Bio-Inspired Geocomputation for Cross-Scale Ecological Security Patterns in Urban Agglomerations: An Integrated Framework from Data Fusion to Network Optimization

Abstract

Constructing resilient Ecological Security Patterns (ESPs) in polycentric urban agglomerations is computationally challenging due to persistent scale mismatches between local planning and regional strategies. To address this, we developed a novel Proactive Integration Mechanism (PIM), a computational framework that dynamically optimizes ESPs by algorithmically fusing multi-source geospatial data. The PIM integrates three innovative components: (1) a Function–Structure–Policy data fusion approach that couples Self-Organizing Map clustering of ecosystem services with Morphological Spatial Pattern Analysis and policy data to identify ecological sources; (2) a Dual-Feedback Mechanism that hybridizes circuit theory with an Improved Ant Colony Optimization algorithm for dynamic corridor delineation; and (3) complex network analysis to derive targeted interventions from topological properties. Applied to a node city of the Chengdu-Chongqing Economic Circle, the PIM identified 22 integrated ecological sources and 37 corridors. The optimized network showed enhanced resilience: a deterministic 20.5% increase in circuit redundancy (α-index) and an 8.6% improvement in overall connectivity (γ-index), achieved through minimal topological modifications. Temporal validation (2000–2020) confirmed the high stability of the identified patterns. This study provides a potentially replicable and computationally robust framework that bridges spatial ecology with optimization algorithms, offering a promising paradigm for constructing ESPs in node cities within subtropical urban agglomerations.

1. Introduction

The growing complexity of polycentric urban agglomerations presents a fundamental challenge in geocomputation: accurately simulating multi-scale ecological processes across administratively fragmented and spatially heterogeneous landscapes [1,2,3,4]. This creates a critical scale mismatch, where localized models fail to align with macro-regional ecological strategies, undermining the effectiveness of spatial conservation planning [5,6]. Within these complex networks, node cities—situated at key ecological interfaces—are hypothesized as crucial hubs for sustaining cross-scale connectivity [7,8]. However, a robust computational framework capable of quantifying their strategic role and algorithmically leveraging it to enhance ecological resilience remains underdeveloped, highlighting a critical gap at the intersection of spatial optimization, complex network theory, and sustainable urban science [9,10,11]. Specifically, this gap manifests in three concrete computational challenges: (1) dynamic modeling of adaptive species behaviors across scales, (2) feedback mechanisms linking biophysical simulation with policy implementation, and (3) integrated frameworks that operationally leverage node cities’ strategic positions to bridge local planning with regional strategies.

Ecological Security Patterns (ESPs) have been established as a crucial spatial strategy for safeguarding regional ecological integrity and sustainable development [12,13]. Methodologically, the field has converged on a fundamental paradigm of “source identification–resistance surface construction–corridor extraction” [14]. For source identification, common practices integrate ecosystem service assessments with structural connectivity analysis, such as Morphological Spatial Pattern Analysis (MSPA) [15]. For corridor delineation, methods like circuit theory and the Minimum Cumulative Resistance model are widely adopted for modeling landscape permeability [16]. Furthermore, graph-theoretic approaches are increasingly applied to analyze the topology and robustness of identified ecological networks [17,18], while bio-inspired computation algorithms, such as Ant Colony Optimization [19,20], have shown promise in solving complex spatial optimization problems, including corridor routing [12,21].

However, despite these advances, prevailing methodologies for multi-scale ESP construction suffer from three core computational limitations that restrict their realism and utility in polycentric urban agglomerations [22,23]: (1) Static Modeling Paradigms: A reliance on static resistance surfaces that cannot incorporate adaptive biological behaviors or dynamic species movement strategies [24,25]. (2) Lack of Bio-Inspired Feedback: An inability to dynamically optimize corridor pathways and their hierarchical importance based on bio-behavioral feedback, limiting ecological realism [26,27]. (3) Model Integration Deficits: A persistent disconnect between disparate modeling approaches (e.g., ecosystem service quantification, structural connectivity analysis, and policy data), leading to fragmented and often contradictory conservation priorities [28]. This results in a tendency toward “isolated conservation” that fails to proactively bridge scale mismatches.

To overcome these algorithmic gaps, we introduce a novel Proactive Integration Mechanism (PIM)—a spatially explicit computational framework that integrates multi-source geospatial data through three synergistic innovations: (1) A Function–Structure–Policy (F-S-P) Data Fusion Module: This module couples Self-Organizing Map (SOM) clustering, MSPA, and policy data for multi-dimensional ecological source identification [29,30]. (2) A Dual-Feedback Mechanism: This mechanism hybridizes circuit theory with an Improved Ant Colony Optimization (IACO) algorithm to dynamically delineate corridors by simulating both physical connectivity and adaptive species movement. (3) A Complex Network Analysis Pipeline: This pipeline translates topological centrality into spatially explicit, targeted interventions for network optimization, thereby bridging the gap between spatial ecology and computational intelligence.

The Chengdu-Chongqing Economic Circle (CCEC), a strategic national urban agglomeration in China characterized by complex transboundary ecological challenges [31,32,33], serves as an ideal validation site. Tongnan District, positioned at its geometric core and designated as a critical ecological node, provides a real-world testbed for the proposed “local optimization—regional enhancement” computational mechanism [34].

Building upon these identified gaps and innovations, this study aims to: (1) propose and implement the computationally integrated PIM framework to enhance cross-scale ESP resilience; (2) develop and apply the ‘F–S–P’ framework for multi-dimensional ecological source identification; (3) construct a POI-enhanced ecological resistance surface to better represent anthropogenic pressure; (4) establish the Dual-Feedback Mechanism for dynamic corridor extraction and barrier point identification; and (5) apply complex network theory to optimize the ecological network’s topology and validate its robustness within the CCEC context (Figure 1).

The remainder of this paper is organized as follows. Section 2 details the study area and data sources. Section 3 presents the PIM framework’s methodological innovations, including the F-S-P source identification, Dual-Feedback corridor extraction, and network optimization, alongside validation strategies. Section 4 presents the empirical results and discusses their implications for cross-scale ecological planning, positioning PIM’s contributions within the broader methodological landscape. Section 5 concludes with key insights and future directions.

2. Materials and Methods

2.1. Study Area

Tongnan District (30°11′–30°19′ N, 105°50′–105°84′ E) is situated at the geometric core of the CCEC (Figure 2). This strategic location designates it as a critical ecological node city, bridging the Chengdu and Chongqing metropolitan areas and serving as a vital conduit for cross-boundary ecological flows. Covering 1583 km² within a subtropical humid monsoon climate zone, the district features a heterogeneous landscape of low hills, mountains, plains, and terraces. This spatial complexity generates fine-grained variation in ecological processes and anthropogenic pressures, providing a rigorous testbed for model sensitivity. A key feature is the Fujiang River, a designated ecological corridor under the CCEC’s “Four Barrier Zones–Six Corridors” spatial plan, which traverses the district and sustains habitats for endangered species (e.g., Procypris rabaudi). Furthermore, Tongnan’s official status as a regional integration pilot zone offers a unique policy context. This allows the operationalization and validation of the “local optimization–regional enhancement” transmission mechanism central to our PIM framework. Consequently, Tongnan District represents an ideal scaled microcosm for computationally addressing the multi-scale spatial governance challenges between local planning and regional ecological strategy.

2.2. Data Sources

The main data and their formats, sources and other key information are shown in

Table 1. Main data used in this study.

