the ‘dual carbon goals’, the ecological security early warning index for these counties will exhibit a ‘gradual stabilisation and upward trend’. The composite early warning index will increase from 54.99 in 2023 to 58.80 in 2028, representing a 6.91% rise. Over the preceding four years, the land ecological security rating gradually approached the upper threshold of the ‘Relatively Safe (Low Alert)’ level (58.01), achieving a historic transition from ‘Relatively Safe (Low Alert)’ to ‘Safe (No Alert)’ by 2028. The warning indicator will shift to blue, reflecting the effectiveness of ecological security governance measures and marking a new phase in regional ecological management—shifting from risk prevention and control to maintaining a secure and stable state. However, the ecosystem remains vulnerable, with risks of land ecological degradation not yet eradicated, necessitating long-term consolidation of governance achievements. Efforts should leverage the national ‘dual carbon’ strategy and ecological compensation mechanisms, anchoring the goal at the ‘Safe (No Alert)’ level. This will amplify the policy dividends of county-level green transformation, guiding land use from intensive development towards high-quality utilisation, and steadily advancing into a new phase of virtuous economic–land–ecological cycles.

3.2.3. Subsystem Early Warning Trends

Employing a CNN-LSTM hybrid neural network predictive model, this study calculated the early warning indices for the land ecological security subsystem of the Top 100 Counties from 2024 to 2028. Based on the model training evaluation metrics, the R2 values for the goodness-of-fit across the DPSIRM subsystems were 0.9954, 0.9860, 0.9951, 0.9907, 0.9847, 0.9970, with mean absolute errors of 0.3283, 0.2907, 0.1148, 0.2233, 0.1072, and 0.3444, respectively. The fitting results were satisfactory. Radar charts and line graphs of the predicted early warning indices for the land ecological security subsystem were plotted to examine the evolution trends of alert conditions. As illustrated in Figure 8, the six subsystems will exhibit a polarised pattern of ‘three rising, two stagnant, one declining’ from 2024 to 2028. Specifically, the driving, management, and pressure subsystems will continuously optimise, while the response and impact subsystems will experience fluctuating stagnation. The State subsystem, however, will follow a declining trajectory.
The high growth in the driving subsystem stems primarily from accelerated investment in green industries. The surge in the management subsystem originates from the comprehensive deployment of digital governance platforms. In contrast, the improvement in the pressure subsystem benefits from the phasing out of outdated production capacity. The closed-loop formation of economic traction, institutional innovation, and emission reduction measures constitutes the fundamental driver behind the sustained strengthening of these three subsystems. The response subsystem’s plateau phase exposed bottlenecks in fiscal expenditure efficiency, while the impact subsystem’s stagnation stemmed from the inefficient conversion of ecological dividends into economic benefits. Concurrently, governance funding was characterised by ‘excessive expenditure and inefficient allocation’, coupled with ecological industries marked by ‘substantial potential but slow realisation’, collectively causing dual stagnation. The core issue in the state subsystem’s decline lies in the lagging recovery of ecological foundations, with fundamental ecological capital suffering from ‘insufficient stock and constrained growth’. The decisive factor for land ecological security in the top 100 counties over the next five years lies in whether the institutional dividends from the ‘three-rise’ subsystem can be extended to the ‘two-lag’ subsystem, thereby enhancing governance efficiency and reversing the downward trajectory of the state subsystem through long-term ecological restoration projects. Otherwise, structural vulnerabilities may resurface under future shocks even if the composite index crosses the ‘Safe (No Alert)’ threshold.

