Construction of Multi-Functional Composite Resilient Ecological Networks in High-Density Cities

Abstract

The rapid development of high-density cities has triggered severe ecological challenges, including habitat fragmentation, urban heat island (UHI) effects, and conflicting demands for public recreation. Traditional ecological networks (ENs) often focus only on “source” landscapes while neglecting degraded “sink” areas. This bias limits the ability of planners to resolve complex spatial conflicts. Therefore, the primary aim of this study is to develop a robust spatial planning framework that mitigates urban ecological conflicts and enhances regional resilience. To achieve this, we constructed a composite ecological network (CEN) for the high-density city of Guangzhou that harmonizes bird habitat conservation, thermal regulation, and cultural recreation. We combined the MaxEnt model, morphological spatial pattern analysis (MSPA), and circuit theory to identify functional “sources” and “sinks” across these three dimensions. Next, using complex network theory, we optimized the CEN and evaluated its structural robustness using low degree addition (LDA) and low betweenness addition (LBA) strategies. The results indicate the following: (1) The CEN effectively captured the complex mosaic landscape of the city. (2) Single-objective networks displayed distinct spatial differences—the recreational network formed a dispersed web of 242 corridors, while habitat and climate networks remained highly clustered. (3) The integrated CEN generated 1137 multi-layered corridors, creating a vital green skeleton to support species dispersal, mitigate UHI effects, and improve cultural access. (4) Optimization simulations verified that the LBA strategy provided the highest stability against targeted attacks by balancing network connectivity with local aggregation. Ultimately, this framework offers a highly adaptable planning tool for dense cities, providing precise spatial guidance to overcome ecological bottlenecks and harmonize urban growth with ecosystem resilience.

1. Introduction

Driven by global climate change and rapid urbanization, modern megacities face unprecedented socio-ecological challenges. Intensive development models have profoundly exacerbated regional human–land conflicts [1]. In high-density urban environments, these conflicts manifest as acute spatial trade-offs between nature conservation and intense human demands. For instance, expanding impermeable surfaces worsens urban heat island (UHI) effects while encroaching on fragile habitats, creating intense competition for limited land resources [2,3]. As China transitions toward high-quality economic development [4], balancing ecological conservation with urban expansion to forge a sustainable ecological security pattern remains a critical scientific challenge [5]. Grounded in landscape ecology, ecological networks (ENs) [6] optimize the spatial configuration of heterogeneous landscapes by reconnecting fragmented patches. By regulating ecosystem service flows, enhancing connectivity, and maximizing functional synergies [7], these networks serve as the backbone for safeguarding urban ecological security [8].

Since Forman [9] introduced the foundational “patch-corridor-matrix” paradigm, research has established a standardized EN construction framework: “source identification–resistance surface construction–corridor extraction–key node determination” [10]. This framework has driven the development of single-objective networks tailored to diverse functions, such as biodiversity conservation [11,12], linear heritage conservation [13], climate regulation [14], and green space recreation [15]. Common methods for identifying ecological sources include morphological spatial pattern analysis (MSPA) and ecosystem service evaluations [16]. To build function-specific resistance surfaces, researchers typically weight natural and socio-economic factors using expert scoring and the analytic hierarchy process (AHP) [17]. For corridor extraction, minimum cumulative resistance (MCR) models and circuit theory are frequently utilized to simulate species dispersal across heterogeneous landscapes [18]. Furthermore, researchers have advanced spatial planning by identifying corridor intersections as critical breakpoints [19]. Circuit theory is also applied to pinpoint ecological pinch points and barriers [20], providing a scientific basis for prioritizing restoration zones [21].

However, high-density cities operate as highly complex, dynamically evolving natural–social–economic systems [3]. Thus, relying on single-objective networks is insufficient to address multifaceted ecological conflicts. In recent years, research on composite ecological networks (CENs) has gained momentum. Rather than merely connecting isolated green spaces, a CEN integrates multiple socio-ecological dimensions, shifting the paradigm from single-function protection to multi-functional synergy. Examples include integrating ecological and recreational functions to design comprehensive urban ENs [5,22], or developing CENs that simultaneously account for biodiversity, hydrology, and human settlements [23]. Nevertheless, a critical limitation persists: current studies predominantly focus on connecting “source” landscapes [24], largely neglecting the analytical significance of “sink” landscapes. In reality, the ecological crises of high-density cities are intensely concentrated within these “sinks.” A traditional paradigm that connects “sources” while circumventing “sinks” fails to capture the spatial agglomeration of urban ecological deficits. Consequently, it cannot provide managers with the targeted spatial guidance required to overcome “ecological resistance bottlenecks” and implement precise micro-restoration.

“Source-sink” theory [25] originates from research on pattern–process relationships in landscape ecology. It categorizes heterogeneous landscapes into “sources”—which provide positive ecosystem services and facilitate ecological flows—and “sinks,” which act as service consumers that exhibit high demand, obstruct flows, or accumulate ecological issues. In an urban context, “sources” export ecological benefits, such as avian woodland habitats, cooling water bodies, or highly accessible cultural parks. Conversely, “sinks” represent degraded zones, including fragmented habitats, severe UHI areas, or recreational blind spots. While this theory excels at revealing complex ecological processes, its current application remains largely confined to identifying cold sources and heat sinks within single-objective climate networks [26,27]. Holistic integrations of “source-sink” dynamics across multiple ecosystem services remain scarce [24]. Therefore, embedding the “source-sink” framework into the construction of multi-functional CENs offers a crucial breakthrough for diagnosing and alleviating the deep-seated contradictions between ecological supply and demand in high-density environments.

Furthermore, quantitatively assessing and optimizing structural stability post-construction is crucial for safeguarding urban ecological resilience. To this end, researchers increasingly use complex network theory [16] to decode the spatial topological structures of ENs. By abstracting patches and corridors as nodes and edges, topological metrics (e.g., degree, global efficiency, betweenness), combined with simulated attack scenarios, enable the precise identification of network vulnerabilities [28]. Although this approach facilitates robustness evaluations and targeted edge-addition strategies [29], a systematic framework for optimizing multi-functional resilient networks remains notably absent. Specifically, coupling “source-sink” landscape patterns with complex network optimization tailored to high-density constraints requires urgent investigation to maximize the overall efficacy of composite green infrastructures.

As a global high-density megacity, Guangzhou faces acute ecological challenges, including habitat degradation, intensifying UHI effects, and conflicting recreational demands [30]. While existing local research predominantly focuses on single-objective networks—such as biodiversity conservation [12] or nature reserves [31]—it leaves a critical gap in multi-functional integration. To address this, the present study adopts a comprehensive “source-sink” framework to construct a CEN for Guangzhou, prioritizing avian habitat conservation, thermal regulation, and cultural recreation. Birds serve as vital bio-indicators of habitat quality, particularly given Guangzhou’s strategic position along the East Asian–Australasian Flyway (EAAF). Concurrently, the climate and recreation networks target UHI mitigation and resident well-being, respectively. In high-density urban contexts where green space is severely constrained, ecological patches must fulfill multiple roles. Thus, fostering functional synergy not only maximizes spatial efficiency but also mitigates complex socio-ecological spatial conflicts.

In summary, to address the deficiencies of existing single-objective networks and the prevalent neglect of “sink” areas, the primary objective of this study is to actively respond to multi-functional spatial conflicts. We aim to construct a resilient CEN for Guangzhou that effectively harmonizes habitat conservation, thermal regulation, and public recreation. Centered on this objective, this research addresses the following three core questions:

(1)

How can “source” and “sink” landscapes for avian habitat conservation, thermal regulation, and cultural recreation be scientifically defined and identified within intricate high-density urban environments?

(2)

How can the “source-sink” framework be leveraged to transcend the limitations of traditional single-target networks and construct a CEN that effectively mitigates spatial conflicts?

(3)

How can complex network theory be employed to quantitatively optimize the topological structure of this CEN and evaluate its structural robustness under various targeted disturbance scenarios?

To answer these questions, this research integrates MSPA, circuit theory, and complex network models to construct resilient green infrastructure that harmonizes habitat conservation, climate adaptation, and recreational well-being. Ultimately, this work aims to provide actionable spatial planning paradigms and precise governance strategies for resolving multifaceted ecological conflicts and advancing sustainable land-use management in high-density cities.

2. Study Area and Data Sources

2.1. Overview of the Study Area

Guangzhou (112°57′–114°03′ E, 22°26′–23°56′ N) is located at the northern margin of the Pearl River Delta, adjacent to the South China Sea. It is the political, economic, and cultural center of Guangdong Province, China. As a core hub of the Guangdong-Hong Kong-Macao Greater Bay Area (GBA), the city comprises 11 municipal districts covering approximately 7434 km² (see Figure 1). With a permanent population of 18.98 million in 2024, Guangzhou is a quintessential high-density megacity.

Topographically, the terrain descends from north to south toward the coast, featuring a dense network of rivers. This creates a unique landscape where mountains, land, and sea converge. Situated in the transition zone between northern temperate and tropical regions, Guangzhou has a subtropical maritime climate with hot, humid summers and mild, dry winters. The annual average temperature is 21–23 °C, and annual precipitation is roughly 2000 mm. Thanks to these natural conditions, Guangzhou supports highly diverse habitats. It also serves as a critical node along the EAAF, providing vital refuge for migratory species, particularly waterbirds, waders, and raptors.