Data Type	Year	Resolution	Data Source	Data URL
CNLUCC	2020	30 m	Resource and Environmental Science Data Platform	https://www.resdc.cn/DOI/doi.aspx?DOIid=54 (accessed on 20 August 2025)
Annual precipitation	2020	1 km	National Earth System Science Data Center, National Science & Technology Infrastructure of China	https://www.geodata.cn/main/face_science_detail?id=56226&guid=113786088533256 (accessed on 15 October 2025)
Evapotranspiration	2020	1 km	National Earth System Science Data Center, National Science & Technology Infrastructure of China	https://www.geodata.cn/data/datadetails.html?dataguid=34595274939620&docId=465 (accessed on 15 October 2025)
Soil attribute data	-	30 m	National Cryosphere Desert Data Center	https://www.ncdc.ac.cn/portal/metadata/1fdf7dc7-7ecb-4e1f-a5df-30f7196756a8 (accessed on 20 August 2025)
DEM	-	30 m	Geospatial Data Cloud	https://www.gscloud.cn/sources/details/aeab8000652a45b38afbb7ff023ddabb?pid=302 (accessed on 20 August 2025)
Road, water, and settlement distribution data	2020	shapefile	OpenStreetMap	https://www.ncdc.ac.cn/portal/metadata/1fdf7dc7-7ecb-4e1f-a5df-30f7196756a8 (accessed on 20 August 2025)
Night light	2020	30 m	National Earth System Science Data Center, National Science & Technology Infrastructure of China	https://www.geodata.cn/main/face_science_detail?guid=8213124601985&docId= (accessed on 15 October 2025)
Population density	2020	30 m	Resource and Environmental Science Data Platform	https://www.resdc.cn/DOI/DOI.aspx?DOIID=32 (accessed on 15 October 2025)
POI	2020	30 m	Bigemap	http://www.bigemap.com/ (accessedon 15 October 2025)
Administrative division	2020	shapefile	YUDITU	https://yuditu.com/bzdt/index.html?Name=%E6%BD%BC%E5%8D%97%E5%8C%BA%E8%A1%8C%E6%94%BF%E5%8C%BA%E5%88%92 (accessed on 20 August 2025)
NDVI	2020	30 m	National Science & Technology infrastructure	https://www.nesdc.org.cn/sdo/detail?id=60f68d757e28174f0e7d8d49 (accessed on 15 October 2025)

. To ensure spatial consistency across all multi-resolution datasets, all input layers were resampled to a unified resolution of 30 m prior to analysis. All data are publicly available at the URLs listed in Table 1; no permissions are required for download.

2.3. Methodology

2.3.1. Identification of Ecological Sources

Ecological sources are the core components of the ESP, defined as strategic areas characterized by high ecosystem services and critical landscape connectivity [17,35]. To overcome the “isolated conservation” practices in traditional ESP construction—where source identification often neglects policy-specified corridors and nonlinear connectivity [22,36,37], this study proposes a novel “F–S–P” three-dimensional framework. This framework integrates: Ecological sources are the core components of the ESP, defined as strategic areas characterized by high ecosystem services and critical landscape connectivity. (1) Functional dimension (anticipating nonlinear ecosystem services synergies via SOM-InVEST), (2) Structural dimension (identifying core habitats via MSPA), and (3) Policy dimension (internalizing cross-scale conservation priorities from planning documents) (Figure 3).

(1)

Functional Ecological Sources

Functional Ecological Sources (FESs) were identified by coupling SOM clustering with the InVEST model suite. This approach maps synergistic Ecosystem Service Bundles (ESBs) aligned with Sustainable Development Goal 15, encompassing four key services: Water Yield, Habitat Quality, Carbon Sequestration, and Soil Conservation (Table S1). To enhance methodological transparency, the specific parameters and data sources used for the InVEST model calculations—such as threat layers for Habitat Quality and carbon pool densities for Carbon Storage—are documented in the Supplementary Methods (Section S1.1; [38,39,40,41,42]) and detailed in Tables S3–S6. Subsequently, the SOM algorithm was applied to these four standardized ecosystem service layers at the grid scale to identify distinct ESBs. The optimal number of clusters was determined to be five, as evidenced by the Davies-Bouldin index reaching its minimum at this value (Figure S1), indicating the best trade-off between intra-cluster compactness and inter-cluster separation [43]. Consequently, five ESBs were retained for subsequent analysis to define the FESs.

(2)

Structural Ecological Sources

Structural Ecological Sources (SESs) derived from MSPA core areas (forests/grasslands/water foreground; [44]), processed via Guidos Toolbox with 8-connectivity (

Table 2. Description of MSPA landscape types.

Landscape Type	Ecological Significance
Core	Serves as a “source” for various ecological processes and is of great significance for species reproduction and biodiversity conservation.
Islet	Disconnected and fragmented small patches with low connectivity between patches, resulting in limited material and energy exchange and transfer within the patches.
Perforation	Transitional zones between core and non-green landscape patches, consisting of small patches at the edges of the interior patches, which are independent and have low connectivity.
Edge	Transitional zone between the edge of a core and the surrounding non-green landscape, helping to reduce the impact of external environmental and anthropogenic disturbances.
Loop	Channels for the exchange of materials and energy within the same core, serving as shortcuts for material and energy exchange within the core area.
Bridge	Narrow areas connecting different core patches, facilitating species migration and landscape connectivity within the area.
Branch	Channels connecting only one end to the main patch, serving as pathways for species dispersion and energy exchange with the surrounding landscape.

Note: This table is adapted from the GuidosToolbox MSPA user manual.

).

(3)

Policy-Integrated Ecological Sources

Policy-Integrated Ecological Sources (PIESs) incorporated CCEC-mandated rivers (e.g., Fujiang River) to rectify MSPA’s fragmentation of linear corridors, ensuring trans-regional connectivity for endangered species migration (Procypris rabaudi; [45]) and sediment transport [46]) per the “Four Barrier Zones–Six Corridors” plan [34]. Composite sources were generated through optimized spatial overlay.

2.3.2. Construction of Ecological Resistance Surface

Landscape resistance integrates natural/socioeconomic factors [17,47,48]. A baseline resistance surface was derived from land-use data by integrating the classical landscape-permeability paradigm of Adriaensen et al. [49] and the “source (low)–buffer (medium)–barrier (high)” resistance logic proposed by Yu, K. et al. [50]. Relative values were further calibrated against published parameters from comparable regional studies in China [17,51] (

Table 3. Base resistance values.

Land Use	Resistance Value
Water body	10
Forestland, Grassland	30
Cultivated land	50
Unused land	70
Construction land	90

). This was refined via natural factors (DEM, slope, NDVI, water proximity) and socioeconomic factors (road proximity, nightlight, population density, POI density). POI classification (industrial/residential/commercial) provided granular human-impact mapping, overcoming nightlight-data limitations. Final ERS was calculated as:

$$RC=\sum _{i=1}^n{\lambda }_i{ \times N}_i/{\mu }_i$$

(1)

$${\lambda }_i={\mu }_i/(\sum _{j=1}^n{\mu }_j)$$

(2)

$$R_{comprehensive}=({RC}_{nature}+{RC}_{socio-eco})/2\times R_{base}$$

(3)

where $RC$ represents the natural or Socio-Economic correction factor; $n$ denotes the number of resistance factors; $N_i$ denotes the value of the $i$th factor; for example; $N_{water}$ represents the resistance value of the water body distance; ${\mu }_i$ represents the weight or standard value of the $i$th factor, reflecting its importance or serving as a benchmark; ${\lambda }_i$ indicates the normalized weight of the $i$th factor, calculated as ${\mu }_i/\left(\sum _{j=1}^n{\mu }_j\right)$, ensuring the sum ${\mu }_i$ of all weights equals. (The weights (${\mu }_i$) for all normalized factors in Equations (1)–(3) were assigned equal values [22,52,53]. This equal-weight scheme was adopted as a neutral baseline, given the absence of species-specific movement data to justify differential weighting [54,55].

2.3.3. Extraction of Ecological Corridors and Ecological Barrier Points

The computational objective of this section was to develop a hybrid algorithm that surpasses the limitations of static corridor models by integrating physical connectivity principles with bio-inspired optimization, thereby dynamically simulating species movement and identifying corridors with hierarchical importance (Figure 4).

(1)

Extraction of Ecological Corridors.

The conventional ACO algorithm exhibits critical limitations, including premature convergence, redundant path crossover, and high parameter sensitivity [56], which reduce its efficacy in simulating complex ecological processes. To address these issues, we employed an IACO algorithm. Specific computational enhancements included the introduction of a 2-opt local search optimization function to iteratively eliminate path crossings and reduce unnecessary detours, which significantly accelerates convergence speed and promotes more efficient path generation. Additionally, the pheromone matrix was dynamically updated based on path quality—such as assigning stronger reinforcement to shorter paths—to avoid premature convergence and enhance global search capability.

These improvements were designed to enhance the algorithm’s efficiency, biological realism, and applicability in modeling ecological networks. The IACO was implemented on the Python platform with the following parameters, which were determined through preliminary sensitivity analysis to balance exploration and exploitation: ant count = 100, iterations = 500, pheromone importance = 1, heuristic factor = 5, and pheromone evaporation coefficient = 0.2. The rationale for selecting these parameter values and the details of the sensitivity analysis are provided in the Supplementary Methods (Section S1.2 and Figure S12).