4. Conclusions and Discussions

4.1. Conclusions

This study established an early warning indicator system for land ecological security based on the DPSIRM framework, conducting a scientific diagnosis of land ecological security across the Top 100 Counties. A CNN-LSTM hybrid neural network model was employed to analyse dynamic early warning systematically. Key findings are as follows.
(1)
From 2010 to 2023, the land ecological security warning levels across all counties in the Top 100 Counties improved progressively. The proportion of counties achieving ‘Relatively Safe (Low Alert)’ or higher levels increased from 2% in 2010 to 68% in 2023. The ‘Critically Safe (Medium Alert)’ proportion rose from 6% to 40% before stabilising at around 30%. Conversely, the proportion below ‘Critically Safe (Medium Alert)’ declined rapidly from 92% in 2010 to 7% in 2023. The highest proportion of counties (21%) transitioned from ‘Relatively Unsafe (High Alert)’ to ‘Relatively Safe (Low Alert)’ from 2010 to 2023. Moreover, based on the magnitude of alert level changes, counties classified as ‘significantly improved’ (those advancing by at least three levels) constituted the largest share (32%). Counties classified as ‘relatively lagging’—where the alert level never exceeded ‘Critically Safe (Medium Alert)’—ranked second (22%). Counties categorised as ‘consistently leading’—maintaining an alert level no lower than ‘Critically Safe (Medium Alert)’ while predominantly achieving ‘Safe (No Alert)’ in recent years—represented the smallest proportion (14%). This indicates that most counties have made significant progress in land ecological security, yet some still face considerable ecological pressures requiring enhanced ecological governance and protection measures.
(2)
From 2010 to 2023, the overall comprehensive early warning index for land ecological security among the Top 100 Counties exhibited a ‘stepwise leap’ trend, achieving a transformative progression from the ‘Relatively Unsafe (High Alert)’ to the ‘Relatively Safe (Low Alert)’ level. This evolution unfolded in four phases: ‘fluctuation and adjustment—rapid improvement—plateau consolidation—steady-state optimisation’. This progression indicates that, driven by policies such as ‘ecological civilisation development’, the ‘dual carbon strategy’, and ‘high-quality county-level development’, the Top 100 Counties have significantly enhanced regional ecological security through greening industrial structures, intensifying resource utilisation, and institutionalising ecological governance.
(3)
From 2010 to 2023, the six significant subsystems’ land ecological security in the Top 100 Counties diverged markedly in their evolutionary trajectories. Dual-core drivers of ‘driving-management’ propelled the warning level from ‘Unsafe (Giant Alert)’ to ‘Safe (No Alert)’, becoming the core engine for ecological security enhancement. This advancement primarily stemmed from synchronous growth in economic scale and digital governance investment. Pressure-impact fluctuations showed positive trends, with the alert level approaching ‘Relatively Safe (Low Alert)’ status. However, structural constraints such as high population density, heavy industry dominance, and inefficient ecological–economic conversion continued to exert limitations. The ‘state-response’ warning level fluctuates within the ‘Critically Safe (Medium Alert)’ range, reflecting coexisting ecosystem fragility and governance response lag. Fiscal investment efficiency and institutional refinement require urgent enhancement.
(4)
Over the next five years, the projected comprehensive land ecological safety early warning index for the Top 100 Counties exhibits a ‘gradual stabilisation and upward trend’, rising from 54.99 in 2023 to 58.80 by 2028, marking a historic transition from ‘Relatively Safe (Low Alert)’ to ‘Safe (No Alert)’ status by 2028. The projected values of the subsystem early warning indices exhibit a polarised pattern of ‘three increases, two stagnations, and one decline’. The sustained optimisation of the driving, management, and pressure subsystems stems from green investment, digital governance, and capacity optimisation. The response and impact subsystems remain stagnant and volatile, exposing fiscal inefficiency and slow conversion of ecological dividends. The state subsystem shows a declining trend due to the lag in ecological baseline recovery. The key to achieving land ecological security development in the Top 100 Counties lies in channelling the spillover benefits from the three improving subsystems to the two stagnant ones while reversing the inertia of the state subsystem through ecological restoration.