However, rapid urban expansion has created severe socio-ecological challenges for the city:

(1)

Avian habitats: Urban sprawl has severely fragmented core habitats, such as northern woodlands and southern wetlands, thereby drastically reducing ecological connectivity.

(2)

Thermal environment: Dense building clusters and intense human activities have triggered severe UHI effects, highlighting the urgent need for ventilation corridors between cold sources and heat sinks.

(3)

Cultural recreation: The massive population generates immense demand for high-quality outdoor leisure spaces.

Together, these three demands cause intense spatial competition and land-use conflicts within a limited urban footprint.

Consequently, Guangzhou urgently needs to shift from traditional single-objective conservation models to a multi-functional CEN. This composite network must synergistically integrate avian habitat conservation, climate adaptation, and public recreation.

2.2. Data Sources

The data used in this study include (see

Table 1. List of data sources.

Data Type	Data Source	Resolution
Remote sensing imagery	The remote sensing imagery data for September 2022 is sourced from the United States Geological Survey (USGS).	30 m
Digital elevation model (DEM)	Geospatial data cloud (https://www.gscloud.cn, accessed on 26 March 2025)	30 m
Land use	The 2020 land use data is sourced from Globe Land 30.	30 m
Bird species distribution points	The 2022 survey data from 100 systematic transects originated from the Institute of Zoology, Guangdong Academy of Sciences, while citizen science species records were sourced from the China Birding Records Center. (http://www.birdreport.cn, accessed on 17 February 2024)	-
Cultural and recreational resource site	2022 List of Grade A Tourist Attractions in Guangzhou (Guangdong Provincial Department of Culture and Tourism), AutoNavi Map POI Data	-
Surface temperature	Inversion of remote sensing image data using ENVI software (version 3.4.4)	30 m
River system	Planning Center, Guangdong Provincial Institute of Water Resources and Hydropower Research	-
Road traffic	National Geographic Information Resource Catalog Service System	-
Annual average temperature	National Climate Data Center (https://www.ncei.noaa.gov/, accessed on 27 March 2025)	1 km
Annual average precipitation	National Climate Data Center (https://www.ncei.noaa.gov/, accessed on 27 March 2025)	1 km
Vegetation cover	Calculating using remote sensing imagery	30 m
Nighttime lighting	National Earth System Science Data Center (https://www.geodata.cn/main/, accessed on 19 April 2025)	1 km
Population density	European Commission Global Human Settlements Layer Dataset (https://human-settlement.emergency.copernicus.eu/, accessed on 19 April 2025)	100 m
Building height	European Commission Global Human Settlements Layer Dataset (https://human-settlement.emergency.copernicus.eu/, accessed on 19 April 2025)	30 m
Building density	China’s First Building Height Dataset [32]	10 m

): Landsat 8 remote sensing imagery, land use data, DEM, bird species distribution data, cultural and recreational resource data, and relevant data such as annual mean temperature, annual mean precipitation, vegetation coverage, nighttime light, river systems, road networks, population density, building height, and building density.

In order to unify the research and analysis scale, all data were projected and resampled using ArcGIS software (version 10.8). The unified projection coordinate system was WGS 1984 UTM Zone 49 N, the spatial resolution was 30 m, and the same administrative division boundaries were used for data clipping.

3. Methods

3.1. Research Framework

Table 2. Research framework.

Research Steps	Objectives	Main Analytical Processes	Core Outcomes and Outputs
Step 1: Identification of single-function “source-sink” landscape patches	Aims to spatially delineate the “source” areas of ecological service supply and the “sink” areas of demand for three core functions.	By comprehensively applying the MaxEnt model, MSPA, LST retrieval, Conefor landscape connectivity analysis, and hotspot analysis, core “sources” were extracted; furthermore, high-resistance hotspots were identified as “sinks.”	Single-function source and sink patches for bird habitat, thermal regulation, and cultural recreation.
Step 2: Construction of single-function resistance surfaces	To establish functionally heterogeneous resistance surface representing natural and anthropogenic barriers.	The AHP analysis method was employed to select, classify, and assign weights to multi-dimensional indicators (e.g., 11 items for bird habitat, 5 for the thermal environment, and 5 for recreation).	Three heterogeneous resistance surfaces tailored to each specific network function.
Step 3: Extraction of single-objective corridors and synthesis of multi-functional CEN	To extract independent networks and integrate them into a multi-functional framework that mitigates spatial conflicts.	Single-target pathways were extracted using circuit-theoretic model; patch centroids and corridor intersections were overlaid—in conjunction with hotspot analysis—to identify composite nodes; a composite resistance surface was generated through buffer analysis, and a composite “source-sink” network was constructed.	Constructing a multifunctional CEN (comprising composite “source-source,” “sink-sink,” and “source-sink” corridors).
Step 4: Analysis of static topological characteristics	To quantitatively assess the connectivity of different network structures and identify structural bottlenecks.	Utilizing Gephi software (version 0.10.1), patches were abstracted as nodes and corridors as edges to construct an undirected, scale-free network, and topological indices were calculated.	Baseline network evaluation metrics (Average degree, centralities, clustering coefficient, and shortest path length).
Step 5: Edge-addition optimization strategies	To direct topological optimization and forge redundant alternative pathways under realistic urban constraints.	Sort node degrees and betweenness centralities in ascending order, and use MATLAB software (version R2025b) to simulate the LDA and LBA edge-addition strategies under a 30% upper threshold.	Optimized composite “source-sink” landscape corridor network configurations.
Step 6: Robustness analysis of the CEN	To evaluate and compare network stability and resilience under external shocks pre- and post-optimization.	Simulated network attack scenarios (random node attacks, degree-based attacks, and betweenness-based attacks); evaluated the degradation rates of network efficiency and connectivity.	Network robustness curves tracking global network efficiency (E) and largest connected component (M) to verify the optimal strategy.

illustrates the overall technical workflow of this study, which comprised six main steps: (1) Source and sink identification: We used the MaxEnt model, MSPA, land surface temperature (LST) retrieval, Conefor, and hotspot analysis to identify functional “source” and “sink” patches across the three dimensions (bird habitat, thermal regulation, and cultural recreation); (2) Resistance surface construction: We selected appropriate ecological resistance factors to build functionally heterogeneous resistance surfaces for each dimension; (3) Single-objective network extraction: We applied circuit theory to map single-objective ecological corridors; (4) Multi-functional CEN synthesis: We integrated the composite nodes and resistance surfaces to generate the comprehensive multi-functional CEN; (5) Topological analysis: Using complex network theory, we abstracted the ecological networks into topological structures to quantitatively analyze their connectivity and complexity; (6) Optimization and robustness evaluation: We optimized the CEN structure utilizing low degree addition (LDA) and low betweenness addition (LBA) strategies, and subsequently evaluated its robustness by simulating both random and targeted attack scenarios.

3.2. Identification of Single-Function “Source-Sink” Landscape Patches

To resolve multi-functional spatial conflicts, we must first break down the complex urban ecosystem into specific functional dimensions. Based on the “source-sink” theory of landscape ecology [25,33], this study conceptualizes ecological conflicts as spatial mismatches between ecosystem service supply and demand. In this framework, “sources” are landscapes that support ecological flows and provide positive services in high-density cities. Conversely, “sinks” are landscapes that obstruct flows, cause negative impacts, or consume ecological functions. These include areas where poor habitat quality fails to sustain populations, urban heat continuously accumulates, or structural barriers cause persistent recreational deficits.

Therefore, sources act as ecological supply areas (e.g., core habitats, cold islands, cultural hotspots). Sinks represent demand areas where services are consumed, obstructed, or structurally absent (e.g., fragmented habitats, UHI hotspots, recreational deserts). Crucially, we identify sinks based on this ecological process, rather than simply extracting areas with high resistance values. Although demand-driven sinks often overlap with high-resistance areas due to intensive urbanization, they play a distinct analytical role. They represent spatially discrete clusters with severe, self-reinforcing service deficits that urgently require restoration. Recent multi-functional source-sink network studies in similar urban contexts have consistently adopted this supply-demand conceptualization [24,34,35]. Explicitly defining these sinks as discrete spatial nodes is essential. It allows us to construct directional source-to-sink corridors and set spatially bounded restoration targets—a goal that a continuous resistance surface alone cannot achieve.

3.2.1. Identification of “Source-Sink” Landscape Patches for Bird Habitat Conservation

In high-density cities, expanding impermeable surfaces cause severe ecological conflicts, primarily by encroaching on and fragmenting natural habitats. To address this, we selected birds as the target species for the habitat conservation network. Birds are highly sensitive bio-indicators of environmental quality and landscape connectivity. Furthermore, because Guangzhou is strategically located along the EAAF, prioritizing avian habitats is crucial for safeguarding both local biodiversity and regional ecological security.