(2)

Extraction of Ecological Barrier Points.

Ecological barriers are critical zones that significantly reduce connectivity, manifesting as abrupt gradients in current density—areas where species migration probability declines sharply in space [12,57]. Within ecological security frameworks, these barriers represent “migration bottlenecks” caused by heterogeneous landscape resistance (e.g., intense human activity or physical obstructions), which disrupt or delay the movement of organisms, genes, or ecological processes. Such barriers can impair connectivity, suppress gene flow, and degrade ESs [57]. This study employed the Barrier Mapper module in Linkage Mapper to identify barriers, which were then classified based on corridor importance: barriers intersecting high-importance corridors were categorized as Level 1, other moderate-importance corridors as Level 2, and barriers on low-importance corridors as Level 3. Higher-level barriers should be prioritized for restoration, as their remediation substantially enhances corridor connectivity.

2.3.4. Determination of the Range of Ecological Corridors

Theoretical justification for integrating Circuit Theory and IACO: The fusion of circuit theory and IACO is designed to capture complementary aspects of ecological connectivity [27]. Circuit theory models landscape connectivity as a random walk process, identifying areas with high probability of movement (current density) and thus representing potential, diffusive connectivity across the entire landscape [16]. In contrast, the IACO algorithm simulates goal-directed, adaptive movement where “ants” iteratively find optimal paths between sources based on pheromone feedback, modeling learned or efficient migration routes [21]. Combining these outputs through kernel density and overlay analysis allows us to identify corridors that are both physically well-connected (circuit theory) and likely to be efficiently utilized by organisms (IACO) [58]. This “Dual-Feedback Mechanism” thus integrates a physical connectivity framework with a bio-behavioral simulation to derive a more ecologically comprehensive corridor network.

In this study, cumulative current density raster was generated using Linkage Mapper, with the top 40% of high-value areas were extracted. The IACO resulting paths underwent kernel density analysis to produce kernel density surface, from which the top 40% of high-density areas were selected. Overlay analysis between high-current-density zones (from Linkage Mapper) and high-density ant colony paths defined the spatial extent of high-priority ecological corridors as the intersection of both datasets. To integrate outputs from both methods, the current density raster (Linkage Mapper) and ant colony kernel density results were normalized to a common scale and fused using an equal-weight fusion method. The fused output was classified into five levels using the Natural Breaks (Jenks) method, with the top three levels designated as the range of moderate-importance corridors. For general Moderate-importance corridors, regions with the highest cumulative current density and lowest resistance values were selected from the Linkage Mapper output.

Kernel density analysis effectively identifies high-use corridor zones, enhancing the spatial precision of corridor delineation [58]. The proposed “Dual-Feedback Mechanism” enables a comprehensive assessment of corridor dimensions, supporting more effective ecological management and conservation planning. By integrating circuit theory with the IACO algorithm, this methodology offers a robust, multi-model framework that balances physical connectivity with adaptive biological behavior, thereby addressing both static and dynamic ecological processes.

2.3.5. Topological Characteristics of the Network

An ecological network, as a complex network of ecological significance, can be systematically investigated through complex network theory to analyze its structural characteristics in greater depth [17,18]. By mapping the ecological network into an undirected graph within a physical network framework—where nodes correspond to the ecological source sites and edges represent ecological corridors—the intrinsic attributes and interrelationships among ecosystem components can be elucidated through topological metric analysis. This approach provides precise guidance for network optimization at the micro-level, enabling targeted enhancements to ecological connectivity and functionality. All the followed metrics were calculated using UCINET 6.212.

2.3.6. Robustness and Uncertainty Assessment Framework

The PIM is a deterministic geocomputational framework that integrates multi-source data, spatial algorithms, and expert knowledge. Quantifying the combined uncertainty of its final outputs (e.g., the exact number of corridors or the percentage increase in α-index) through classical statistical error propagation is challenged by the framework’s complex structure, which includes nonlinear clustering (SOM), rule-based policy integration (PIES), heuristic optimization (IACO), and topological decision-making. Instead of providing probabilistic confidence intervals, which may not align with the practical use of such a planning-support tool, we adopt a multi-faceted robustness validation strategy to ensure the reliability and credibility of our results.

(1)

Parameter Sensitivity Analysis

We systematically test the core computational algorithm (IACO) to evaluate the sensitivity of key outputs (e.g., corridor paths and total length) to parameter choices, demonstrating stability within a reasonable parameter space.

(2)

Temporal Persistence Validation

We leverage long-term time-series data (2000–2020) to examine whether the identified spatial components of the ESP (sources, corridors) maintain their configuration and ecological quality over time, providing strong evidence that they reflect persistent landscape characteristics rather than stochastic model artifacts.

(3)

Comparative Methodological Validation

We compare the network optimized by the full PIM against networks generated by traditional methods and intermediate stages of our framework. The coherent, unidirectional, and systematic improvement across a suite of complementary graph-theoretical metrics (α, β, γ, centrality indices) provides strong evidence that the observed enhancements are structurally robust and methodologically driven, rather than stochastic variations.

(4)

Spatial Consistency with Authoritative Plans

We assess the spatial congruence between our identified high-priority corridors and macro-regional ecological conservation strategies (e.g., the CCEC’s “Six Rivers” plan). This high degree of alignment provides external, policy-relevant validation for the model’s outputs. This comprehensive strategy cross-validates the robustness of the PIM’s outputs from multiple independent perspectives (parametric, temporal, methodological, and policy), offering a rigorous alternative to classical uncertainty quantification for complex spatial integration frameworks.

3. Results

3.1. Construction of the Ecological Security Pattern

3.1.1. Ecological Sources Identification

(1)

Functional Ecological Sources identification

The application of the SOM at the grid scale provided a data-driven and spatially explicit means to transcend subjective thresholding in ecosystem service value assessment. This geocomputational approach successfully identified five distinct ESBs (Figure S3), revealing clear spatial heterogeneity and patterns of synergies and trade-offs (Figure S2). Bundles 3 (69% Habitat Quality), 4 (a multifunctional bundle), and 5 (balanced Water Yield and Carbon Sequestration) were rigorously identified as FESs. The explicit exclusion of Bundles 1 (homogenized agriculture) and 2 (urban-dominated) underscores the utility of our SOM-based method in moving beyond land-cover proxies to pinpoint areas with synergistic, high-capacity ESs for conservation prioritization.

(2)

Structural Ecological Sources identification

MSPA-derived core areas, selected as SESs, covered 6666.21 hm² (57.54% of the foreground). These cores, overlapping with high-NDVI FESs, were primarily concentrated in forests and grasslands, while water bodies were also included due to their ecological significance. Critically, spatial analysis revealed a pattern of “dispersed high-quality habitat patches with scarce connecting nodes,” indicating insufficient landscape connectivity despite high habitat quality (Figure S4). This finding highlights the urgent need for the corridor optimization undertaken in this study.

(3)

Policy-Integrated Ecological Sources identification

The F-S-P integration framework fundamentally addressed a critical spatial data and policy misalignment. We identified that MSPA’s technical limitation regarding minimum edge width had artificially fragmented the policy-mandated Qiongjiang River into 347 isolated segments, creating a cartographic representation that deviated from hydrological reality and regional planning objectives (Figure S5). To reconcile this systemic discrepancy, our proactive PIESs method involved topologically restoring designated water bodies per the “Six Rivers” Ecological Corridor Construction Plan. This spatially intelligent integration expanded the source area by 3884.76 hm² (a 37.6% increase) and critically recovered the Qiongjiang and Fujiang Rivers as continuous ecological corridors. The resulting 22 composite sources effectively balance ecological functionality with spatial governance feasibility, while computationally amplifying the strategic “leverage effect” of the Tongnan node city within the macro-scale CCEC ESP.

3.1.2. Ecological Resistance Surface Construction

The base resistance surface, constructed from land-use data, was refined using natural and socio-economic factors, with POI data providing granular mapping of human impact (Figure 5). The final resistance surface clearly demonstrated the dominant role of human activities: low resistance values aligned with forests, grasslands, and water bodies, while high resistance values coincided with farmland and construction land. Of particular concern, high resistance areas were identified around the Fujiang and Qiongjiang Rivers—key corridors in the CCEC’s plan—suggesting that urban expansion poses a significant threat to their connectivity. This finding underscores the urgency of implementing the ecological restoration strategies identified in this study.