4.2. Discussions

This study reveals that the overall ecological security of land in China’s top 100 counties significantly improved from 2010 to 2023, aligning with findings from numerous regional-scale studies in China [45,46,47]. This consistency confirms that ecological governance across different administrative tiers in China has achieved notable results under the strong impetus of the national ecological civilisation strategy.
It should be noted that the value of this study lies particularly in its focus on the unique characteristics of the Top 100 Counties as economically dense regions, revealing a distinctive dual-core model driven by both ‘driving’ and ‘management’. The research not only observed an increase in the composite index but also, through subsystem analysis, identified the ‘management’ subsystem and the ‘driving’ subsystem as jointly forming the dual-core driving development. Firstly, it validates the necessity and scientific merit of incorporating the ‘management’ dimension into research frameworks, addressing the shortcomings of traditional frameworks in capturing the agency of ‘governance’. Secondly, it empirically demonstrates that in economically developed countries, robust economic strength (driving) can become the primary engine for advancing ecological security improvements when effectively channelled into refined environmental governance (management), rather than functioning solely as a pressure source as traditionally perceived. This finding deepens discussions on the complex relationship between socio-economic factors and ecological security within regional ecological security assessment research [48], providing theoretical support for a win–win economic growth and environmental protection model in economically developed countries. Thirdly, it confirms the necessity of integrating ecological civilisation construction into the overarching design of high-quality economic development.
Furthermore, this study possesses advantages in methodological integration. The employed DPSIRM-CNN-LSTM integrated model demonstrates significant strengths in diagnosis and early warning. Compared to traditional time series models and shallow machine learning algorithms, the CNN-LSTM hybrid model exhibits higher accuracy when processing the complex panel data in this study. This finding corroborates conclusions affirming its superiority when applying similar models to landscape diversity risk prediction [49]. Integrating DPSIRM and CNN-LSTM represents not a mere accumulation of models, but a synergistic combination of a ‘systematic diagnostic framework’ and a ‘high-performance spatio-temporal forecasting tool’. The former ensures the logical completeness of the early warning indicator system, while the latter guarantees the ability to extract reliable future signals from complex data.
Finally, the subsystem’s polarised pattern of ‘three increases, two stagnations, and one decline’ offers precise regulatory levers for future county land use decisions. This pattern indicates: Firstly, the ‘three increases’ (driving, management, pressure) represent areas where policy dividends are evident and should be further consolidated. For instance, green financial instruments can strengthen driving forces, digital governance platforms can optimise management, and reforms like the ‘per-mu productivity benchmark’ can continuously alleviate pressure. Second, the ‘two stagnations’ (response, impact) reveal current governance bottlenecks. Inefficient fiscal ecological expenditure (response) indicates a need to shift funding from ‘extensive irrigation’ to ‘precision drip irrigation’. In contrast, the poor conversion of ecological benefits into economic gains (impact) necessitates innovation in ‘ecological product value realisation mechanisms’ within land use planning. This includes explicitly demarcating and cultivating ‘ecological industrialisation’ zones in territorial spatial planning. Thirdly, the ‘decline’ (state) signals the most severe warning. Despite effective end-point governance, the recovery of ecosystem fundamentals (such as soil, water bodies, and biodiversity) remains lagging, which may be linked to the persistent encroachment of urban construction land on ecological spaces. Future land use decisions must shift from an ‘incremental expansion’ mindset towards ‘stock optimisation’ and ‘flow management’. The decline in the ecological baseline can be fundamentally curbed by delineating and strictly enforcing ecological conservation red lines and implementing integrated land remediation models such as ‘ecological restoration plus’. The dynamic early-warning model developed in this study can be employed to assess the potential impacts of major land use planning adjustments on subsystem structures, enabling intelligent governance through the ‘assessment before decision-making’ principle.

5. Policy Recommendations and Research Prospects

5.1. Policy Recommendations

Based on the above conclusions and discussions, this study proposes a three-tiered progressive policy recommendation framework to achieve sustained improvement and systemic stability in land ecological security levels for the Top 100 Counties and comparable county-level regions.
First, governments should formulate regionally differentiated ecological security strategies and implement a tiered ‘red–green–blue’ response mechanism. Red-light counties (unsafe-level) should initiate emergency interventions through land ecological security rescue plans, with relevant experts stationed for focused supervision and to formulate tailored transformation strategies (‘one county, one policy’). Green-light counties (critically safe-level) should engage in land ecological security compliance competitions to ensure steady improvement. Blue-light counties (safe-level) should establish land ecological security demonstration zones to consolidate benchmark leadership.
Second, the development of the land ecological security subsystem should be optimised through targeted improvement measures. Addressing the subsystem’s projected ‘three increases, two stagnations, one decline’ trend for 2024–2028: Firstly, leverage momentum to strengthen the ‘three increases’ subsystem, consolidating the foundation for high-quality development. Regarding driving, deepen county-level investment in green technological innovation, transforming economic driver advantages into ecological capital. Regarding management, advanced intelligent reforms are being implemented by establishing a ‘digital platform for ecological governance’. Regarding pressure, implement a ‘negative list for industrial ecologisation’ and innovate intensive land use models. Secondly, the ‘two stagnations’ subsystem must be broken to clear governance bottlenecks. In response mechanisms, establish an ‘ecological security emergency bank’ to address environmental emergencies; regarding impact, identify key industrial pollution nodes for targeted remediation and create ‘ecology-industry’ demonstration zones. Thirdly, urgent intervention should be made in the ‘decline’ subsystem to halt land ecological degradation. Prioritise state subsystem transitioning from low-amplitude oscillation to decline, implement red-and-yellow card warnings for ecologically sensitive areas, and establish an innovative ecological diversity protection network for land.
Third, enhance dynamic early warning and long-term planning for land ecological security to ensure sustained ecological conservation and improvement measures. The comprehensive early warning index for land ecological security is projected to exhibit a ‘gradual stabilisation and upward trend’ in the future. Based on this, the strategy should anchor the 2028 ‘no alert transition’ objective and implement a three-phase plan: the foundation phase (2024–2025) will establish a special breakthrough initiative for land ecological security; the challenge phase (2026–2027) will launch the ‘Blue Light’ sprint initiative, and the leap phase (2028) will establish a stable protective network. Simultaneously, deepen the application of early warning models to create a smart governance hub. It must be clarified that early warning is not about prediction but identifying intervention windows, enabling the land ecological security levels of the Top 100 Counties to transition from gradual improvement to systematic stability.