Bird habitat “source” landscapes: These denote high-quality habitat patches that sustain avian reproduction and support population dispersal within complex urban environments. To accurately identify these sources, we used the MaxEnt model (version 3.4.4). Unlike traditional empirical scoring, MaxEnt is highly advantageous for urban ecological modeling because it relies strictly on presence-only data. Grounded in maximum entropy theory, the MaxEnt model [36] uses machine learning to evaluate the relationships between species occurrence records and their geographical environments. It infers the probability distribution (ranging from 0 to 1) of species presence based on limited known data. Using this model alongside avian distribution data, we selected environmental variables across three dimensions (climate, habitat, and human disturbance; see Table S1) to simulate and predict suitable avian habitats within the study area. The fundamental equation is expressed as follows:

$$H(p)=-{\sum }_{x\in X}P\left(x\right)\ln P\left(x\right)$$

(1)

where the study area was discretized into multiple pixel points x, collectively denoted as the set X. The entropy of the study area is expressed as H(P), where P(x) represents the probability distribution of the pixel x.

We imported the simulation results into ArcGIS and classified them into four categories using the natural breaks (Jenks) method: non-suitable, low-suitable, moderately suitable, and high-suitable areas. Next, we assigned habitat patches in the moderate- and high-suitability zones as foreground (value = 2) and designated the remaining categories as background (value = 1) to generate a binarized raster image. We then processed this image in Guidos Toolbox using an eight-neighbor analysis. Based on established thresholds for ecological source extraction [37] and the specific conditions of our study area, we set a core area threshold of ≥70 hm². This threshold retains the vast majority of effective core areas while keeping the number of patches manageable. We further evaluated these extracted patches using Conefor 2.6 software to calculate the Probability of Connectivity (PC) and the Integral Index of Connectivity (IIC). Based on avian movement ranges and regional landscape characteristics, we set the optimal dispersal distance threshold at 10 km with a connectivity probability of 0.5. Ultimately, we identified ecological patches with a dPC ≥ 0.1 as “source” areas for avian habitat conservation. These areas primarily consist of woodlands and wetlands. They experience minimal human disturbance and serve as vital migratory corridors for diverse bird species [12].

Bird habitat “sink” landscapes: These represent clustered, high-risk disturbance zones that negatively impact avian survival in high-density cities. According to classical source-sink population dynamics theory [33], a habitat “sink” is not merely a physical barrier. Rather, it is a low-quality area where the mortality rate exceeds the reproduction rate, causing continuous population depletion. Under intense urbanization, these “sink” landscapes carry a high risk of becoming “ecological traps” [38]. Typical examples include hardened urban edges prone to predator aggregation, severely polluted industrial zones, and contiguous built-up areas subjected to frequent human disturbance. All these environments are highly detrimental to species survival. Based on this premise, we first constructed a comprehensive habitat resistance surface. We then used hot spot analysis (Getis-Ord Gi*) to extract high-resistance hot spots as habitat “sinks.” Corresponding to our selected resistance factors, these “sink” areas are primarily characterized by high building density, elevated nighttime light indices, dense populations, and high landscape ecological risks (

Table 3. Bird migration resistance factors and weights.

Resistance Factors	Weight	Indicator Grading	Resistance Value
Landscape ecological risk	0.18	High risk (>0.07)	80
		Higher risk (0.04–0.07)	60
		Moderate risk (0.03–0.04)	40
		Lower risk (0.02–0.03)	20
		Low risk (<0.02)	1
Landscape types	0.16	Construction land	80
		Bare ground	60
		Arable land	40
		Grassland, shrub	20
		Woodland, wetland, water bodies	1
Elevation	0.04	>800 m	80
		600–800 m	60
		400–600 m	40
		200–400 m	20
		<200 m	1
Slope	0.03	>35°	80
		25–35°	60
		15–25°	40
		8–15°	20
		0–8°	1
Vegetation coverage	0.14	0–0.2	80
		0.2–0.4	60
		0.4–0.6	40
		0.6–0.8	20
		0.8–1.0	1
Water body buffer distance	0.06	>1500 m	80
		800–1500 m	60
		500–800 m	40
		200–500 m	20
		<200 m	1
Road buffer distance	0.07	>1500 m	80
		800–1500 m	60
		500–800 m	40
		200–500 m	20
		<200 m	1
Normalized nighttime light index	0.08	>0.3	80
		0.1–0.3	60
		0.04–0.1	40
		0.01–0.04	20
		<0.01	1
Population density	0.06	>2000	80
		1000–2000	60
		500–1000	40
		50–500	20
		<50	1
Building height	0.08	>100 m	80
		28–100 m	60
		24–28 m	40
		10–24 m	20
		<10 m	1
Building density	0.1	>60	80
		45–60	60
		35–45	40
		15–35	20
		<15	1

, Figure 2a).

$$G_i^{\ast }={\displaystyle \frac{{\sum }_{j=1}^n{W}_{i,j{x}_j}-\overline{x}{\sum }_{j=1}^n{W}_{i,j}}{S\sqrt{[n{\sum }_{j=1}^n{W}_{i,j}^2-}{\left({\sum }_{j=1}^n{W}_{i,j}\right)}^2]/(n-1)}}$$

(2)

$$\overline{X}=\left({\sum }_{j=1}^n{x}_j\right)/n$$

(3)

$$S=\sqrt{{\displaystyle \frac{{\sum }_{j=1}^n{x}_j^2}{n}}-{\left(\overline{X}\right)}^2}$$

(4)

where xj is the comprehensive ecological resistance value of landscape unit j; W(i,j) is the spatial weight between landscape units i and j; $\overline{X}$is the mean comprehensive ecological resistance value; and n is the total number of landscape units.

3.2.2. Identification of “Source-Sink” Landscape Patches for Thermal Environment Regulation

The conflict between dense built environments and human thermal comfort is another major ecological crisis in megacities. Therefore, mitigating the UHI effect is a central component of our composite framework. We adopted the ecosystem services supply-demand framework and landscape source-sink theory to define the nodes of the thermal regulation network. We defined patches that provided and exported UHI mitigation services as “cold sources.” Conversely, we defined high-temperature built-up areas that consumed these services and exhibited high cooling demand as “heat sinks.” This network-based “cold source-heat sink” conceptualization aligns strongly with recent advanced urban thermal mitigation network studies [39].

Cold island “source” landscapes: These are patches that possess a cooling effect and export cold air to their surroundings. Using the Landsat 8 OLI/TIRS Collection 2 Level 2 dataset (acquired during Guangzhou’s high-temperature summer in September 2022), we retrieved the LST using ENVI 5.6 software. We identified landscapes with temperatures below the regional average as cold island patches, because these areas help mitigate the urban thermal environment. We used the mean-standard deviation method [40] to classify the thermal environment into seven levels (Table S2). Next, we extracted the sub-low, low, and extremely low-temperature zones as foreground elements for the MSPA. Finally, we designated highly connected regions within these core areas as cold “source” landscapes. These primarily consist of cooling spaces like woodlands and water bodies.

Thermal environment “sink” landscapes: These are areas that continuously generate and accumulate heat, severely obstructing or dissipating the cold island effect. We applied the Getis-Ord Gi* statistic to extract heat island spaces with high thermal regulation resistance. Spatially, these areas correspond to contiguous built-up zones characterized by high impervious surface indices, high building density, and scarce vegetation cover (

Table 4. Cold source diffusion resistance factors and weights.

Resistance Indicator	Weight	Indicator Grading	Resistance Value
Building density	0.3	>60	80
		45–60	60
		35–45	40
		15–35	20
		<15	1
NDVI	0.25	0–0.2	80
		0.2–0.4	60
		0.4–0.6	40
		0.6–0.8	20
		0.8–1.0	1
ENDISI	0.2	Impermeable surface	80
		Other types	1
Elevation	0.15	<200 m	80
		200–400 m	60
		400–600 m	40
		600–800 m	20
		>800 m	1
Slope	0.1	>35°	80
		25–35°	60
		15–25°	40
		8–15°	20
		0–8°	1

, Figure 2b). To minimize the fragmentation of these “sink” areas, we conducted a landscape connectivity analysis on the 100 largest patches. We ultimately extracted the regions demonstrating the highest connectivity levels as thermal “sinks.”

3.2.3. Identification of “Source-Sink” Landscape Patches for Cultural Recreation

In addition to natural conservation, the spatial mismatch between dense urban populations and accessible recreational green spaces creates a significant socio-ecological conflict. Fostering equitable cultural recreation is essential for urban resilience.

Cultural recreation “source” landscapes: These are patches endowed with high aesthetic value and social service functions that attract visitors and stimulate recreational behaviors. We used the MaxEnt model to simulate the spatial distribution patterns of cultural ecosystem services (CES) [41] across six dimensions: landscape aesthetics, leisure and recreation, cultural heritage, science education, religious/spiritual, and health/healing. Initially, we selected 88 A-level and above tourist attractions as sample plots from the 2022 List of A-level Tourist Attractions in Guangzhou. Through expert consultation, we invited professionals and professors from fields such as landscape architecture and urban planning to classify the CES functions of these attractions via a questionnaire. This established our foundational classification for recreational resources (Table S3).