3.1.3. Ecological Corridors Extraction and Ecological Barrier Points Identification

(1)

Hierarchical Spatial Patterns of Ecological Corridors

The application of our Dual-Feedback Mechanism (DFM) generated a hierarchical and spatially complex corridor network, advancing beyond static models. By integrating kernel and current density analyses, we classified 37 corridors (total length 221.7 km) into three functional tiers (Figure 6 and Figure S6). The 18 high-importance corridors (61.7% of total length) form long-distance connectivity trunks in northeastern and southeastern forests and along the Fujiang River. The six moderate-importance corridors, though shorter, critically enhance local connectivity along key rivers, while the 13 low-importance corridors (37.2% of length) in southern forests reveal significant potential for future ESP optimization. This hierarchical output provides a spatially explicit and prioritized blueprint for conservation resource allocation.

(2)

Spatial Configuration of Critical Ecological Barriers

The Barrier Mapper module identified 25 ecological barrier points, which were classified based on the hierarchical importance of the corridors they disrupt. This classification establishes a clear spatial priority framework for targeted restoration. Level 1 barriers (14 sites) are predominantly (92.3%) located within crop-intensive areas adjacent to high-importance corridors, quantitatively highlighting the significant fragmenting impact of agricultural land use on the most critical ecological links. A single Level 2 barrier on the Qiongjiang River obstructs along-river species migration, while the 10 Level 3 barriers are situated within fragmented southern forest patches and the Qiongjiang River Basin (Figure 6).

3.1.4. Range Delineation and Classification of Ecological Corridors

The spatial extent of corridors was delineated through the fusion of current density and kernel density surfaces (Figure 7). High-importance corridors exhibit highly continuous and clustered spatial ranges, resulting from a strong coupling between high current density and path density, which reflects a distinct core-edge structure and enables them to play a central role in maintaining network connectivity. Moderate-importance corridors demonstrate a more even distribution overall but contain localized narrowed segments, necessitating the design of buffer zones to compensate for these functional deficiencies; they primarily serve as buffers and supplements within the ecological network. Low-importance corridors are fragmented and widely distributed, reflecting currently low connectivity but indicating considerable potential for future restoration efforts.

3.2. Topological Analysis of the Ecological Network

Modeling ecological sources as nodes and corridors as edges allowed us to abstract the spatial ESP into a complex network for topological analysis. Using UCINET, we calculated key metrics—betweenness, clustering, closeness, and degree centrality—to move beyond mere spatial description and quantify the structural properties and functional roles within the ecological network of Tongnan District (Figure 8b).

3.2.1. Degree Centrality

Node 8 exhibited the highest degree centrality value of 9, establishing it as the network’s primary hub for resource aggregation and distribution. This high-degree connectivity signifies Node 8’s role as a runoff convergence center and a core exporter of ESs. Nodes 12 displayed the second-highest degree centrality value of 8, underpinning critical linkage functions. Node 14 displayed the third-highest degree centrality value of 6. The degree centrality values of the other nodes are no more than 5 (Figure S7).

3.2.2. Betweenness Centrality

The betweenness centrality analysis revealed a network centralization index of 52.86%, which indicates a structural vulnerability: the network’s connectivity is strongly reliant on a limited set of core nodes. Node 12 exhibited the highest betweenness centrality (normalized value 58.64%), followed by Node 8 (48.86%) and Node 14 (16.42%). All other nodes displayed normalized values under 10% (Figure S7). This topological finding quantifies the system’s dependence on key connectors and identifies potential points of systemic failure.

3.2.3. Closeness Centrality

Analysis of the study area revealed a network centralization of 37.78%, indicating greater reliance on a limited subset of nodes for ecological flows. From the perspective of specific nodes, Node 12 emerged as the core hub for ecological propagation, with a closeness centrality value of 56.76. Moreover, Nodes 8 and 14, both exhibiting closeness centrality values of 48.84, functioned as secondary hubs. Node 22 displayed the lowest closeness centrality (24.14), reflecting its peripheral network position (Figure S7).

3.2.4. Clustering Coefficient

The overall graph clustering coefficient of 0.60 indicates a moderate tendency for nodes to form tightly interconnected groups. However, the lower weighted overall graph clustering coefficient (0.33) reveals significant heterogeneity in actual connection strengths. Nodes 1, 2, 3, 4, 6, and 9 formed densely interconnected local clusters (clustering coefficient = 1.0), yet their low degree centrality (value = 2) suggests they function as local cliques rather than network-wide hubs. Conversely, the identification of Nodes 10 and 22 as isolated nodes with no ecological flow interactions highlights critical discontinuities in the network (Figure S7).

3.3. Optimization of the Ecological Network

The network optimization was designed by integrating centrality-based topological ranking with spatially explicit land-use analysis, translating system-level diagnostics into targeted interventions for enhancing ecological connectivity (Figure 8c,f).

3.3.1. Critical Node Identification and Priority Interventions

A synthesis of the four-centrality metrics identified Nodes 12, 8, and 14 as structurally pivotal and ecologically significant, warranting prioritized intervention (Figure S7). Node 12 (Qiongjiang River), with the highest betweenness and closeness centrality, was identified as the primary bridge connecting the northern and southern subnetworks. To safeguard its role in maintaining global network efficiency, a 200 m riparian protection buffer is proposed. Node 8 (Fujiang River), the highest in degree centrality, functions as a secondary hub connecting multiple subnetworks and is recommended for a similar buffering strategy. Node 14, a terrestrial connectivity hub in the south, should be designated for strict conservation and targeted restoration. These interventions are directed by network theory to fortify the most critical points of the spatial system.

3.3.2. Intervention for Isolated Nodes and Connectivity Enhancement

Node 10 and Node 22, both exhibiting the lowest degree centrality (value = 1), are at high risk of isolation from the broader network. To address this, a connectivity reinforcement strategy is proposed through edge addition and corridor enhancement. For Node 10, its nearest neighbor excluding Node 12 is Node 11. A new corridor is planned between Node 10 and Node 11, incorporating Stepping Stone 1, a forested patch located between the two nodes. This stepping stone serves as an intermediate habitat to facilitate organism movement and strengthen ecological flow. Similarly, for Node 22, the nearest neighbor aside from Node 21 is Node 19, which contains the Chongqing Tongnan Wuguishan Municipal-level Forest Nature Park. A new corridor between Node 22 and Node 19 is proposed, including Stepping Stone 2, a forested area that will serve as a stopover point to enhance the effectiveness of the new linkage.

3.3.3. Long-Corridor Optimization Using Stepping Stones

The corridor connecting Nodes 1 and 2 was identified as the longest in the current network configuration, potentially limiting its ecological function. To optimize this corridor, Stepping Stone 3, a forested patch located along the route, is introduced. This stepping stone will provide critical resting and foraging habitat for species during migration and enhance the overall permeability of the corridor.

3.3.4. Addressing Low Clustering and Ecological Blind Spots

To address the systemic vulnerabilities identified by low clustering coefficients around key hubs (Nodes 8 & 12) and the ecological blind spot in the central-northern region, we propose a proactive network modification: the creation of a new ecological source node (Node 23). This node, formed by merging contiguous grassland segments, will be connected via new corridors to Nodes 1, 2, and 3. This intervention is designed to increase local redundancy around the core hubs and enhance the structural robustness and functional integration of the previously under-connected northern subnetwork.

3.4. Integration of Ecological Security Pattern

The final ESP, structured around the “One Core, One Ring, Dual Belts, Three Zones” (Figure 9b), is the spatially explicit embodiment of our integrated geocomputational framework. It demonstrates how the PIM translates multi-source data and algorithmic outputs into a cohesive and policy-relevant spatial strategy that strategically reinforces the CCEC’s ESP (Figure 9a) through a node-city-centric configuration.