5.2. Research Prospects

This study provides a systematic solution for diagnosing and dynamically monitoring land ecological security in China’s top 100 counties by integrating the DPSIRM framework with a hybrid CNN-LSTM neural network model. Nevertheless, certain limitations remain that warrant further refinement in future research. Firstly, while the constructed indicator system is relatively comprehensive, it inadequately captures ‘soft’ indicators within the social sustainability dimension, such as community participation levels and public perceptions of ecological wellbeing. This limitation primarily stems from the availability and consistency of social data at the county level. Future research may explore integrating multi-source information data to incorporate social sustainability assessments into land ecological security evaluations. Secondly, while the employed model excels at revealing statistical correlations between variables, it falls short in elucidating precise underlying causal mechanisms. For instance, although we identified the core role of the ‘management’ subsystem, determining which specific policy instruments prove most effective and their transmission pathways requires deeper analysis through qualitative methods such as case studies and field investigations.
To address these limitations, future research may explore the following directions: (1) developing integrated evaluation frameworks incorporating social dimensions to reflect the three pillars of sustainable development more comprehensively; (2) advancing the integration of mechanistic and data-driven models to enhance causal inference and scenario simulation capabilities.

Author Contributions

Conceptualization, Y.C. and Y.W.; methodology, F.X.; software, Y.C.; validation, F.X. and Y.C.; formal analysis, F.X.; data curation, F.X.; writing—original draft preparation, F.X. and Y.C.; writing—review and editing, F.X. and Y.W.; visualization, Y.C.; supervision, F.X.; project administration, Y.W.; funding acquisition, Y.W. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by Special Project for Cultivating Leading Talents in Philosophy and Social Sciences of Zhejiang Province (23QNYC15ZD).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The original contributions presented in this study are included in the article. Further inquiries can be directed to the corresponding author.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Shaker, R.R.; Mackay, B.R. Hidden Patterns of Sustainable Development in Asia with Underlying Global Change Correlations. Ecol. Indic. 2021, 131, 108227. [Google Scholar] [CrossRef]
  2. Li, L.; Fu, M.; Zhu, Y.; Kang, H.; Wen, H. The Current Situation and Trend of Land Ecological Security Evaluation from the Perspective of Global Change. Ecol. Indic. 2024, 167, 112608. [Google Scholar] [CrossRef]
  3. Prăvălie, R. Exploring the Multiple Land Degradation Pathways across the Planet. Earth-Sci. Rev. 2021, 220, 103689. [Google Scholar] [CrossRef]
  4. Maestre, F.T.; Guirado, E.; Armenteras, D.; Beck, H.E.; AlShalan, M.S.; Al-Saud, N.T.; Chami, R.; Fu, B.; Gichenje, H.; Huber-Sannwald, E.; et al. Bending the Curve of Land Degradation to Achieve Global Environmental Goals. Nature 2025, 644, 347–355. [Google Scholar] [CrossRef]
  5. Fund, S. Sustainable Development Goals. 2015. Available online: https://www.un.org/sustainabledevelopment/inequality (accessed on 5 August 2025).
  6. Carnohan, S.A.; Trier, X.; Liu, S.; Clausen, L.P.W.; Clifford-Holmes, J.K.; Hansen, S.F.; Benini, L.; McKnight, U.S. Next Generation Application of DPSIR for Sustainable Policy Implementation. Curr. Res. Environ. Sustain. 2023, 5, 100201. [Google Scholar] [CrossRef]
  7. Li, G.; Fang, C.; Li, Y.; Wang, Z.; Sun, S.; He, S.; Qi, W.; Bao, C.; Ma, H.; Fan, Y.; et al. Global Impacts of Future Urban Expansion on Terrestrial Vertebrate Diversity. Nat. Commun. 2022, 13, 1628. [Google Scholar] [CrossRef]