Given the precise geolocation and clear categorization advantages of point of interest (POI) data [42], we incorporated big data analytics to uncover potential cultural and recreational sites in Guangzhou. After cleaning the data, we obtained 9762 classified recreational resource occurrence points (Table S3). Next, we used the MaxEnt model to simulate and predict the six CES functions individually. Based on the study area’s actual conditions, we selected environmental raster data—comprising natural and human well-being indicators—to reflect the spatial relationships between cultural recreation sites and their geographical environments (Table S4). We then weighted and overlaid these individual simulation results to derive the comprehensive cultural service value of the study area, categorizing it into seven levels using the natural breaks (Jenks) method. We extracted patches within the moderately high-, high-, and extremely high-value zones for the MSPA. Ultimately, we designated highly connected core areas as recreational “source” landscape patches, which represent critical hotspots for historical and cultural resources.

Cultural recreation “sink” landscapes: These primarily refer to high-resistance spaces that hinder public access to recreational services. Based on the hot spot analysis results, we applied the Getis-Ord Gi* statistic to identify spatial clusters with high comprehensive recreational resistance. To extract the specific recreational “sink” areas, we conducted a landscape connectivity analysis on the 100 largest patches within these clusters. These areas feature a low density of recreational resources, resulting in diminished recreational attractiveness and accessibility (

Table 5. Recreation resistance factors and weights.

Resistance Indicator	Weight	Indicator Grading	Resistance Value
Accessibility of recreational patches	0.2	>2000 m	80
		2000 m	60
		1500 m	40
		1000 m	20
		500 m	1
Road classification	0.15	Non-road area	80
		High way	60
		Expressway	40
		Secondary roads, branch roads	20
		Main road	1
The cultural service value of the path	0.2	1–2	80
		3–4	60
		5–6	40
		7–8	20
		9–10	1
Density of scenic spots along the path	0.15	1	80
		2	60
		3	40
		4	20
		5	1
Water buffer distance	0.3	>600 m	80
		600 m	60
		300 m	40
		100 m	20
		50 m	1

, Figure 2c). The physical and psychological barriers formed by these high-resistance landscapes drastically reduce recreational service flows, rendering them typical process-blocking “sink” areas.

3.3. Construction of Resistance Surfaces for Single-Function ENs

To construct the resistance surfaces for each single-function EN, this study integrated AHP with expert consultation. This approach determined the resistance weights and classification criteria for various spatial factors. The consistency ratio (CR) of all judgment matrices was strictly maintained below 0.1. The specific resistance factor classification standards and weight assignments for each single-function network are detailed in the following sections.

3.3.1. Construction of the Resistance Surface for the Bird Habitat Conservation Network

Referring to relevant empirical studies [43,44] and comprehensively considering the actual conditions in Guangzhou, 11 resistance factors—including the landscape ecological risk index, landscape types, elevation, and slope—were selected as resistance indicators for the avian habitat conservation network. These indicators characterize the impacts of ecological risks, the natural environment, and anthropogenic disturbance factors on the habitat conservation network in the study area. All indicators were subjected to classification, value assignment, and weight analysis (see Table 3).

3.3.2. Construction of the Resistance Surface for the Thermal Environment Regulation Network

Drawing upon relevant research [26] and considering the specific conditions of the study area, five factors were selected as indicators to evaluate the thermal environment within the high-density city. These factors encompass building density, the normalized difference vegetation index (NDVI), the enhanced normalized difference impervious surface index (ENDISI) [45], elevation, and slope. Collectively, these variables effectively characterize the built environment, land cover types, and topography of the study area (see Table 4).

3.3.3. Construction of the Resistance Surface for the Cultural Recreation Network

Drawing upon relevant research [22] and considering the specific context of this study, five resistance factors were selected to construct the comprehensive resistance surface for cultural recreation within the study area (see Table 5). These factors encompass recreational patch accessibility, road hierarchy, CES value along the routes, attraction density along the routes, and river buffer zones. Collectively, these indicators effectively characterize the accessibility of cultural recreation nodes, the inherent recreational value of the pathways, and the potential for realizing recreational value.

3.4. Construction of Single-Function “Source-Sink” ENs

Grounded in circuit theory [46], the Linkage Mapper toolbox (version 3.0) was employed to identify and extract multi-layered ENs. Unlike traditional MCR models that only identify a single optimal path, circuit theory simulates the random walk characteristics of species and ecological flows. This stochastic approach is highly suited for high-density cities, as it identifies multiple alternative pathways and ecological pinch points within highly fragmented landscapes. These networks specifically comprised “source-source,” “sink-sink,” and “source-sink” corridors, which connected the respective landscape patches within each single-function network [24].

3.5. Construction of Multi-Functional “Source-Sink” CENs

Following the construction of single-function networks, a multi-functional CEN was developed through multi-criteria integration and spatial overlay analysis. This CEN aimed to harmonize spatial conflicts among diverse ecological functions. This goal was achieved by identifying multi-functional composite nodes and integrated resistance surfaces, ultimately enhancing the overall resilience and service efficiency of the urban ecosystem.

3.5.1. Identification of Composite “Source-Sink” Landscape Nodes

The multi-functional CEN consists of several interconnected layers, including composite nodes, composite resistance surfaces, and composite corridors [5]. To identify composite “source-sink” landscape nodes, we first overlaid the “source-sink” patches of different functions to extract centroids representing dual-functionality. Subsequently, we extracted the intersections of ecological corridors across various functions. After removing redundant points, we used Getis-Ord Gi* hot spot analysis to identify multi-functional hotspots. We focused on areas with a Gi Z-score > 0, Gi p-value < 0.10, and Gi_Bin ≥ 2. Furthermore, the intersection zones of the three primary source types formed multi-functional composite node areas. We extracted the centroids of these zones and integrated them with the previously identified nodes to form the final composite “source-sink” landscape nodes for this study.

Among these, multi-functional landscape “sources” represent regions that provide integrated ecosystem services. However, these composite “sources” do not always exhibit functional synergy. They may also harbor spatial conflicts, such as trade-offs between low-disturbance habitats and high-frequency recreational activities. Conversely, multi-functional “sinks” represent composite regions with prominent ecological problems and an acute demand for ecosystem services.

3.5.2. Construction of Composite Resistance Surfaces and Extraction of Multi-Functional Corridor Networks

Based on the identification of single-function “source-sink” corridors, the corridor buffer zone method was utilized to generate buffers of <30 m, 30–50 m, 50–80 m, and >80 m, representing resistance thresholds of 1, 20, 40, 60, and 80, respectively. These buffers were then coupled and overlaid with the integrated resistance surfaces of the three networks to derive the composite resistance surface. On this basis, the Linkage Mapper toolbox was employed to identify and extract composite “source-source,” “sink-sink,” and “source-sink” ecological corridors. By integrating multi-functional landscape nodes and corridor networks, the efficiency of energy flow between “source-sink” landscapes in high-density cities is significantly enhanced.

3.6. Comprehensive Evaluation and Optimization of the Multi-Functional “Source-Sink” CEN

Following the spatial construction of the multi-functional CEN, quantitatively evaluating its structural characteristics and implementing targeted optimizations were critical steps to ensure the overall resilience of the urban ecosystem. By integrating complex network theory, the spatial connectivity and vulnerability of the network were initially revealed through static topological metrics. Subsequently, the network structure was directionally optimized by utilizing LDA and LBA strategies. Finally, by simulating random and deliberate attack scenarios, the robustness of the network pre- and post-optimization was quantitatively assessed, providing a scientific basis for the efficient construction and refined management of ENs in high-density cities.

3.6.1. Analysis of Static Topological Characteristics of the Complex Network

As a complex network, the “source-sink” EN indirectly influences ecological flows; thus, analyzing its topological structure reveals the network’s inherent attributes and relational characteristics. In this study, the EN was represented as an undirected and scale-free network, where source patches and corridors were abstracted as nodes and edges, respectively. With default parameter settings, Gephi (version 0.10.1) was utilized to calculate topological metrics—including average degree, closeness centrality, betweenness centrality, average clustering coefficient, and average shortest path length (see

Table 6. “Source-sink” corridor network topological structure evaluation indicators.