3.4.1. One Core

The One-Core (Southern Subnetwork Backbone Core) configuration comprises ecological source nodes 11, 13, 14, 15, 16, 18, and 19, each buffered by a 100 m zone to enhance ecological protection. This core area is located in the upstream region of the Qiongjiang River, within the southern sector of the Tongnan District. Node 11 corresponds to the Chongqing Tongnan Ma’anshan Municipal-level Forest Nature Park, and Node 19 to the Chongqing Tongnan Wuguishan Municipal-level Forest Nature Park, both of which exhibit high levels of biodiversity and are prioritized for strict conservation. Node 14 demonstrates notable structural importance within the ecological network, ranking third in both degree and betweenness centrality, and second in closeness centrality. Additional ecological sources in the southern region—specifically Nodes 17, 20, 21, and 22—can be functionally integrated into the core via connecting Nodes 14, 16, or 19. The 100 m buffer applied to each core node is designed to mitigate edge effects and support habitat continuity. Strategically positioned, the Core functions as the ecological nucleus of the southern subnetwork and plays a pivotal role in promoting the long-term sustainability of the Southern Ecological Conservation and Farming Zone.

3.4.2. One Ring

The One-Ring (High-priority Connectivity Ring) structure is primarily composed of priority ecological corridors and their associated ranges, along with the ecological nodes they interconnect. This circular framework integrates seamlessly with the Two Belts, forming a cohesive spatial configuration. The importance corridors, identified through the “Dual-Feedback Mechanism”—a methodology that integrates the strengths of circuit theory and bio-behavioral feedback—serve as the structural backbone of the ecological network in the Tongnan District. Functionally, the One-Ring supports the primary flows of matter and energy within the region and aligns effectively with the CCEC’s ESP (Figure 9a), particularly through its synergy with the hydrological dynamics of the Two Belts.

3.4.3. Two Belts

The Two-Belts framework computationally formalizes the synergy between local network topology and macro-regional policy. The Fujiang Belt (Node 8) and Qiongjiang Belt (Node 12) are not merely policy designations but were identified as topologically central hubs through our complex network analysis. Their inherent hydrological interconnection, coupled with their pivotal network roles (e.g., Node 8 as a secondary hub, Node 12 as the primary bridge), creates a synergistic “leverage effect” (Figure 9a). This finding validates that strategic node cities can be computationally engineered to amplify regional ecological coherence, directly aligning local optimization with the CCEC’s broader security strategy.

3.4.4. Three Zones

The Three-Zone framework comprises the Northern Ecologically Restored Agricultural Zone, the Central Green Development Zone, and the Southern Ecological Conservation and Farming Zone. Across all zones, the protection of cultivated land is a priority. All permanent basic farmland should be designated within farmland protection areas, with the construction of high-standard farmland prioritized to enhance agricultural resilience and productivity.

The Northern and Southern Zones are characterized predominantly by hilly and mountainous terrains, respectively. These areas are well-suited for the development of agroforestry systems that integrate ecological restoration with agricultural productivity. In addition, moderate-scale herbivorous livestock farming is considered appropriate for the Southern Zone, given its ecological and topographic conditions.

The Central Zone, which encompasses the urban core of the Tongnan District, features relatively flat terrain and accommodates urban and town development needs. This zone is recommended for the advancement of urban agriculture, including the cultivation of vegetables and specialty crops such as lemons.

Due to its location between the Fujiang and Qiongjiang Belts, the Central Zone plays a critical ecological role. Particular attention should be given to the prevention and management of agricultural non-point source pollution and urban point-source pollution to safeguard the ecological integrity of both the Central Zone and the adjacent ecological belts.

The innovative “PIM” framework, incorporating “F–S–P” multidimensional ecological source identification, the “Dual-Feedback Mechanism” or corridor optimization, and the Complex Network approach for topology-driven optimization, effectively integrates Tongnan’s ESP and the CCEC’s broader ecological security strategy. This enhances both local and regional ESP while providing a potentially replicable paradigm for constructing ESPs in node cities within subtropical urban agglomerations.

3.5. Validation of Framework Robustness

3.5.1. Ecological Relevance of Network Components

To address the ecological functionality of the identified ESP, we evaluated two key aspects: the habitat quality of ecological sources and the spatial relationship between corridors and species-relevant conservation zones.

(1)

Habitat Suitability of Ecological Sources

The composite ecological sources identified by the F-S-P framework exhibited persistently high vegetation vigor (mean NDVI > 0.72 from 2000 to 2020, Section 4.4.1) and were predominantly composed of forest and water bodies (Figure 10). These characteristics—stable, well-vegetated, and aquatic habitats—represent the core environmental requirements for the focal species mentioned in regional plans. For instance, forested riparian zones provide critical habitat for Andrias davidianus [59], while undisturbed river segments are essential for Procypris rabaudi [45,60].

(2)

Spatial Congruence with Conservation Frameworks

The optimized high-importance corridors, especially the Fujiang and Qiongjiang Belts, show perfect spatial alignment with the CCEC’s “Six Rivers” ecological corridor plan (Figure 9a). This plan is explicitly designed to protect watershed connectivity and aquatic biodiversity. While precise occurrence records for rare species like Procypris rabaudi are scarce due to its critically endangered status and monitoring challenges [60,61,62], the corridors’ placement within this policy-mandated network ensures they are prioritized for restoration and conservation actions aimed at benefiting such species.

3.5.2. IACO Parameter Sensitivity and Stability Validation

To justify the selected IACO parameters and demonstrate the robustness of the framework, we conducted a sensitivity analysis focused on the key operational parameters: ant colony size and iteration limit. First, we tested ant counts ranging from 30 to 500 while keeping other parameters fixed (α = 1, β = 5, γ = 0.2, iterations = 500). As shown in Figure S12a, the algorithm exhibited strong robustness in solution quality across this wide parameter range, with optimal path distances consistently within a narrow, high-quality band (204,480.55 to 204,706.28). However, computation time increased linearly with ant count (Figure S12a). The selected ant count of 100 represented a pragmatic compromise, reliably achieving optimal or near-optimal solutions within a reasonable computation time of approximately 5 min. This sensitivity analysis confirms that the IACO-derived corridor network is robust to variations in its key operational parameters. The optimal path cost remained within a narrow band (e.g., 204,480.55 to 204,706.28, a variation in less than 0.11%) across a wide range of ant colony sizes, indicating that the core finding—the extraction of efficient ecological corridors—is not an artifact of a specific parameter choice. Second, we assessed the stability of the algorithm under the finalized parameters (ant count = 100, iterations = 500) through five independent runs (Figure S12b). All runs successfully converged to high-quality ecological corridor configurations. The optimal path distances clustered tightly between the two best-observed values, and convergence occurred within 67 to 232 iterations—well below the 500-iteration limit. This demonstrates both the consistency of the IACO implementation and the adequacy of the chosen iteration limit, which provides a substantial safety margin above the observed maximum convergence iteration. While the ant count and iteration parameters were tuned through this case-specific analysis, the values for α, β, and γ were adopted from well-established ranges in the ACO literature for combinatorial optimization problems, ensuring a principled foundation for the algorithm’s exploratory behavior.

3.5.3. Temporal Persistence of the ESP (2000–2020)

To assess the long-term reliability of the constructed ESP, we examined the temporal stability of its core components (sources, corridors, ranges, and river buffers) from 2000 to 2020 (Figure 10a–d and Figures S8–S11). Ecological sources maintained a stable spatial configuration, predominantly within protected areas. Their land cover showed positive trends: forest coverage consistently exceeded 34%, water bodies increased from 57.42% to 65.30%, and non-ecological land uses (cultivated/construction/unused land) disappeared within sources by 2020. Vegetation vigor remained high and stable, with mean NDVI values between 0.725 and 0.743 across the two decades.

Ecological corridors and their spatial ranges were consistently dominated by cultivated land and forest, with NDVI values persistently around 0.74, indicating sustained vegetation health within these connectivity pathways. Similarly, the key river buffer zones (Fujiang and Qiongjiang) maintained stable land-use composition and vegetation conditions (NDVI > 0.57). This remarkable temporal consistency across all ESP components provides strong evidence that the patterns identified by the PIM framework represent persistent and resilient features of the regional landscape, rather than transient or model-specific artifacts.

4. Discussion

4.1. Proactive Integration Mechanisms: A Computational Solution to Cross-Scale Mismatches

Rapid urbanization creates spatial–temporal asynchrony across nested scales, leading to “mismatched connections” that undermine conservation efficacy [63,64,65,66]. Conventional ESP approaches—dominated by the “source-resistance-corridor” paradigm—typically operate with static algorithms at single scales, neglecting cross-scale synergistic interactions [67,68,69].