  8. Gayen, J.; Datta, D. Application of Pressure–State–Response Approach for Developing Criteria and Indicators of Ecological Health Assessment of Wetlands: A Multi-Temporal Study in Ichhamati Floodplains, India. Ecol. Process. 2023, 12, 1–20. [Google Scholar] [CrossRef]
  9. Sadeghi, S.H.; Vafakhah, M.; Moosavi, V.; Pourfallah Asadabadi, S.; Sadeghi, P.S.; Khaledi Darvishan, A.; Bagheri Fahraji, R.; Mosavinia, S.H.; Majidnia, A.; Gharemahmudli, S.; et al. Assessing the Health and Ecological Security of a Human Induced Watershed in Central Iran. Ecosyst. Health Sustain. 2022, 8, 2090447. [Google Scholar] [CrossRef]
  10. Li, R.; Cui, W.; Si, L.; Bao, S.; Yang, F. Ecological Security Early Warning and Obstacle Factor Analysis in the Zhejiang Greater Bay Area. Ecol. Environ. 2023, 32, 1861. [Google Scholar] [CrossRef]
  11. Zhao, L.; Liu, G.; Xian, C.; Nie, J.; Xiao, Y.; Zhou, Z.; Li, X.; Wang, H. Simulation of Land Use Pattern Based on Land Ecological Security: A Case Study of Guangzhou, China. Int. J. Environ. Res. Public Health 2022, 19, 9281. [Google Scholar] [CrossRef]
  12. Dong, J.; Lyu, Y. Appraisal of Urban Land Ecological Security and Analysis of Influencing Factors: A Case Study of Hefei City, China. Env. Sci. Pollut. Res. 2022, 29, 90803–90819. [Google Scholar] [CrossRef] [PubMed]
  13. Liu, J.; Cao, X.; Zhao, L.; Dong, G.; Jia, K. Spatiotemporal Differentiation of Land Ecological Security and Its Influencing Factors: A Case Study in Jinan, Shandong Province, China. Front. Environ. Sci. 2022, 10, 824254. [Google Scholar] [CrossRef]
  14. Gong, D.; Huang, M.; Lin, H. Construction of an Ecological Security Pattern in Rapidly Urbanizing Areas Based on Ecosystem Sustainability, Stability, and Integrity. Remote Sens. 2023, 15, 5728. [Google Scholar] [CrossRef]
  15. Zhang, L.; Peng, W.; Zhang, J. Assessment of Land Ecological Security from 2000 to 2020 in the Chengdu Plain Region of China. Land 2023, 12, 1448. [Google Scholar] [CrossRef]
  16. Zhang, X.; Huang, X.; Li, G.; Lu, Y.; Ren, D.; Zhang, S. Evaluating the Land Ecological Security of Urban Areas in the Yangtze River Basin in Hubei. Env. Dev. Sustain. 2025. [Google Scholar] [CrossRef]
  17. Zhang, X.; Wang, Y.; Yuan, X.; Yang, Y. Regional Land Ecological Security Evaluation and Ecological Poverty Alleviation Practice: A Case Study of Yangxian County in Shaanxi Province, China. J. Geogr. Sci. 2022, 32, 682–700. [Google Scholar] [CrossRef]
  18. Miao, H.; Zhou, H. Evaluation of County-Level Economic Efficiency and Its Spatiotemporal Differentiation in Hohhot-Baotou-Ordos-Yulin Urban Agglomeration in China. Humanit. Soc. Sci. Commun. 2023, 10, 99. [Google Scholar] [CrossRef]
  19. Teng, G.; Chen, L.; Zhang, T.; Li, L.; Xiao, J.; Ma, L. The Spatiotemporal Evolution of Land Use Ecological Efficiency in the Huaihai Economic Zone: Insights from a Multi-Dimensional Framework and Geospatial Modeling. Land 2025, 14, 883. [Google Scholar] [CrossRef]
  20. Chen, Z.; Chen, H.; Yang, M.; Wang, X.; Jiang, Y.; Zhang, W. Heterogeneity and Optimization of Ecological Security Pattern on a Mountain Town: A Case of Tianzhu County in the Hexi Corridor, China. Front. Environ. Sci. 2023, 10, 1106379. [Google Scholar] [CrossRef]
  21. Liu, H.; Wang, Z.; Zhang, L.; Tang, F.; Wang, G.; Li, M. Construction of an Ecological Security Network in the Fenhe River Basin and Its Temporal and Spatial Evolution Characteristics. J. Clean. Prod. 2023, 417, 137961. [Google Scholar] [CrossRef]
  22. Shan, Y.; Wei, S.; Yuan, W.; Miao, Y. Evaluation and Prediction of Land Ecological Security in Shenzhen Based on DPSIR-TOPSIS-GM(1,1) Model. PLoS ONE 2022, 17, e0265810. [Google Scholar] [CrossRef] [PubMed]
  23. Li, F.; Lu, S.; Sun, Y.; Li, X.; Xi, B.; Liu, W. Integrated Evaluation and Scenario Simulation for Forest Ecological Security of Beijing Based on System Dynamics Model. Sustainability 2015, 7, 13631–13659. [Google Scholar] [CrossRef]