Indicator Name	Definition	Formula	Note
Average degree [47]	The importance of a node reflects the number of connections to the node.	$K={\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{k}_i$	Where N is the total number of nodes in the network, and ki is the number of edges directly connected to node i.
Closeness centrality [48]	Reflects the proximity of a node to other nodes in the network.	$C_i={\displaystyle \frac{n - 1}{{\sum }_{j\ne i}{d}_{\mathit{ij}}}}$	Where dij represents the shortest path length between node i and node j.
Betweenness centrality [49]	Measures the degree to which a node plays the role of an information hub in the network.	${\mathit{BC}}_i={\displaystyle \sum _{k=0}^n}s \ne i \ne t{\displaystyle \frac{{\sigma }_{\mathit{st}}i}{{\sigma }_{\mathit{st}}}}$	Where σst is the total number of shortest paths from node s to node t, and σsti is the total number of paths through node i among the shortest paths of these nodes.
Average clustering coefficient [50]	The degree of aggregation of network nodes reflects the frequency of ecological information transmission between nodes.	$C_i={\displaystyle \frac{2L}{k_i(k_i - 1)}}$	Where Ci is the clustering coefficient, ki is the actual number of edges connecting node i, and L is the actual number of edges from node i to adjacent nodes.
Average shortest path length [51]	Measures the average distance of all nodes in the network and reflects the tightness of the network connection.	$L={\displaystyle \frac{1}{N(N - 1)}}{\displaystyle \sum _{i\ne j}}{d}_{\mathit{ij}}$	Where dij is the shortest path length from node i to j.
Eigenvector centrality [52]	Reflects the importance of connected nodes in the network.	$\mathit{EVC}\left(i\right)={\displaystyle \frac{1}{\lambda }}{\sum }_{\mathit{Aij}}\mathit{EVC}\left(j\right)$	Where $\lambda$is a constant (eigenvalue), Aij is an element in the adjacency matrix, indicating whether nodes i and j are connected.
Global network efficiency [53]	Measures the efficiency of information transmission between nodes in the network. The higher the value, the higher the transmission efficiency.	$E_{\mathit{glob}}={\displaystyle \frac{1}{N(N - 1)}}{\sum }_{i\ne j}{\displaystyle \frac{1}{d_{\mathit{ij}}}}$	Where N is the total number of network nodes, dij is the shortest path length from node i to j.

)—to evaluate and compare the topological structural characteristics of various single-function “source-sink” corridor networks and the multi-functional CEN.

3.6.2. Edge-Addition Optimization Strategies for the Composite Corridor Network

Complex network theory [54] optimizes topological structures primarily through four methods: edge deletion, edge rewiring, edge addition, and edge orientation. Because our EN is undirected and our circuit theory model establishes connections via least-cost paths, edge deletion and rewiring risk disrupting the established corridor networks. Therefore, we applied edge-addition strategies—specifically the LDA and LBA strategies—using MATLAB (version R2025b) to optimize the multi-functional CEN. We proceeded as follows:

First, the node degree and betweenness centrality of the network were ranked, followed by the application of the two edge-addition strategies:

(1)

LDA strategy: The degree values of all nodes were counted and sorted in ascending order, prioritizing the connection of nodes with the lowest degrees.

(2)

LBA strategy: The betweenness centrality values of all nodes were counted and sorted in ascending order. Edges were then added between the two unconnected nodes possessing the lowest betweenness centrality.

Given the practical difficulties of constructing urban ecological corridors, we set the target for added edges at 30% of the existing composite “source-sink” corridors [54]. We terminated the edge-addition process once the newly added edges reached this 30% threshold, yielding the final optimized composite “source-sink” landscape corridor network. We selected this threshold based on commonly used edge-addition ratios from previous network optimization studies [53,54]. Reaching a 30% addition threshold significantly improves the network structure while preventing excessive redundancy and prohibitive planning costs.

3.6.3. Robustness Analysis of Composite Corridor Network

Network robustness refers to the ability of a composite corridor network to maintain its normal structure and function after the removal of nodes or edges [55]. To evaluate the stability and efficiency of ecological flows within the network, we employed random and deliberate attack scenarios [56]. Specifically, we simulated attacks on random nodes, high-degree nodes, and high-betweenness nodes within the composite “source-sink” network, both before and after optimization. We used two primary indicators to measure the network’s robustness [53].

Global network efficiency (E): This metric reflects the operational efficiency of the complex network. Higher efficiency means that nodes require fewer resources to communicate, which translates to a higher transmission efficiency of information and ecological flows across the network. Largest connected component (M): As external shocks degrade network nodes and edges, the network’s integrity may decline, causing it to fragment into disconnected sub-networks and ultimately collapse [16]. The metric M illustrates the impact of these external shocks and effectively reflects the network’s overall resilience [53]. The calculation formulas are as follows:

$$\mathrm{M} = {\displaystyle \frac{{\mathrm{N}}^{\prime }}{\mathrm{N}}}$$

(5)

where N′ is the number of nodes in the largest connected subgraph, and N is the total number of nodes in the EN.

4. Results

4.1. Spatial Patterns of Single-Function “Source-Sink” Landscape Patches and ENs

4.1.1. Spatial Distribution Characteristics of Single-Function “Source-Sink” Landscape Patches

Figure 3 illustrates the spatial distribution of the “source-sink” landscape patches across the three functional dimensions. Specifically:

Bird habitat conservation “source-sink” landscapes: We identified 72 “source” patches (1351.57 km²) and 31 “sink” patches (597.33 km²). Spatially, the most extensive and highly connected “sources” were concentrated in the northeastern woodlands, an area with minimal human disturbance. In the central-eastern and southern regions, green spaces and wetlands provided moderately connected but crucial roosting and breeding grounds. Although the “sources” in the central urban core were smaller and fragmented, they functioned as indispensable “stepping stones” for migratory birds. Conversely, “sink” patches clustered predominantly in intensively developed areas with frequent human activity. Driven by high ecological risks, these dense zones represent priority targets for habitat restoration.

Thermal environment regulation “source-sink” landscapes: LST retrievals revealed a stark spatial dichotomy. We identified 55 cooling “sources” (2133.04 km², accounting for 29.49% of the study area), primarily anchored in the northeastern and central woodlands, as well as the southern Pearl River system. With their low impervious surface coverage, these expanses act as substantial cooling engines for the megacity. In stark contrast, 84 thermal “sinks” (969.80 km²) clustered tightly within the central urban core and southern construction zones. Constrained by dense built environments, these areas exhibit weak thermal regulation, underscoring the urgent need to establish “sink-sink” or “source-sink” ventilation corridors.

Cultural recreation “source-sink” landscapes: We found 98 recreational “sources” (218.77 km²) clustered primarily in highly accessible historical and central districts (e.g., Yuexiu, Tianhe, and Haizhu). Crucially, these hubs spatially intersect with bird habitat and thermal “sources” in the northeastern region. Ecologically, they act as vital socio-ecological buffers that absorb intense human recreation, thereby shielding fragile natural habitats from unregulated anthropogenic pressure. Conversely, the 85 recreational “sinks” (58.87 km²) concentrated predominantly in the southern periphery (e.g., Nansha), severely overlapping with habitat and thermal “sinks.” This spatial overlap highlights acute socio-ecological conflict hotspots: a structural lack of recreational infrastructure drives disorderly human encroachment into vulnerable areas, further exacerbating habitat degradation and thermal stress.

4.1.2. Spatial Patterns of Single-Function “Source-Sink” ENs

Figure 4 illustrates the construction results of the single-function “source-sink” landscape corridor networks. Specifically:

Bird habitat conservation corridors: We extracted 180 “source-source” migratory corridors (1094.03 km). Anchored primarily in the north, these ecological arteries bolstered population connectivity among core habitats, serving as crucial pathways for migration, reproduction, and genetic exchange. In contrast, 70 “sink-sink” corridors (659.17 km) traversed the central-southern regions. This subsidiary network connected marginal habitats and transitional dwelling environments. Establishing ecological buffer zones within these high-resistance areas could effectively facilitate temporary avian foraging and transit. Bridging the gap, 266 “source-sink” corridors (1178.33 km) formed a clustered distribution. By linking high-quality core habitats with marginal zones, these vital routes guided avian dispersal outward. Consequently, strictly controlling human disturbances along these transit lines is essential to enhance overall ecosystem connectivity and conservation efficiency.

Thermal environment regulation corridors: We identified 114 climate “source-source” corridors (522.38 km) as primary ventilation channels. These predominantly connected peripheral cold-source areas, such as urban woodlands and major water bodies. Conversely, 203 “sink-sink” corridors (417.90 km) clustered tightly within heat-sink regions, notably the dense urban construction lands. Functioning as inefficient ventilation zones or thermal “grey spaces,” these fragmented paths represent priority targets for remediating thermally vulnerable neighborhoods. Most critically, 337 climate “source-sink” corridors (673.37 km) acted as active heat-exchange conduits. By bridging cold sources with heat island regions, they formed indispensable channels for transferring fresh, cold air from the urban periphery into the overheated central core, thereby playing a pivotal role in mitigating the UHI effect.

Cultural recreation corridors: We mapped 242 “source-source” landscape corridors (1727.61 km) that formed a dense web in the central study area. They successfully interconnected highly accessible historical and cultural districts, such as Yuexiu and Tianhe. Further integrating regional resources along these axes could foster a continuous, clustered cultural tourism experience. Meanwhile, 209 “sink-sink” corridors (822.09 km) clustered across the southern expanse. By delineating pathways through regions with weak cultural vitality and high recreational resistance, these corridors provide spatial blueprints for developing future cultural infrastructure. Finally, the most extensive network consisted of 468 “source-sink” corridors (2161.17 km), radiating from core resource areas out to deficient zones. This outward connectivity strategically integrates regions with historically low recreational attractiveness into the broader, city-wide cultural network.

Overall, the divergent spatial distributions of these single-function networks closely align with Guangzhou’s typical “mountain-water-city” development pattern. The continuous mountains in the north preserve intact habitats and stable cold-island sources, while the central built-up areas form a dense network of highly concentrated recreational landscapes.