To address this gap, we propose the Proactive Integration Mechanism (PIM). Unlike existing cross-scale frameworks that typically reconcile discrepancies between independently constructed local and regional ESPs through post hoc adjustment [67,68,69], PIM fundamentally restructures the workflow by embedding upper-scale policy directives as computational constraints during the initial source identification stage. This proactive integration eliminates the iterative reconciliation process, reducing scale mismatch errors by design rather than correction.

The PIM synergistically combines three innovative components: (1) the F-S-P framework for multi-dimensional source identification, (2) the DFM for dynamic corridor extraction, and (3) complex network analysis for topological optimization. By leveraging the strategic position of node cities, the PIM quantifies their “leverage effect” to catalyze regional ecological synergies, achieving deterministic improvements in network robustness (α-index: +20.5%; γ-index: +8.6%) with minimal topological modifications.

4.2. “F-S-P” Framework: A Paradigm Shift from Isolated to Integrated Source Identification

Accurate ecological source identification is fundamental to effective conservation planning; however, conventional methodologies—such as simplistic MSPA and InVEST overlay approaches—suffer from significant limitations. These include neglecting spatial heterogeneity in ecosystem services, which leads to erroneous prioritization [70,71,72], and a narrow local focus that overlooks broader policy imperatives, thereby disconnecting site-level assessments from regional or national conservation strategies.

To overcome these constraints, we propose the F-S-P framework, which embodies a conceptual and operational advance by systematically integrating Functional, Structural, and Policy-guided analyses. The core innovation of F-S-P lies in its tripartite integrative design: first, through SOM clustering, it incorporates functional heterogeneity by quantifying differentiated ecosystem service capacities across landscape patches, superseding the assumption of uniform provision—a critical enhancement for ecologically complex regions such as arid zones and biodiversity hotspots [70,73]. Second, the policy-driven designation mechanism explicitly embeds macro-strategic directives—exemplified by the CCEC’s “Six Rivers” initiative—by introducing PIES, a novel computational element that aligns model outputs with administrative conservation targets. Third, while MSPA delineates structural cores, the framework compensates for its technical drawbacks, such as artificial fragmentation of linear landscapes, through complementary functional and policy inputs, achieving a more holistic structural representation. Recent studies have similarly pursued multi-dimensional source identification by coupling SOM with MSPA [15,74]. However, these approaches remain biophysical, neglecting the policy dimension that critically determines conservation feasibility in governed landscapes [75]. The F-S-P framework’s explicit incorporation of PIES as computational inputs—rather than post hoc validation criteria—represents a decisive advance, as demonstrated by the restoration of 3353 hm² of continuous river corridors that fragmented methods failed to recover.

The demonstrated superiority of the F-S-P framework is unequivocal (Figure 8): it increased identified ecological source areas by a factor of 3.30 compared to conventional MSPA-InVEST overlay (7405.77 hm² vs. 2244 hm²), while achieving 100% spatial congruence with national policy corridors versus 47% in conventional approaches. Notably, it successfully restored the integrity of the previously fragmented Qiongjiang and Fujiang Rivers into continuous PIESs corridors spanning 3353 hm²—an outcome that aligns seamlessly with national ecological priorities (Figure 9a). This reintegration establishes vital connectivity between upstream erosion-prone regions and downstream habitats, offering crucial support for endangered species such as Leptobotia elongata and reinforcing large-scale initiatives like the Yangtze River upstream ecological barrier [23,66]. By transcending the algorithmic confines of traditional models through the principled incorporation of policy intelligence, the F-S-P framework enables a transformative approach to spatial conservation planning—one that is both scientifically robust and pragmatically aligned with governance realities.

4.3. “Dual-Feedback Mechanism”: Capturing Ecological Realism in Corridor Dynamics

Ecological corridors are vital for connectivity, yet traditional extraction methods (e.g., MCR, Circuit Theory) are often constrained by static resistance surfaces and an over-reliance on physical connectivity [76,77]. They fail to dynamically capture the influence of landscape heterogeneity on species’ adaptive behaviors and migration strategies, neglecting crucial biological realism [16,78]. Furthermore, functional corridor effectiveness depends critically on spatial attributes (width, shape, integrity) often overlooked, impacting edge effects and migration efficiency.

To address these critical gaps, we developed the innovative “DFM”. This method represents a significant advance by:

(1)

Coupling Physical and Biological Logic

Integrating Circuit Theory (providing a global connectivity framework based on current density) with the IACO algorithm (dynamically simulating species migration routes through pheromone feedback mechanisms). This explicitly models multi-path ecological flows and adaptive organism behavior.

(2)

Dynamic and Adaptive Simulation

Overcoming the static nature of traditional models by simulating species’ path selection and optimization in response to landscape features.

Application of the DFM identified 37 ecological corridors (total length 221.7 km). Crucially, 16 corridors are spatially aligned with the national “CCEC ‘Six Rivers’ Ecological Corridor Plan”, primarily connecting the Qiongjiang River, Fujiang River, and surrounding wetlands. This alignment validates the model’s ability to reconcile regional conservation needs with broader strategic objectives. Beyond extraction, the DFM framework enabled the development of a novel three-tier corridor classification system (High-Priority, Important, General-Important), based on integrated analysis of kernel density (biological movement preference) and current density (physical connectivity potential).

High-Priority: Strong significance in both physical structure and biological preference. Important: Strong biological preference but weaker physical structure. General-Important: Primarily structural connectivity. This classification provides a nuanced and comprehensive understanding of corridor functionality, moving beyond simple linear connectivity. It enables highly targeted conservation strategies (e.g., prioritizing width management for High-Importance corridors, habitat improvement for moderate-importance corridors). The DFM methodology thus significantly advances corridor design optimization and functional evaluation, providing a robust scientific basis for defining optimal pathways and spatial extents within node city ESPs, ultimately enhancing overall ecological security and resilience.

The parameter sensitivity analysis further validates the IACO framework’s design. The relative insensitivity of solution quality to ant count variation underscores the algorithm’s robustness in our ecological context, where the landscape resistance surface provides strong guidance for corridor formation. Our parameter selection—combining established conventions for α, β, and γ with empirically tuned values for ant count and iterations—effectively balances exploration and exploitation [79,80]. A colony size of 100 enables sufficient parallel exploration of potential corridor routes, while the 500-iteration limit, informed by observed convergence behavior, allows thorough pheromone feedback development without excessive computational cost (Figure S12). The improvements of IACO over a standard ACO are embedded in its core mechanisms. The integration of the 2-opt local search operator directly addresses path-crossing inefficiencies common in naive ant simulations, leading to shorter, more ecologically plausible corridors. This is evidenced by the algorithm’s consistent identification of the superior solution class of path distance (204,480.55). The adaptive adjustment of search parameters during iterations prevents premature stagnation and mimics adaptive foraging behavior, contributing to the variation in convergence speed observed in our stability tests (Figure S12).

4.4. Validation of ESP Construction and Optimization

4.4.1. Temporal Persistence and Stability of the ESP Components

The temporal validation (2000–2020) presented in Section 3.5.3 provides compelling evidence for the long-term ecological reliability of the ESP constructed by the PIM framework. The remarkable stability in land-use composition and sustained high vegetation vigor across all identified components—sources, corridors, and buffers—over two decades is a critical finding.

This persistence signifies that the PIM framework does not merely capture transient landscape patterns or model artifacts; rather, it successfully identifies enduring structural elements of the regional ecological fabric. Such temporal robustness is a key attribute for any spatial planning tool, as it increases confidence that conservation investments based on these patterns will yield long-term benefits. The stability of ecological sources within protected areas and the consistent quality of corridor pathways further validate the framework’s ability to integrate intrinsic landscape resilience into its computational design.

Importantly, this long-term validation serves as a powerful proxy for the overall robustness of our methodology. In complex, deterministic modeling frameworks like PIM, where classical statistical uncertainty quantification is challenging, demonstrating that outputs are stable over time provides a pragmatic and strong alternative form of credibility assessment. Traditional ESP validation relies primarily on algorithmic sensitivity analysis or single-metric accuracy assessment (e.g., corridor length consistency). The PIM validation strategy—integrating parametric robustness (IACO stability), temporal persistence (20-year land cover stability), structural coherence (multi-metric network improvement), and policy alignment (Six Rivers congruence)—provides a more comprehensive credibility assessment for deterministic geocomputational frameworks where classical statistical uncertainty quantification is infeasible.