  24. Bury, T.M.; Sujith, R.I.; Pavithran, I.; Scheffer, M.; Lenton, T.M.; Anand, M.; Bauch, C.T. Deep Learning for Early Warning Signals of Tipping Points. Proc. Natl. Acad. Sci. USA 2021, 118, e2106140118. [Google Scholar] [CrossRef] [PubMed]
  25. Peng, W.; Sun, Y.; Liu, C.; Liu, D. Study on Urban Land Ecological Security Pattern and Obstacle Factors in the Beijing–Tianjin–Hebei Region. Sustainability 2023, 15, 43. [Google Scholar] [CrossRef]
  26. Zhang, G.; Zhang, J.; Zhou, G. Ecological Security Level of Resource-Based Cities in the Yellow River Basin and Trend of Change. Resour. Sci. 2023, 45, 762–775. [Google Scholar] [CrossRef]
  27. Meng, L.; Yang, R.; Sun, M.; Zhang, L.; Li, X. Regional Sustainable Strategy Based on the Coordination of Ecological Security and Economic Development in Yunnan Province, China. Sustainability 2023, 15, 7540. [Google Scholar] [CrossRef]
  28. Fan, Z.; Liu, B.; Zhong, P. Spatial-Temporal Evolution of Land Ecological Health and Influencing Factors in China’s Ecological Civilization Pilot Area. J. Environ. Eng. Landsc. Manag. 2025, 33, 259–272. [Google Scholar] [CrossRef]
  29. Zhou, J.-R.; Li, X.-Q.; Yu, X.; Zhao, T.-C.; Ruan, W.-X. Exploring the Ecological Security Evaluation of Water Resources in the Yangtze River Basin under the Background of Ecological Sustainable Development. Sci. Rep. 2024, 14, 15475. [Google Scholar] [CrossRef]
  30. Liu, L.; Zhang, S.; Liu, W.; Qu, H.; Guo, L. Spatiotemporal Changes and Simulation Prediction of Ecological Security Pattern on the Qinghai–Tibet Plateau Based on Deep Learning. Land 2024, 13, 1073. [Google Scholar] [CrossRef]
  31. Fan, Y.; Zheng, Y.; Jia, C.; Song, Y. The Impact of China’s Urbanization on Ecosystem Service Value from the Perspective of Gross Ecosystem Product: A Case Study of Beijing-Tianjin-Hebei Region. Sci. Rep. 2024, 14, 15954. [Google Scholar] [CrossRef]
  32. Du, W.; Liao, X.; Tong, Z.; Rina, S.; Rong, G.; Zhang, J.; Liu, X.; Guo, E. Early Warning and Scenario Simulation of Ecological Security Based on DPSIRM Model and Bayesian Network: A Case Study of East Liaohe River in Jilin Province, China. J. Clean. Prod. 2023, 398, 136649. [Google Scholar] [CrossRef]
  33. Sun, D.; Wu, J.; Zhang, F.; Su, W.; Hui, H. Evaluating Water Resource Security in Karst Areas Using DPSIRM Modeling, Gray Correlation, and Matter–Element Analysis. Sustainability 2018, 10, 3934. [Google Scholar] [CrossRef]
  34. Wan, B.-T.; Bao, X.-Y.; Zhao, J.-C.; Li, A.-C. Establishment and Application of Performance Evaluation and Obstacle Diagnosis Model for Regional Water Ecological Civilization Construction. Huan Jing Ke Xue 2021, 42, 2089–2100. [Google Scholar] [CrossRef] [PubMed]
  35. Peng, G.; Zhang, F.; Yang, X.; Ma, D.; Tan, H. Spatial and Temporal Evolution and Drive Mechanism of Tourism Ecological Security Based on DPSIRM-SBM Model. Environ. Res. Commun. 2024, 6, 085021. [Google Scholar] [CrossRef]
  36. Zadeh, S.; Ghadiri, M.; Mirrokni, V.; Zadimoghaddam, M. Scalable Feature Selection via Distributed Diversity Maximization. In Proceedings of the AAAI Conference on Artificial Intelligence, San Francisco, CA, USA, 11 February 2017; Volume 31. [Google Scholar] [CrossRef]
  37. Wang, L.; Yi, P.; Li, W.; Dong, Q. Smart Cities Dynamic Performance Assessment in China Using Vertical and Horizontal Deviation Maximization Method and Driving Indicator Analysis. Env. Dev. Sustain. 2024, 26, 11847–11880. [Google Scholar] [CrossRef]
  38. Sun, M.; Li, X.; Yang, R.; Zhang, Y.; Zhang, L.; Song, Z.; Liu, Q.; Zhao, D. Comprehensive Partitions and Different Strategies Based on Ecological Security and Economic Development in Guizhou Province, China. J. Clean. Prod. 2020, 274, 122794. [Google Scholar] [CrossRef]