4.2. Construction Results of the Multi-Functional “Source-Sink” CENs

4.2.1. Spatial Distribution of Composite "Source-Sink" Landscape Nodes

Constructing a CEN by merely superimposing multiple single-objective networks would yield an overly complex structure and incur prohibitive spatial planning costs. Therefore, we overlaid the “source-sink” landscape patches of the three functions and extracted their centroids. Through this process, we initially identified 18 composite “source” nodes (primarily in the northern forest areas, Figure 5b) and 35 composite “sink” nodes (heavily concentrated in the southern region, Figure 5c). These spatial intersections form composite nodes with multi-dimensional attributes.

While composite “sources” serve as epicenters for integrated ecosystem service supply, composite “sinks” act as concentrated zones of multiple ecological deficits. However, this high degree of spatial overlap inevitably triggers multi-functional spatial conflicts. While accommodating high ecological demands, these composite nodes face compatibility dilemmas and spatial resource trade-offs. For example, there is an inherent contradiction between the low-disturbance requirements of habitat conservation and the high-accessibility demands of cultural recreation.

To further refine the network, we analyzed the pairwise intersections of the functional corridor networks and screened them using hot spot analysis. Ultimately, we identified a final set of 101 composite “source” nodes and 35 composite “sink” nodes within the corridor intersection zones. Distributed predominantly across the central urban area (Figure 5a), these critical nodes represent composite “sources” providing integrated services, alongside composite “sinks” burdened by severe ecological issues and acute service demands.

4.2.2. Spatial Patterns of the Multi-Functional “Source-Sink” CENs

By integrating the resistance surfaces of the three individual networks and employing circuit theory, the multi-functional CEN was successfully delineated.

For the primary ecological connections, 350 composite “source-source” corridors (spanning 2721.46 km) were established. Anchored primarily in the northeastern expanse of the study area, these corridors weave together high-value habitats, cold island zones, and cultural attractions. Collectively, they forge a continuous urban green skeleton that simultaneously supports species dispersal, mitigates the UHI effect, and provides cultural slow-mobility pathways (Figure 5b). Conversely, 210 composite “sink-sink” corridors (1364.07 km) were extracted to link ecologically degraded areas. By pinpointing and connecting regions plagued by severe avian habitat fragmentation, intense thermal stress, and cultural resource scarcity, these pathways establish a crucial spatial planning foundation for delineating urban resilience restoration belts and cultural potential renewal zones (Figure 5c). Finally, acting as the overarching integration mechanism, 577 composite “source-sink” corridors (3939.07 km) bridged the ecologically superior and vulnerable regions. Endowed with the synergistic functions of species dispersal, heat island mitigation, and cultural penetration, these conduits significantly enhance urban ecological resilience, thermal comfort, and cultural equity. Ultimately, the construction of this multi-functionally integrated green infrastructure culminates in a composite urban landscape paradigm that seamlessly embodies ecological security, environmental livability, and cultural diversity (Figure 5a).

4.3. Comprehensive Evaluation and Optimization Results of the Multi-Functional “Source-Sink” CENs

4.3.1. Static Topological Characteristics of the Identified Complex Network

To quantitatively evaluate the structural connectivity and coordination across the different functional networks, we selected seven core topological metrics, including average degree, global network efficiency, average clustering coefficient, and betweenness centrality. We imported the adjacency matrices for both the single-function and composite “source-source,” “sink-sink,” and “source-sink” corridor networks into Gephi software (version 0.10.1) for calculation.

Table 7. Evaluation metrics for the static topological structure of the complex networks.

Network Type	Indicator Name
	Average Degree	Eigenvector Centrality	Global Network Efficiency	Average Clustering Coefficient	Average Shortest Path Length	Closeness Centrality	Betweenness Centrality
Habitat “source” network	5.070	0.106	0.345	0.514	3.671	0.277	0.038
Climate “source” network	4.145	0.096	0.328	0.447	3.751	0.301	0.048
Recreation “source” network	4.938	0.074	0.285	0.469	4.600	0.222	0.037
Composite “source” network	5.982	0.069	0.293	0.483	4.396	0.232	0.029
Habitat “sink” network	4.516	0.164	0.457	0.486	2.741	0.372	0.060
Climate “sink” network	4.833	0.080	0.413	0.565	2.808	0.368	0.022
Recreation “sink” network	4.917	0.089	0.313	0.467	4.111	0.248	0.037
Composite “sink” network	6	0.098	0.364	0.508	3.521	0.288	0.037
Habitat “source-sink” network	5.215	0.086	0.302	0.517	4.235	0.240	0.032
Climate “source-sink” network	4.848	0.056	0.340	0.542	3.319	0.320	0.016
Recreation “source-sink” network	5.114	0.051	0.221	0.456	6.085	0.167	0.028
Composite “source-sink” network	6.171	0.052	0.253	0.480	4.992	0.203	0.021

and Figure 6 summarize the results.

Within the “source” networks, the multi-functional CEN achieved the highest average degree (5.982). This significantly outperformed the single-function networks for avian habitat (5.070), thermal regulation (4.145), and cultural recreation (4.938). This result indicates that multi-functional integration creates denser connection pathways, thereby enhancing spatial connectivity potential. However, the composite network’s average clustering coefficient (0.483) and global network efficiency (0.293) were lower than those of the habitat “source” network. This disparity suggests that within the constrained space of high-density cities, multi-functional integration forces nodes to compromise across diverse functional thresholds. Consequently, this structural discreteness weakens local aggregation and reduces transmission efficiency. Similarly, while the composite “sink” network exhibited the highest average degree (6.000), its clustering coefficient (0.508) and global efficiency (0.364) fell short of the optimal levels found in the single-function networks. Ultimately, these findings highlight the inherent challenges of coordinating micro-structures while pursuing macro-connectivity enhancement.

The comprehensive “source-sink” network further validated this inherent structural trade-off. Although the composite “source-sink” network achieved an exceptionally high connection density (average degree: 6.171), its global efficiency (0.253) remained suboptimal, accompanied by a prolonged average shortest path length (4.992). Crucially, the network exhibited remarkably low betweenness centrality (0.021), signifying a highly decentralized and balanced topological architecture. This distinct configuration implies an absence of absolute dominant bottleneck nodes, allowing for functional ecological flows to distribute uniformly across multiple redundant pathways.

Overall, while the composite network enhances functional overlay and connectivity, there remains a critical need to optimize its local coordination and efficiency structures.

4.3.2. Network Optimization Effects Under Different Edge-Addition Strategies

To structurally optimize the composite corridor networks, the LDA and LBA strategies were independently applied, with the edge-addition threshold capped at 30% of the existing corridors. This yielded 105, 63, and 173 newly simulated edges (via the MCR model) for the composite “source-source,” “sink-sink,” and “source-sink” networks, respectively (Figure 7). Spatially, these two strategies revealed distinctly divergent optimization pathways. The LDA strategy effectively shortened the average path length by linking isolated peripheral nodes, thereby maximizing global transmission efficiency. Conversely, the LBA strategy focused on reinforcing “bridge” patches characterized by lower betweenness centrality. By weaving critical redundant pathways, it successfully bolstered the network’s topological robustness against targeted attacks without severely compromising local aggregation.

Subsequent calculations of topological indices (Figure 8) demonstrated that both strategies significantly elevated overall network performance, driving a universal surge in average degree. Notably, the LBA strategy consistently outperformed the LDA strategy across the “source,” “sink,” and comprehensive “source-sink” networks. It pushed the average degrees to 7.658, 7.457, and 7.903, respectively (compared to 7.059, 7.285, and 7.262 under LDA). This universal increase indicated a substantial enhancement in overall spatial connectivity.

Concurrently, both optimization paradigms markedly improved global network efficiency and universally shortened the average shortest path lengths. Furthermore, closeness centrality increased notably. For instance, in the comprehensive “source-sink” network, it surged from 0.203 to 0.284 (LDA) and 0.312 (LBA), implying a highly compacted spatial structure. Crucially, however, these macro-level gains required localized structural trade-offs. The clustering coefficients decreased across all network types, reflecting a deliberate weakening of local aggregation to support broader connectivity. Similarly, the decline in eigenvector centrality indicated a shift toward a more balanced, decentralized distribution of node influence.

Overall, compared to the unoptimized baselines, the LDA strategy inclined toward maximizing pure transmission efficiency. In contrast, the LBA strategy exhibited superior structural stability. By fortifying redundant pathways alongside comprehensive connectivity enhancements, it struck an optimal structural balance.

4.3.3. Comparison of Robustness Performance Under Simulated Network Attacks

To evaluate the robustness of the comprehensive “source-sink” corridor network before and after optimization, we simulated attacks under three distinct modes: random node, degree-based, and betweenness-based attacks. Figure 9 illustrates the corresponding variations in global network efficiency and the proportion of the largest connected component. The results demonstrate that both the LDA and LBA strategies successfully enhanced the network’s overall robustness. As anticipated, the unoptimized baseline network exhibited the sharpest drop in both efficiency and the largest connected component across all three attack scenarios, highlighting its severe structural vulnerability.