4.4.2. Network Optimization Validation: A Comparison of Traditional, Pre-Optimization, and Optimization Methods

To rigorously validate the effectiveness of the core innovation–the PIM embedding cross-scale nesting early in ESP construction (specifically source identification)–we compared ecological networks built using three approaches: Traditional (conventional methods), Pre-optimization (PIM sources + traditional corridors), and Optimization (Full PIM: PIM sources + DFM corridors + network optimization). A suite of graph-theoretic metrics assessed changes in connectivity, centralization, heterogeneity, and clustering (

Table 4. Structural properties of ecological networks under different approaches.

Indices	Traditional	Pre-Optimization	Optimization
Nodes	20	22	23
Edges	35	37	42
Freeman’s degree centralization	29.74	28.57	25.32
Normalized heterogeneity (%)	1.40	1.57	1.03
Un-normalized centralization	2242	2331	2509
Freeman’s betweenness centralization (%)	59.00	52.86	49.37
Freeman’s closeness centralization (%)	34.68	37.78	37.07
Overall graph clustering coefficient (Unweighted)	0.562	0.602	0.563
Weighted Overall graph clustering coefficient	0.391	0.331	0.351

).

The results reveal a systematic and significant structural evolution (Table 4, Figure 8a–c): (1) Enhanced Decentralization & Resilience: Concurrent decreases in degree centralization (from 29.74% to 25.32%) and betweenness centralization (from 59.00% to 49.37%), coupled with increases in nodes (from 20 to 23) and edges (from 35 to 42), demonstrate a fundamental shift. The network evolved from a centralized, potentially brittle structure reliant on few critical hubs towards a flatter, more distributed topology. This signifies increased redundancy and robustness, where ecological flows can utilize multiple, alternative pathways, enhancing resilience to perturbations like habitat loss or fragmentation. (2) Optimized Global-Local Connectivity Trade-off: While unweighted clustering remained high (peak 0.602 Pre-optimization), closeness centralization increased (from 34.68% to 37.07%). This indicates the optimized network preserved valuable local cohesion (modularity for stability) while simultaneously improving global accessibility (connectivity for adaptability). The optimization enhanced inter-module connectivity, facilitating the flow of organisms and processes across the landscape [81,82,83]. (3) Stabilized Integration: The initial rise in heterogeneity (from 1.40% to 1.57% Pre-optimization) followed by a drop to 1.03% (Optimization), alongside a rebound in the weighted clustering coefficient (from 0.331 to 0.351), suggests a transition from uneven connectivity expansion to a more balanced, mature state. This reflects a strategic rebalancing of interaction strengths, optimizing both structural integration and functional efficiency.

Complementary insights from classical network indices (α, β, γ—

Table 5. Classical network indices across approaches.

Approaches	α	β	γ
Traditional	0.405	1.667	0.614
Pre-optimization	0.410	1.682	0.617
Optimization	0.488	1.826	0.667

) confirm these trends: (1) Steady Increase: All indices rose progressively (Traditional → Pre-optimization → Optimization). (2) Enhanced Redundancy & Completeness: The significant rise in α index (from 0.405 to 0.488) indicates vastly increased loop redundancy (more independent cycles/alternative pathways). The concurrent increase in γ index (from 0.614 to 0.667) signifies improved topological completeness (a higher proportion of possible connections realized), moving the network closer to its full connectivity potential. (3) Increased Efficiency: The rise in β index (from 1.667 to 1.826) reflects higher link density per node (connectivity), underpinning the growth in both redundancy (α) and completeness (γ).

Crucially, the most substantial improvements occurred during the transition from Pre-optimization to Optimization. This highlights the profound impact of integrating the full PIM framework–particularly the DFM corridors and complex network optimization–building upon the foundation laid by the proactive, cross-scale source identification (F-S-P). Remarkably, these significant gains in network resilience and connectivity (α, β, γ) were achieved through relatively minor topological additions (1 source, 4 corridors, 2 stepping stones), demonstrating the exceptional efficiency of the PIM approach. The reported improvements in classical network indices (e.g., the 20.5% increase in α and 8.6% increase in γ) are deterministic outcomes derived from comparing the pre-optimization and optimized network states. They quantify the structural efficacy of our PIM-based optimization. While these single values do not convey statistical confidence intervals, their reliability is affirmed by three key observations: (1) the coherent, unidirectional improvement across all complementary metrics (Table 4 and Table 5), (2) the robustness of the underlying corridor extraction algorithm (IACO) to parameter variation (Section S1.2), and (3) the temporal persistence of the network components (Section 4.4.1). Collectively, this multi-faceted validation provides strong evidence that the reported percentage gains are robust indicators of enhanced network resilience, not artifacts of stochastic computation.

4.5. Comparative Analysis of the PIM Framework

4.5.1. Systematic Comparison with Alternative Approaches

To explicitly position PIM’s added value within the methodological landscape of spatial conservation planning, we systematically compare its performance characteristics against three prevalent alternative approaches.

(1)

Systematic Conservation Planning (Marxan & Marxan Connect)

SCP tools excel at solving cost-efficiency problems—identifying near-optimal reserve networks that meet pre-defined biodiversity representation and connectivity targets within a fixed budget [26,84]. A critical prerequisite, however, is that users must supply explicit targets (e.g., protect 20% of each habitat type) [85]. In the Tongnan application, Marxan would require pre-specifying protection targets for 22 ecological sources and 37 corridors—data unavailable in regional planning documents. PIM’s generative design achieved equivalent network coverage (α = 0.488 vs. Marxan-estimated α ≈ 0.45–0.50 for comparable scenarios) without target pre-specification, at the cost of 6.5 h runtime versus Marxan’s typical 2–3 h [26]. The trade-off is that PIM’s optimization is heuristic (centrality-based ranking plus IACO), lacking the formal optimality guarantees provided by the mixed-integer programming core of Marxan.

(2)

Hierarchical Multi-scale ESP Frameworks

Recent studies emphasize analyzing ESPs at multiple administrative or ecological scales [37,86,87]. While hierarchical frameworks analyze scales sequentially and reconcile discrepancies through post hoc overlay [37], PIM’s F-S-P module preemptively eliminates scale mismatch by constraining local source identification with upper-scale policy corridors. This reduces the reconciliation error rate from the typical 15–30% [37] to zero in policy corridor alignment.

(3)

Advanced Graph-Theoretic Models

Graph theory is powerful for diagnosing network topology and identifying critical nodes/links [17,88,89]. PIM incorporates this in its final optimization pipeline. Its distinction lies in coupling topological analysis with preceding ecological process simulation. Unlike pure graph optimization that maximizes topological efficiency irrespective of ecological process (e.g., shortest path algorithms), PIM’s topology emerges from simulated species movement (IACO) and ecosystem service distribution (SOM), ensuring that network centrality metrics reflect functional importance rather than abstract connectivity. However, PIM’s current network optimization remains more heuristic than some formal graph optimization algorithms.

(4)

Synthesis

The PIM framework’s primary advantage is its integrated, simulation-driven workflow designed for complex, polycentric urban contexts where aligning local action with regional policy and simulating adaptive ecological flows are paramount. It is less suited for problems dominated by strict budget constraints or well-defined species representation targets, where systematic planning tools excel.

4.5.2. Explicit Articulation of Added Value

Drawing on the comparative analysis above and the empirical results, we explicitly articulate the added value of the PIM framework relative to existing research. The novelty is threefold:

Methodological added value–from passive reconciliation to proactive integration: Existing multi-scale ESP studies typically reconcile scales post hoc, whereas PIM introduces a “nested integration” design that proactively internalizes upper-scale policy mandates as computational primitives. To our knowledge, no previous ESP framework has systematically embedded policy directives directly into source identification.

Algorithmic added value–hybridizing physical connectivity with adaptive behavior: The DFM hybridizes circuit theory with IACO, capturing both global connectivity backbones and adaptive, pheromone-mediated path selection. This combination—absent in purely physical or purely bio-inspired models—yields corridors that are both structurally coherent and behaviorally realistic.

Practical and transferable added value–a replicable paradigm for node cities: PIM provides a structured workflow to translate qualitative policy mandates into quantifiable spatial inputs, achieves high optimization efficiency (20.5% α-index increase with only minor topological additions), and demonstrates temporal robustness over two decades (2000–2020). These features make it a replicable computational paradigm for node cities in subtropical urban agglomerations.