  39. Jing, X.; Tao, S.; Hu, H.; Sun, M.; Wang, M. Spatio-Temporal Evaluation of Ecological Security of Cultivated Land in China Based on DPSIR-Entropy Weight TOPSIS Model and Analysis of Obstacle Factors. Ecol. Indic. 2024, 166, 112579. [Google Scholar] [CrossRef]
  40. Xi, W. Measurement and Influencing Factors of the Construction Level of Chinese Path to A Modernized Industrial System. Sci. Technol. Prog. Policy 2024, 41, 54–65. [Google Scholar] [CrossRef]
  41. Zhu, Q.; Cai, Y. Assessment Framework and Simulation Prediction for Ecological Security Based on “Risk-Health-Services”: A Case Study of the Huaihe River Basin. J. Nat. Resour. 2024, 39, 2676. [Google Scholar] [CrossRef]
  42. Yang, H.; Zou, R.; Hu, Y.; Wang, L.; Xie, Y.; Tan, Z.; Zhu, Z.; Zhu, A.-X.; Gong, J.; Mao, X. Sustainable Utilization of Cultivated Land Resources Based on “Element Coupling-Function Synergy” Analytical Framework: A Case Study of Guangdong, China. Land Use Policy 2024, 146, 107316. [Google Scholar] [CrossRef]
  43. Li, S.; Zhang, C.; Chen, C.; Yang, C.; Zhao, L.; Bai, X. Optimization Simulation and Comprehensive Evaluation Coupled with CNN-LSTM and PLUS for Multi-Scenario Land Use in Cultivated Land Reserve Resource Area. Remote Sens. 2025, 17, 1619. [Google Scholar] [CrossRef]
  44. Halbouni, A.; Gunawan, T.S.; Habaebi, M.H.; Halbouni, M.; Kartiwi, M.; Ahmad, R. CNN-LSTM: Hybrid Deep Neural Network for Network Intrusion Detection System. IEEE Access 2022, 10, 99837–99849. [Google Scholar] [CrossRef]
  45. Zhai, G.; Ren, P.; Zhang, R.; Wang, B.; Zhang, M.; He, T.; Zhang, J. Evaluation of Land Ecological Security and Driving Factors in the Lower Yellow River Flood Plain Based on Quality, Structure and Function. Sci. Rep. 2025, 15, 2674. [Google Scholar] [CrossRef] [PubMed]
  46. Lee, C.-C.; He, Z.-W.; Luo, H.-P. Spatio-Temporal Characteristics of Land Ecological Security and Analysis of Influencing Factors in Cities of Major Grain-Producing Regions of China. Environ. Impact Assess. Rev. 2024, 104, 107344. [Google Scholar] [CrossRef]
  47. Wang, N.; Li, S.; Kang, Q.; Wang, Y. Exploring the Land Ecological Security and Its Spatio-Temporal Changes in the Yangtze River Economic Belt of China, 2000–2020. Ecol. Indic. 2023, 154, 110645. [Google Scholar] [CrossRef]
  48. Xu, Y.; Yang, L.; Sun, K.; Zhu, J. Synergistic Security Relationships and Risk Measurement of Water Resources-Social Economy-Ecological Environment in Beijing-Tianjin-Hebei Region. Ecol. Indic. 2025, 175, 113512. [Google Scholar] [CrossRef]
  49. Yang, Y.; Wang, H.; Li, X.; Qu, T.; Su, J.; Luo, D.; He, Y. Assessment of Landscape Diversity in Inner Mongolia and Risk Prediction Using CNN-LSTM Model. Ecol. Indic. 2024, 169, 112940. [Google Scholar] [CrossRef]
Figure 1. Distribution of the Top 100 Counties.
Figure 1. Distribution of the Top 100 Counties.
Sustainability 17 09271 g001
Figure 2. Theoretical Framework for Land Ecological Security Systems Based on the DPSIRM Model.
Figure 2. Theoretical Framework for Land Ecological Security Systems Based on the DPSIRM Model.
Sustainability 17 09271 g002
Figure 3. (a) Proportion of early warning levels by year for Top 100 Counties; (b) String diagram of warning level transitions for Top 100 Counties, 2010–2023.
Figure 3. (a) Proportion of early warning levels by year for Top 100 Counties; (b) String diagram of warning level transitions for Top 100 Counties, 2010–2023.
Sustainability 17 09271 g003
Figure 4. Comprehensive Early Warning Index of Land Ecological Security in the Top 100 Counties 2010–2023.
Figure 4. Comprehensive Early Warning Index of Land Ecological Security in the Top 100 Counties 2010–2023.
Sustainability 17 09271 g004
Figure 5. Stacked chart of early warning index of land ecological security subsystem in the Top 100 Counties, 2010–2023.
Figure 5. Stacked chart of early warning index of land ecological security subsystem in the Top 100 Counties, 2010–2023.
Sustainability 17 09271 g005
Figure 6. Land ecological Security Alert Indicator Mapping for the Top 100 Counties, 2010–2028.