Specifically, while the LDA strategy yielded marginal improvements against random perturbations, its performance under targeted degree and betweenness attacks remained limited. This vulnerability suggests that although LDA successfully fortifies connections among peripheral nodes, it fails to forge a resilient core topology. In stark contrast, the LBA strategy significantly delayed the structural collapse of both network efficiency and connectivity. It exhibited exceptional stability—particularly against degree and betweenness attacks—demonstrating superior resilience when confronting the targeted removal of critical hub nodes.

Overall, the LBA optimization strategy effectively bolstered the structural stability and functional resilience of the CEN. Consequently, we established it as the optimal strategy for this study.

5. Discussion

5.1. Constructing a CEN Tailored to the Underlying Landscape Patterns of High-Density Cities Based on the “Source-Sink” Theory

The “source-sink” theory classifies landscapes into “source” and “sink” areas based on their ecological processes—sources promote these dynamics, while sinks impede them [25]. Most existing research on high-density urban ENs focuses on single functions, such as habitat networks for biodiversity conservation [12] or singular urban ventilation corridors [26]. In contrast, our CEN integrates the multi-functional synergies of habitat conservation, climate regulation, and cultural recreation. This holistic approach comprehensively addresses both ecological preservation and urban development demands.

Furthermore, our study reveals the dialectical relationship between “spatial competition” and “functional complementarity” across these three dimensions. By superimposing spatial layers and integrating corridors, we can direct cooling benefits toward heat island “sinks,” facilitate species dispersal into marginal habitats, and achieve equitable coverage of recreational services. Network structural analysis indicates that the recreational network forms a highly dispersed web, driven by social activities and road infrastructure in Guangzhou’s urban core. Conversely, natural topography constrains the habitat and climate “sources,” forming clustered agglomerations in the northern mountainous regions. By narrowing our primary research scope to Guangzhou to tightly control macro-climatic variables, we ensure that this distinct contrast accurately reflects spatial differentiation driven strictly by local natural and anthropogenic forces.

We must also explicitly address a potential methodological concern: because we identified sink patches using hot-spot analysis of resistance-related variables, one might question whether sinks are a genuinely independent ecological category or merely a cartographic by-product of the resistance surface. We argue that the two constructs are analytically distinct. A resistance surface produces a continuous gradient that prescribes corridor routing. In contrast, sink delineation extracts spatially discrete clusters using spatial statistical tools (e.g., hot-spot analysis), area thresholds, and landscape connectivity conditions. This effectively filters out isolated or statistically insignificant high-resistance pixels. This distinction aligns with recent multi-functional CEN studies that apply Gi* sink identification alongside—rather than as a derivative of—resistance surface construction [24,34,35]. Critically, the planning outputs differ qualitatively: resistance surfaces dictate where corridors should be routed, whereas sink patches pinpoint where restoration investments should be concentrated. This fundamental gap justifies explicit sink delineation as an independent analytical step.

Crucially, our network incorporates three distinct layers of corridors, each possessing clear spatial planning connotations:

First, the “source-source” corridors connect high-value ecological core areas. Acting as the “main arteries” for maintaining biodiversity, the cold island effect, and cultural heritage, they ensure the continuous supply of core urban ecological functions.

Second, the construction of “sink-sink” corridors transcends the traditional paradigm of merely protecting “existing ecological high-grounds.” Instead, it shifts toward the proactive identification of “problematic depressions.” Rather than serving as efficient ecological flow channels, these corridors spatially manifest as critical restoration opportunity belts. For instance, the “sink-sink” corridors traversing the high-density built-up areas of central and southern Guangzhou delineate zones afflicted by severe UHI effects and habitat fragmentation. Guided by least-cost paths derived from circuit theory, they provide optimal spatial blueprints for implementing precise micro-interventions, such as interstitial greenway connections, pocket parks, and vertical greening.

Finally, the “source-sink” corridors function as “synergistic transmission belts” bridging ecologically superior and vulnerable regions. They are strategically designed to drive the directional flow of ecosystem services, ultimately enhancing the equity and resilience of the entire urban system.

5.2. Evaluating the Structural Characteristics of the CEN in High-Density Cities Based on Complex Network Theory

By integrating complex network theory, we evaluated the structural dynamics of the multi-functional CEN. Notably, the composite network’s average clustering coefficient and global efficiency were lower than those of specific single-function networks. This indicates that while multi-functional integration bolsters overall macroscopic connectivity, the cross-functional diversification of connections inadvertently weakens the tight aggregation of local nodes and prolongs the transmission distances of specific pathways. Fundamentally, this phenomenon reveals an inherent structural trade-off within multi-functional networks. Although cross-functional spatial integration broadens connection channels, the heterogeneous distribution of distinct “source-sink” nodes and the circuitous routing of pathways inevitably compromise local cluster compactness and the transmission efficiency of specific ecological flows [57].

Our findings present a distinct departure from the existing literature on single-function or generalized ENs that employ similar topological methods. Generally, most studies report a synchronous improvement in both clustering coefficient and global efficiency following structural optimization. For instance, Zhang et al. [58] demonstrated that adding 10 strategic corridors to the EN in Anyang City elevated the average clustering coefficient by 5.55% and global efficiency by 8.74%. Concurrently, node and edge recovery robustness increased by 17.44% and 18.08%, respectively, illustrating a positive synergy between local aggregation and transmission efficiency. Similarly, Song et al. [59] found that edge-addition optimization in Harbin’s central urban area significantly improved both network connectivity and local aggregation.

In stark contrast, the structural trade-off observed in our study—characterized by a decreased clustering coefficient and efficiency within the composite network—challenges these conventional positive synergies. This discrepancy highlights the severe spatial competition induced by functional heterogeneity in high-density cities. In these environments, conflicting physical geography and socio-economic forces differentially drive habitat suitability, thermal regulation, and cultural attractiveness. Ultimately, this finding enriches the current understanding of multi-functional integration. It demonstrates that merely pursuing global connectivity cannot guarantee structural stability. Rather, planners must concurrently safeguard local efficiency and ensure the unhindered transmission of function-specific pathways.

5.3. Optimizing the Composite “Source-Sink” Network via Edge-Addition Strategies and Assessing Its Robustness

Our findings demonstrate that while the LDA strategy predominantly enhances global transmission efficiency, the LBA strategy uniquely prioritizes structural balance alongside local aggregation. By maintaining baseline connectivity and comprehensively improving network stability, the LBA strategy exhibits the most robust resilience across all three simulated attack modes. The mechanistic superiority of the LBA approach lies in its focus on connecting “peripheral” or “bridging” nodes (those with lower betweenness centrality) rather than further overloading highly centralized hubs. By strategically forging redundant alternative pathways, this paradigm disperses the pressure of ecological flows that would otherwise concentrate on a few critical bottlenecks. Ultimately, this dispersal significantly mitigates the risk of cascading collapses triggered by localized disruptions. These findings strongly suggest that in high-density urban contexts, fortifying the connectivity of strategic peripheral nodes is a highly cost-effective intervention to bolster systemic resilience.

Contextualizing these results within the broader literature on edge-addition optimizations reveals critical nuances. For instance, Song et al. [59] reported that a low-degree priority strategy performed optimally in robustness tests; by connecting low-degree nodes, it effectively dispersed pressure and enhanced the network’s anti-interference capacity. This aligns broadly with the decentralized mechanism of our LBA strategy. Crucially, however, high-intensity urban development has typically exhausted the spatial resources surrounding core ecological nodes in megacities like Guangzhou. Consequently, establishing new large-scale corridors between these hubs face prohibitive land costs and immense implementation resistance. In contrast, the LBA strategy explicitly circumvents this spatial bottleneck by strategically connecting low-betweenness nodes, which are typically secondary or marginalized patches. As noted by Xiang et al. [29] in their research on cascading failures, channeling ecological flows through such decentralized and redundant pathways can significantly reduce the vulnerability of the entire network to targeted disturbances. Therefore, for high-density cities, this LBA-driven optimization paradigm not only structurally delays network collapse but also provides a highly feasible spatial planning pathway. By relying on urban micro-renewal rather than large-scale demolition, it demonstrates profound applicability to the severe, realistic constraints of high-density urban environments.

5.4. Management of Multi-Functional Spatial Conflicts and Spatial Planning Implementation Pathways for CENs in High-Density Cities

Through spatial overlay analysis, we mapped the composite “source” areas of the multi-functional CEN. Spatially, these high-value zones form clustered agglomerations across the contiguous northeastern woodlands (e.g., Liuxi River and Daling Mountain) and in the core segments of the Pearl River system. While these regions exhibit immense value for biodiversity, thermal regulation, and recreation, they simultaneously face severe spatial conflicts. These primarily manifest as functional incompatibilities on the supply side. For instance, biodiversity conservation strictly demands minimal human interference, whereas recreational services inherently rely on high human influx and enhanced accessibility. Consequently, the spatial overlap of these competing objectives triggers a direct trade-off between the rigid baselines of ecological protection and the flexible demands of human utilization.

Conversely, within composite “sink” areas, conflicts predominantly arise from resource allocation bottlenecks on the restoration demand side. The superimposition of severe ecological deficits necessitates integrated governance for both UHI mitigation and habitat restoration. However, within highly compact built-up environments, severely constrained spatial resources struggle to concurrently satisfy these dual goals. This scarcity inevitably triggers intense competition regarding the spatial layout and prioritization of various ecological recovery projects.

Therefore, grounded in the identified spatial conflict and synergy characteristics, we propose three differentiated mitigation and restoration strategies tailored to high-density urban ecological planning:

(1) Composite “source-source” high-value core areas: Implementing a planar-zonal isolation strategy (see Figure 10a).

Serving as the “ballast stones” of the urban ecosystem baseline, these areas require a “strict protection and quality enhancement” paradigm. By delineating and rigorously enforcing ecological redlines, planners can guarantee the integrity and connectivity of core habitats and cold sources. Furthermore, under the premise of absolute protection, these zones can transition from mere “conservation areas” to “high-quality ecological product suppliers.” This is achievable by optimizing stand structures and sensitively upgrading facilities for nature education and wellness tourism.

(2) Composite “sink-sink” ecologically vulnerable restoration areas: Implementing spatio-temporal dynamic and functional substitution strategies (see Figure 10b).

These areas correspond to “ecological gaps” or “potential restoration corridors” within high-density environments. To address this, we propose transforming them into urban resilience restoration belts. Through functional substitution pathways, the baseline function of ecological restoration should be prioritized in zones with acute conflicts, while recreational functions make moderate concessions. Specifically, by utilizing “micro-renewal and networking” interventions, abandoned lands and street spaces can be systematically converted into resilience restoration belts. Consequently, implementing spatio-temporal dynamic management during critical periods (e.g., peak summer UHI stages or avian migration seasons) effectively resolves superimposed issues like heat islands and habitat fragmentation.

(3) Composite “source-sink” functional synergistic transmission belts: Implementing a vertical-multidimensional diversion strategy (see Figure 10c).

At conflict nodes where multi-functional corridors inevitably intersect, vertical-multidimensional diversion pathways must be deployed. For example, by introducing elevated light walkways, ecological overpasses, or underpasses, planners can achieve the vertical separation of high-frequency human activities from wildlife migratory flows. This intervention guarantees spatial continuity while maintaining low development intensity. Concurrently, efforts should focus on integrating these structures with urban ventilation corridors, cultural heritage routes, and waterfront greenways. Ultimately, this “multi-line integration” successfully transforms originally mutually exclusive functions into non-interfering composite spaces, establishing a robust green infrastructure backbone that synthesizes ecological flow, thermal regulation, and slow-mobility recreation.

5.5. Limitations

The multi-functional CEN and optimization framework we developed not only serves as a scientific tool for unraveling complex urban ecological relationships, but also as a robust decision-support system that translates into actionable spatial planning guidelines. By advocating for a systematic “protection-restoration-connection” intervention paradigm, we delineate a feasible pathway for high-density cities to synergistically enhance ecological security, climate adaptation, and resident well-being. Although grounded in the empirical case of Guangzhou, our composite “source-sink” methodology possesses inherent universality, offering a replicable framework for ecological network planning in other megacities worldwide. Nonetheless, certain limitations remain, warranting further refinement in future research and practice:

First, regarding resistance surface construction, we employed the AHP combined with expert scoring. While all judgment matrices strictly passed the consistency test (CR < 0.1) to ensure logical self-consistency, we omitted a comprehensive weight sensitivity analysis due to length and computational constraints. Future research should incorporate objective weighting methods and multi-scenario simulations to quantitatively assess how varying weight combinations impact corridor routing in complex urban environments. Furthermore, future studies should precisely quantify the spatial overlap among single-function networks to explicitly demonstrate the land-cost-saving advantages of the composite strategy.

Second, constrained by high-intensity surrounding development, the central and southern portions of the study area exhibit significant marginal ecological degradation. Future cross-regional planning should integrate “edge effects” zoning to establish robust ecological buffer zones, thereby shielding core habitats from external human disturbances. Crucially, edge effects should not be viewed merely as a localized issue in highly disturbed areas; rather, they represent a universal factor inherent to all connectivity network modeling. Additionally, because boundary dynamics are intrinsically reciprocal, future cross-regional planning must fully account for the mutual ecological influences and bidirectional flows between adjacent municipalities.

Third, physical connectivity inherently functions as a double-edged sword. Blindly establishing corridors in “sink” areas with poor ecological baselines could inadvertently facilitate the spread of invasive species or pests across degraded landscapes. Consequently, practical implementation must strictly adhere to the fundamental prerequisite of “restoration prior to connectivity.”

Fourth, spatial scale mismatches present an ongoing challenge in multi-functional network implementation. While we successfully constructed a macro-level ecological network, translating these broad corridors into micro-urban forms requires overcoming the “last mile” accessibility gap. Ensuring that residents can effectively access and realize the benefits of macro-corridors necessitates future research focused on bridging regional networks with community-scale green infrastructure (e.g., pocket parks and street trees). Furthermore, although we intuitively identified synergies between different conservation objectives, we lacked a formal, quantitative mathematical analysis of their spatial complementarity, which subsequent studies should refine.

Finally, as a dynamically evolving complex system, the ecological network requires continuous interdisciplinary tracking. Future studies should incorporate multi-scenario simulation models (such as the PLUS or FLUS model) coupled with future global climate change scenarios (e.g., CMIP6). This integration will enable the quantitative prediction of the spatiotemporal evolution of composite “source-sink” networks under varying urbanization trajectories and extreme climate events, thereby providing forward-looking guidance for resilient spatial planning. Furthermore, transitioning from macro-topological optimization to micro-site implementation requires a rigorous spatial cost–benefit analysis. Future research could integrate high-resolution socio-economic data, urban land values, and ecological engineering costs to quantitatively evaluate the economic feasibility of the LBA strategy. Such an interdisciplinary approach would assist urban managers in prioritizing ecological restoration projects under the tight budget constraints typical of high-density megacities.

6. Conclusions

Taking Guangzhou as a case study, we systematically constructed and optimized a multi-functional CEN for a high-density city by integrating the “source-sink” theory of landscape ecology with complex network models. The primary conclusions are as follows:

(1)

The EN constructed based on the “source-sink” theory exhibited a clear and hierarchical structure. This approach facilitated the identification and reinforcement of key ecological elements across multiple scales [24], effectively reflecting the complex mosaic landscape characteristics of high-density megacities. Furthermore, it transcended the traditional single-objective conservation paradigm by proactively identifying ecological core “sources” and restoration-demanding “sinks” across three dimensions: bird habitat conservation, thermal regulation, and cultural recreation. This directly answered the first research question by providing a scientifically rigorous mechanism to define and identify multi-dimensional “source” and “sink” landscapes within intricate urban environments, establishing a precise spatial foundation for targeted governance.

(2)

Driven by natural and anthropogenic forces, the single-function networks exhibited pronounced spatial heterogeneity. The recreational network, characterized by its dispersed web structure, featured the longest corridors and most extensive coverage. In contrast, the habitat and climate networks emphasized the clustered configurations and connectivity of avian habitats and urban cold islands. By integrating these layers, we established a vital urban green skeleton comprising 1137 composite corridors. This framework effectively balanced spatial conflicts in high-density areas by directing cooling effects toward heat sinks, facilitating species dispersal into marginal habitats, and ensuring equitable recreational coverage. This result directly answered the second research question, demonstrating how the “source-sink” framework successfully transcended traditional single-target limitations to construct a multi-functional CEN that actively and effectively mitigated socio-ecological spatial conflicts.

(3)

Compared to other analytical methods, complex network analysis excelled in identifying critical nodes or structural bottlenecks, providing robust decision support for systematic regulation [54,59]. Regarding topological performance, the CENs generally exhibited a higher average degree (6.171) than the single-function ENs [60], reflecting stronger node connectivity and a greater diversity of functional flow paths. However, the single-function ENs outperformed them in average clustering coefficient (0.480) and global network efficiency (0.253). This disparity underscored the necessity of coupling multi-functional integration with targeted structural optimization to safeguard the transmission efficiency of specific pathways. This addressed the topological evaluation aspect of the third research question, signifying that quantitative assessment was indispensable because structural integration did not automatically guarantee optimal functional efficiency.

(4)

Optimization and robustness simulations confirmed that the LBA strategy was the optimal approach for the CEN. By prioritizing the reinforcement of peripheral or bridging nodes, this strategy elevated the network’s average degree to 7.903 and achieved a clustering coefficient of 0.447, successfully balancing global connectivity with local aggregation. This paradigm significantly delayed network collapse under targeted disturbances, establishing it as our best optimization scheme. This conclusively answered the quantitative optimization and evaluation components of the third research question, demonstrating how complex network theory equips planners with a mathematically validated strategy (the LBA approach) to safeguard the CEN’s structural robustness under various targeted disturbance scenarios.

The findings of this study provide valuable theoretical and methodological references for the construction of multi-functional CEN in high-density cities.