Collectively, the PIM framework moves beyond three common limitations in ESP research: (1) scale isolation → proactive nested integration; (2) static, passive connectivity → adaptive, behavior-informed corridor simulation; (3) policy-blind modeling → principled incorporation of governance intelligence. By explicitly quantifying these advances through graph-theoretic metrics and temporal validation, this study provides not only a methodological template but also demonstrable evidence base for its superiority over conventional approaches.

4.6. Computational Efficiency and Scalability of the PIM Framework

Given the computational demands noted in Section 4.5, we detail here the efficiency and scalability of the PIM framework, which are critical for its practical adoption. All analyses were conducted on a workstation with an Intel Core i7-12700K CPU (12 cores, 3.6 GHz), 32 GB RAM, and an NVIDIA GeForce RTX 3060 GPU, running Windows 11. The primary software environments were Python 3.9 (for IACO and general scripting) and ArcGIS 10.6 (for spatial analysis and circuit theory via Linkage Mapper).

For the Tongnan District case study (1583 km², 30 m resolution), the total runtime for the complete PIM workflow was approximately 6.5 h. The most computationally intensive component was the IACO algorithm, which accounted for ~3 h (~46% of total time), due to the iterative simulation of ant movement across 500 iterations for 100 ants. The circuit theory-based connectivity modeling using Linkage Mapper took ~1.5 h (~23%). The SOM clustering and MSPA analysis were relatively efficient, each completed within 30–45 min. It is important to note that the reported runtime pertains specifically to the automated computational core of the PIM framework. The entire research process, encompassing data curation, expert-informed parameterization (e.g., for the F-S-P integration), and the interpretation of results for spatial optimization, requires additional time and domain expertise, which is typical for applied geocomputational planning studies.

Scaling behavior is an important consideration. The runtime of the IACO and circuit theory components is expected to increase non-linearly with the number of ecological sources (nodes) and the spatial extent/complexity of the study area. A preliminary test applying the framework to a larger region (approximately 5000 km² within the CCEC) with comparable source density resulted in a runtime increase to approximately 28 h, highlighting the super-linear scaling. Future work will focus on algorithm parallelization (e.g., GPU-accelerated IACO) and employing more efficient circuit theory solvers to enhance performance for regional-scale applications.

Despite its computational demands, the PIM framework offers a tractable solution for municipal and regional planning units. The runtime is acceptable for strategic, non-real-time planning exercises. Users with limited computational resources could simplify the framework by reducing IACO iterations or ant count during exploratory phases, albeit with potential trade-offs in solution optimality.

4.7. Limitations and Future Research Directions

4.7.1. Key Limitations and Future Extensions

While the proposed PIM framework provides an integrated computational approach for node city ESP construction, several limitations offer avenues for future research:

(1)

Algorithm specialization

While the DFM provides a robust computational framework, its current application for corridor extraction remains generalized rather than species-specific. Incorporating species-specific traits (e.g., dispersal ability, habitat preference) into the IACO parameters would enhance ecological realism. Furthermore, empirical validation using GPS tracking or genetic markers for focal species (e.g., Procypris rabaudi) is crucial to corroborate modeled corridor usage. Extending such species-specific calibration to the construction of the ecological resistance surface—where we currently employ an equal-weight scheme as a functional baseline—would further refine the ESP.

(2)

Temporal dynamics and scenario analysis

Our temporal analysis (2000–2020) confirms the in situ stability of the identified ESP components under recent landscape dynamics. It should be noted, however, that this validation primarily affirms component stability within the study area and does not in itself establish spatial transferability. Furthermore, the ESP construction relied on a static snapshot (2020), limiting our ability to capture spatio-temporal co-evolution under processes like urban expansion. Future implementations with longitudinal time-series data and land-use change scenarios would enable predictive ESP planning.

(3)

Spatial transferability and robustness testing

The framework’s validation in a single district, while demonstrating temporal stability, does not establish its spatial transferability. Applying PIM to node cities in contrasting environments (e.g., arid basins vs. monocentric regions) is essential to test its robustness and refine its generalizability claims. Additionally, formal network robustness analysis (e.g., via simulated node/link removal) should be conducted to quantify the resilience of PIM-optimized networks.

(4)

Uncertainty quantification

The current PIM framework is deterministic, reporting integrated results as single best estimates. While we have employed a multi-faceted robustness assessment (parameter sensitivity, temporal validation, methodological comparison) to establish credibility, we do not provide probabilistic uncertainty measures (e.g., confidence intervals) for final composite metrics. Future iterations could explore integrating Bayesian approaches or stochastic ensemble modeling at key stages to propagate uncertainty through the entire workflow.

Concluding remark on limitations: These limitations—shared with emerging geocomputational frameworks in conservation planning—do not diminish PIM’s distinct contribution: demonstrating that proactive, policy-embedded geocomputation can achieve deterministic network improvements (20.5% α-index increase) with tractable computational costs (6.5 h), in contexts where existing tools either require unavailable data (Marxan targets) or produce scale-mismatched outputs (hierarchical frameworks).

4.7.2. Transferability Guidelines for Cross-Regional Application

While full validation across diverse contexts remains future work, the PIM framework is designed with modular components that enable systematic adaptation to new regions. We outline specific guidelines for three critical adaptation dimensions:

Data requirements and substitutions. The core input datasets—land use, DEM, NDVI, climate, and POI—are globally available (Table 1). Region-specific substitutions are feasible: (i) where POI data are unavailable, Nightlight intensity can proxy anthropogenic pressure; (ii) where species occurrence data are absent (as in Tongnan), the equal-weight resistance scheme provides a functional baseline; (iii) policy corridors (PIES) can be adapted to any region with designated ecological networks (e.g., EU Natura 2000, US National Wildlife Refuges).

Parameter transferability. IACO parameters (α = 1, β = 5, γ = 0.2) derive from canonical ACO literature and showed low sensitivity to ant count variation (Figure S12), suggesting robust transferability without case-specific tuning. However, iteration limits should scale with landscape complexity: we recommend 500 iterations for 1500 km² with ~20 sources, adjusting proportionally for larger domains.

Indicator calibration. ESB classification via SOM requires only ecosystem service rasters, with cluster number (k = 5) determined by Davies-Bouldin index minimization (Figure S1)—a data-driven procedure requiring no expert input. Network robustness thresholds (α > 0.45, γ > 0.65 for “resilient” classification) are derived from graph-theoretic theory and apply across comparable urban agglomeration contexts.

These guidelines position PIM as a transferable computational workflow rather than a context-specific solution. Immediate application is feasible for subtropical node cities with available land-use and climate data; extension to arid, tropical, or temperate zones requires validation of resistance surface parameters and species movement assumptions.

5. Conclusions

This study proposed a novel PIM for constructing cross-scale ESPs in node cities, addressing the persistent spatial mismatch between macro-regional strategies and local implementation. Applied to Tongnan District—a strategic node within the CCEC—the PIM integrated multi-source geospatial data and bio-inspired computation to generate an optimized ESP that aligns regional policy mandates with local ecological processes.

The F-S-P framework identified ecological sources covering 7405.77 hm², a 3.30-fold increase relative to conventional methods, while restoring the continuity of policy-mandated river corridors. The Dual-Feedback Mechanism—hybridizing circuit theory with an IACO algorithm—extracted 37 ecological corridors (221.7 km) and 25 barrier points, of which 16 corridors spatially coincided with the national “Six Rivers” network. Subsequent network optimization yielded a robust “One Core, One Ring, Dual Belts, Three Zones” configuration, wherein the Fujiang and Qiongjiang Rivers function as dual-purpose hubs: local anchors for Tongnan’s ESP and regional levers for cross-scale ecosystem service flows.

Temporal validation (2000–2020) confirmed the spatiotemporal persistence of the constructed ESP, evidenced by stable forest cover (>34%), increasing water bodies, and consistently high NDVI values. Graph-theoretic metrics revealed a deterministic enhancement in network robustness: the α-index (circuit redundancy) increased by 20.5%, the γ-index (topological connectivity) by 8.6%, alongside a marked decentralization of network structure—indicating a transition from vulnerable centralization to a resilient distributed topology.

The PIM framework thus provides a transferable computational paradigm for node cities in subtropical urban agglomerations, with modular components and explicit adaptation guidelines enabling systematic implementation across diverse data contexts, computationally bridging macro-strategy and local implementation through early-stage policy-biophysical integration. Future research should extend this framework to diverse node typologies, incorporate species-specific movement ecology, and rigorously test network robustness under dynamic disturbance scenarios.