Figure 6. Land ecological Security Alert Indicator Mapping for the Top 100 Counties, 2010–2028.
Sustainability 17 09271 g006
Figure 7. Comprehensive Early Warning Index for Land Ecological Security Forecast Map.
Figure 7. Comprehensive Early Warning Index for Land Ecological Security Forecast Map.
Sustainability 17 09271 g007
Figure 8. (a) Radar chart for predicting the early warning index of the land ecological security subsystem; (b) Line chart for predicting the early warning index of the land ecological security subsystem.
Figure 8. (a) Radar chart for predicting the early warning index of the land ecological security subsystem; (b) Line chart for predicting the early warning index of the land ecological security subsystem.
Sustainability 17 09271 g008
Table 1. County-Level Land Ecological Security Early Warning Indicator System.
Table 1. County-Level Land Ecological Security Early Warning Indicator System.
TypeIndicatorUnitAttributeWeights
Drivinggross regional productten thousand yuan+0.0387
Distance to higher administrative centreskm0.0377
Total population at the end of the yearten thousand people0.0399
Disposable income per capitayuan+0.0411
PressurePopulation densityperson per square kilometre0.0402
Share of secondary sector in GDP%0.0418
Industrial emissions of SO2ton0.0417
Surface PM2.5 mass concentrationμg/m30.0468
StateAnnual precipitationmm+0.0480
Average annual sunshine hoursh+0.0443
Water areahectares+0.0463
Cultivated land areahectares0.0480
ImpactNight light datanWcm-7sr-10.0481
Savings balance for urban and rural residentsten thousand yuan+0.0491
Growth rate of retail sales of consumer goods%+0.0479
Value added of the tertiary sector as a percentage of GDP%+0.0475
ResponseGeneral budget expenditures of local financesten thousand yuan+0.0474
Number of persons employed
in agriculture, forestry and fisheries
person+0.0419
Impervious surface areahectares0.0443
Total industrial output value above scaleten thousand yuan0.0403
ManagementTotal investment in fixed assetsten thousand yuan+0.0402
Total public library collectionsthousand volumes+0.0394
Number of beds in hospitals and health centresbed+0.0395
Table 2. Classification Criteria for Land Ecological Safety Alert Levels in the Top 100 Counties.
Table 2. Classification Criteria for Land Ecological Safety Alert Levels in the Top 100 Counties.
Early Warning IndexWarning LevelVigilanceIndicator Lamp
(0.00, 42.90)UnsafeGiant AlertRed Light
[42.90, 48.03)Relatively UnsafeHigh AlertOrange Light
[48.03, 52.67)Critically SafeMedium AlertGreen Light
[52.67, 58.01)Relatively SafeLow AlertCyan Light
[58.01, 100.00)SafeNo AlertBlue Light
Table 3. Performance comparison of different forecasting models.
Table 3. Performance comparison of different forecasting models.
Model NameCoefficient of DeterminationMean Absolute Error
ARIMA0.50960.5525
SVR0.69180.7249
LSTM0.81640.2940
CNN-LSTM0.95880.1347
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Xu, F.; Cui, Y.; Weng, Y. Research on Land Ecological Security Diagnosis and Dynamic Early Warning for China’s Top 100 Counties. Sustainability 2025, 17, 9271. https://doi.org/10.3390/su17209271

AMA Style

Xu F, Cui Y, Weng Y. Research on Land Ecological Security Diagnosis and Dynamic Early Warning for China’s Top 100 Counties. Sustainability. 2025; 17(20):9271. https://doi.org/10.3390/su17209271

Chicago/Turabian Style

Xu, Fei, Yalun Cui, and Yijing Weng. 2025. "Research on Land Ecological Security Diagnosis and Dynamic Early Warning for China’s Top 100 Counties" Sustainability 17, no. 20: 9271. https://doi.org/10.3390/su17209271

APA Style

Xu, F., Cui, Y., & Weng, Y. (2025). Research on Land Ecological Security Diagnosis and Dynamic Early Warning for China’s Top 100 Counties. Sustainability, 17(20), 9271. https://doi.org/10.3390/su17209271

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop