Impacts of Sea Level Rise and Urbanization on Ecological Source of the Greater Bay Area

Abstract

This study focuses on the Guangdong-Hong Kong-Macao Greater Bay Area and employs a multi-model coupling method of InVEST-Bathtub-GeoSOS-FLUS to predict and analyze the impacts of sea level rise and rapid urbanization on ecological source areas by the year 2100. The InVEST model is used to delineate areas with higher habitat quality scores as ecological source areas. The Bathtub inundation model predicts the impact ranges under three different sea level rise scenarios by 2100. The FLUS model simulates the land-use pattern of the Greater Bay Area in 2100. Finally, the raster calculator is used to conduct overlay analysis and accurately calculate the impact on ecological source areas under the combined effects of sea level rise and urban expansion. The results show that by 2100, the proportion of cultivated land in the Greater Bay Area is expected to decrease from 24.95% to 10.55%, while the proportion of urban land will increase from 7.69% to 26.84%. Under the dual impacts of the three sea level rise scenarios and urbanization, the affected areas of ecological source areas will reach 109.88 km², 125.05 km², and 255.10 km², respectively. This study provides an important basis and decision-making support for the sustainable planning and scientific management of ecological source areas in the Greater Bay Area.

1. Introduction

In the context of the increasingly severe global climate change, sea level rise has become a major environmental issue that urgently needs to be addressed. The Intergovernmental Panel on Climate Change (IPCC) has reported that by the end of this century, the sea level is projected to rise by 0.3 to 1.1 m [1]. This change is expected to have comprehensive and profound impacts on the ecosystems, economic development, and social structures of coastal cities [2]. Sea level rise will lead to an increase in extreme weather events, including storm surges [3], extensive loss of coastal wetland ecosystems [4], exacerbation of flood or drought events [5], changes in sediment transport within river basins [6], and saline intrusion, which will reduce the availability of freshwater resources and affect the supply of agricultural water and drinking water, thereby directly impacting the quality of life of urban residents. Meanwhile, the accelerating global urbanization process has led to the overexploitation of land resources and the overconsumption of natural resources, resulting in a series of serious problems such as increased landscape fragmentation, continuous degradation of ecological functions [7], blocked ecological flows [8], and loss of biodiversity [9]. These issues pose a significant threat to the sustainable development of regional systems. However, most of the research results have two limitations: First, domestic scholars mostly focus on the direct impact of sea level rise, while ignoring the superimposed erosion effect of rapid urbanization on ecological space. For example, studies in the Pearl River Delta region have confirmed that urbanization has led to a reduction of 187.67 km of natural coastline, but the synergistic mechanism between sea level rise and urban expansion has not been quantified [10]. Second, studies on complex urban agglomerations such as the Greater Bay Area generally lack the analysis of protective thresholds guided by the “risk bottom line”. Although existing research has identified the spatial distribution of ecological source areas, it has not revealed their minimum safety boundaries under the dual pressures of extreme climate and urbanization, resulting in a lack of rigid constraints for planning strategies. To address these research gaps, the innovative breakthrough of this paper lies in proposing the concept framework of “Minimum Risk Protection Baseline”, emphasizing the quantification of irreversible ecological loss thresholds under the coupled effects of sea level rise and urbanization, ignoring extreme events (storm surges, earthquakes) and artificial protection. A multi-model chain coupling system of “InVEST-Bathtub-GeoSOS FLUS” is constructed, using the InVEST habitat quality module to identify ecological source areas (>0.94 high-value areas), breaking through the complexity limitations of traditional connectivity analysis, integrating the FLUS model to predict the land-use pattern in 2100, and using the Bathtub model to simulate the inundation range under three sea-level scenarios (0.5 m/1 m/2 m), using a simplified model to reveal the underlying risks. This study aims to focus on the Greater Bay Area of Guangdong, Hong Kong and Macao, achieving a spatially explicit assessment of the vulnerability of ecological source areas under multiple scenarios and long time series (2000–2100), providing a quantifiable decision-making red line for territorial space resilience planning.

The InVEST (Integrated Valuation of Ecosystem Services and Trade-offs) model, as a professional tool for analyzing and assessing complex ecological and environmental systems, has been widely applied in both academic and practical fields. Cheng Aiguo and Jia Fangfang have respectively used the InVEST model to evaluate the habitat quality in Hefei City and the level of ecosystem services in the Ganjiang River Basin [11,12]. Additionally, the InVEST model is capable of being applied to scenario analyses of the past and future, providing scientific basis and technical support for policymakers to formulate long-term ecological protection policies. Wang Zuo applied it to the temporal evaluation of habitat quality in Anhui Province [13]. Numerous scholars both domestically and internationally have employed the InVEST model in regional habitat quality assessment studies, fully verifying the model’s effectiveness and reliability in habitat quality evaluation. Currently, research on urban wetland communities has certain limitations in terms of dimensionality. Domestic scholars mostly focus on analyzing the impacts on coastal habitats from the single perspective of sea level rise, while the role of urbanization in this process is somewhat neglected, which to some extent restricts the practical exploration of systematic and comprehensive protection of ecological sources. The FLUS model (Future Land Use Simulation model) developed by Liu et al. has successfully achieved an organic integration of System Dynamics (SD) and Cellular Automata (CA) based on neural networks. It can effectively address the issue of land transformation probability under the combined influence of natural factors and human activities, providing a powerful technical means for in-depth analysis of landscape pattern evolution laws [5,14]. This model has also been successfully applied in many relevant studies.

The Greater Bay Area features an extensive coastline and a rich variety of ecosystem types, which form an important material basis for its sustainable development. However, with the intensification of climate change and the rapid advancement of urbanization, the advantageous habitat space in the Greater Bay Area has been significantly encroached upon, and the natural coastline and coastal tidal wetlands have experienced a substantial reduction in area. This paper takes the Greater Bay Area as the research object and, based on the InVEST model, integrates land-use patterns with sea level rise projections to conduct an in-depth analysis of the impacts brought about by sea level rise and urbanization. The aim is to provide scientific and precise decision-making basis for the sustainable development of the Greater Bay Area.

2. Study Area and Research Data

The Guangdong-Hong Kong-Macao Greater Bay Area (21°–25° N, 111°–116° E) is located in the central and southern part of Guangdong Province, bordering the South China Sea to the south. It includes nine cities—Guangzhou, Shenzhen, Zhuhai, Foshan, Dongguan, Huizhou, Jiangmen, Zhongshan, and Zhaoqing—as well as the two Special Administrative Regions of Hong Kong and Macao, covering a total area of 56,000 square kilometers. The Greater Bay Area has an extensive coastline and a rich variety of ecosystem types. The wetlands cover an area of 8650 square kilometers, accounting for 15.45% of the total area. These wetlands have high productivity and harbor abundant biological resources, making them an essential material basis for the sustainable development of the Greater Bay Area. However, in recent years, with the impact of climate change and rapid urbanization, the advantageous habitat spaces in the Greater Bay Area have been extensively encroached upon. The areas of natural coastlines and coastal mudflat wetlands have been significantly reduced, and the region has long been affected by various marine disasters (Figure 1).

This paper selects the Guangdong-Hong Kong-Macao Greater Bay Area as the research object. The study area includes both urban construction land and a large amount of non-urban and rural construction land, such as forests, waters, and cultivated land, making it highly representative. The DEM (Digital Elevation Model) data (30 m resolution) were obtained from the Geospatial Data Cloud [15]. Data on highways, railways, main roads, and administrative divisions were sourced from the Resource and Environmental Data Center of the Chinese Academy of Sciences [16]. Land-use data for the years 2000 and 2020 were obtained from GlobeLand30 [17]. According to the research needs, land-use types were classified into six categories using ArcGIS 10.8.1: cultivated land, forest land, grassland, waters, urban and rural construction land, and others. These categories were assigned sequential numbers starting from 1. All data pixels were uniformly processed to a resolution of 100 × 100 to ensure consistent row and column numbers and were converted into the GeoTIFF format with a WGS 1984/World Mercator projection [18].

3. Research Methods

3.1. Habitat Quality Assessment Method Based on the InVEST Model

Habitat refers to the potential of an ecosystem to provide the conditions necessary for the survival and reproduction of species. The Habitat Quality model in InVEST 3.12.0 establishes a connection between land-use data and threat factors, generating a habitat quality index based on the degree of threat posed by different threat factors. This index is used to assess habitat quality and degradation under different landscape patterns. Higher habitat quality indicates greater biodiversity in the corresponding study area. The natural breaks classification method is used to convert high-scoring raster cells into polygons, which are then identified as ecological source areas [19,20].

The formulas for calculating habitat quality and degradation are shown in Equations (1) and (2), respectively.

$$Q_{xj}=H_j(1-{\displaystyle \frac{D_{xj}^z}{D_{xj}^z+k^2}})$$

(1)

$$\left\{\begin{array}{l}{D}_{xj}={\displaystyle \sum _{r=1}^R{\displaystyle \sum _{y=1}^{Y_r}(\frac{W_r}{{\displaystyle {\sum }_{r=1}^R{W}_r}})r_y{i}_{rxy}{\beta }_x{S}_{jr}}}\hfill \\ i_{rxy}=1-(d_{xy}/d_{r,\max })\hfill \\ i_{rxy}=\exp (-d_{xy}(2.99/d_{r,\max }))\hfill \end{array}\right.$$

(2)

Among them, $Q_{xj}$ refers to the habitat quality index of grid $x$ in land-use type $j$; $D_{xj}$ refers to the habitat threat level; $H_j$ refers to the habitat suitability; $k$ is the half-saturation constant; $z$ is the default parameter of the model. $R$ is the number of threat factors; $W_r$ is the weight of each threat factor; $Y_r$ is the number of grids in the threat factor layer; $r_y$ is the intensity of the threat factor; ${\beta }_x$ is the habitat disturbance resistance level; $S_{jr}$ is the sensitivity of the threat factor; $i_{rxy}$ is the threat level of $r_y$ to grid $x$; $d_{xy}$ is the straight-line distance between the grid and the threat source; $d_{r,\max }$ is the maximum influence distance of the threat factor.

This study, in combination with the InVEST model and by reviewing the parameter values of relevant research materials concerning the Guangdong-Hong Kong-Macao Greater Bay Area [21,22], defines urban land, cultivated land, water bodies, and unused land, which are the most concentrated areas of human activities and have a significant impact on habitats, as threat factors. The selection of these four factors as threat factors is based on three considerations: the high-intensity urbanization process (urbanization rate of 89.0%) [23], water and soil pollution caused by excessive pesticide use [24], and the interference and damage to the ecosystems of rivers [25], lakes, and reservoirs caused by human activities. The maximum influence on weights and distances of different threat factors (

Table 1. The weight and the maximum influence distance of the threat source.

Threat Factors	Weight	${\mathit{d}}_{\mathit{r},\mathbf{max}}$ (km)	Decay Type
urban land	1	8	linear
cultivated land	0.7	3	linear
waters	0.6	10	exponential
unused land	0.4	5	exponential

), as well as the habitat suitability and the sensitivity of different habitats to threat factors (

Table 2. Sensitivity of land-use types to threat agents.

Land Cover Types	Habitat Suitability	Sensitivity
		Urban Land	Cultivated Land	Waters	Unused Land
cultivated land	0.5	0.5	0	0.5	0.35
forest land	1	0.6	0.3	0.7	0.7
grassland	0.9	0.3	0.5	0.5	0.55
waters	1	0.8	0.1	0.7	0.65
urban land	0	0	0	0	0
unused land	0.7	0.1	0.1	0.2	0.2

), were determined.

The habitat quality index generated by the Habitat Quality model in InVEST ranges from [0, 1], with higher values indicating better habitat quality in the corresponding study area [26]. This study adopted the natural break method to convert the grids with high assessment scores into vector surface elements and took the areas with high habitat quality scores as ecological source areas for in-depth research.

This study focuses on regions with high habitat quality scores as ecological source areas for two reasons: Although areas with high habitat quality scores do not necessarily possess perfect qualities as ecological source areas (for example, there may still be certain deficiencies in connectivity), within the study area, apart from these areas, there are no other regional resources that can undertake and play the role of ecological source areas. After all, areas with high habitat quality scores possess more functional connotations and roles that an ecological source area should have than those with low scores.

3.2. Sea Level Rise Inundation Model

According to literature reviews, long-term (end-of-century) sea level rise projections are more common than mid-term (2050) ones, with approximately 78% of countries simulating sea level rise scenarios extending to 2100 [27]. This study employs scenario analysis and selects three sea level rise scenarios—0.5 m, 1.0 m, and 2.0 m—to represent low, medium, and high scenarios, respectively. The 0.5 m and 1.0 m scenarios fall within the latest IPCC-projected range of global sea level rise by the end of this century (0.28 m–1.01 m). These values have also been widely used in previous studies on sea level rise projections. The 2.0 m scenario is based on the IPCC 2019 report, which indicates that the maximum possible global mean sea level rise by 2100 could reach 2.0 m if the ice sheets below sea level in Antarctica were to melt [28].

The aim of this study is to identify the minimum risk protection baseline to reveal the extent to which the dual effects of sea level rise and urban expansion erode ecological resources even in the absence of extreme weather, earthquakes, tides and protective facilities.

Therefore, in this study, a simple “Bathtub model” (simple Bathtub model) was chosen to simulate the degree of inundation due to sea level rise. This method is based on the Digital Elevation Model (DEM), assuming that all areas with elevations lower than the set sea level are submerged, without considering the influence of factors such as hydrological connectivity, tidal dynamics, and coastal protection facilities [28].

The process of simulating the potential impacts of sea level rise on the Guangdong-Hong Kong-Macao Greater Bay Area using ArcGIS 10.8.1 is as follows: The elevation data of the study area are input into the software. The “Extract by Attribute” tool is then selected, and commands specifying “VALUE ≤ 0.5,” “VALUE ≤ 1.0,” and “VALUE ≤ 2.0” are sequentially entered. Based on the principle that areas with elevation values lower than the set sea level are considered inundated, the corresponding inundation ranges are derived. This provides a foundation for subsequent analyses [29].

3.3. GeoSOS-FLUS Landscape Pattern Prediction Model

The GeoSOS-FLUS model is a land-use change simulation model based on Artificial Neural Networks (ANNs) and Cellular Automata (CA). It can account for the transition rules of land-use types, spatial neighborhood effects, the influence of driving factors, and land-use constraints, generating high-precision land-use change simulation results.

3.3.1. Suitability Probability Based on ANN

The FLUS model uses the Artificial Neural Network algorithm (ANN) to derive the suitability probability of various land-use types within the study area from initial land-use data, incorporating a variety of driving factors that include both human activities and natural effects [30].

The evolution of land-use patterns is the result of the combined effect of natural geographical factors and social factors. Among them, natural geographical factors, as the objective basis for land-use changes, are stable and have significant cumulative effects in the long term. Social factors, based on the natural environment, influence the current land-use pattern and are generally regarded as the dominant factors driving land-use changes in the short term [31]. Because the aim is to predict the minimum risk protection baseline in 2100, more stable driving factors need to be selected. Therefore, at the natural factor level, three indicators, namely Digital elevation Model (DEM), slope and aspect, are selected. At the social factor level, three indicators, namely distance from railway, distance from main road and distance from expressway, are selected for the prediction of land-use pattern in 2100.

3.3.2. Simulation Based on Cellular Automata

In the GeoSOS-FLUS model, the land-use transformation probability depends not only on the suitability probability generated by the ANN but also on the neighborhood density, inertia coefficient, transformation cost, and competition among land types, which together determine the total probability of land-use transformation.

The formula for calculating neighborhood density is as follows:

$${\Omega }_{p,k}^t={\displaystyle \frac{{\displaystyle {\sum }_{N\times N}con(c_p^{t-1}=k)}}{N\times N-1}}\times w_k$$

(3)

Among them, ${\Omega }_{p,k}^t$ represents the neighborhood density, ${\displaystyle {\sum }_{N\times N}con(c_p^{t-1}=k)}$ indicates the total number of pixels of land-use type $k$ at the end of the previous iteration $(t-1)$ within the window of $N\times N$, and $w_k$ represents the weight of neighborhood influence for different land-use types.

The formula for calculating the inertia coefficient is as follows:

$$Inerti{a}_k^t=\left\{\begin{array}{ll}Inerti{a}_k^{t-1}\hfill & \left|D_k^{t-2}\right|\le \left|D_k^{t-1}\right|\hfill \\ Inerti{a}_k^{t-1}\times (D_k^{t-2}/D_k^{t-1})\hfill & 0>D_k^{t-2}>D_k^{t-1}\hfill \\ Inerti{a}_k^{t-1}\times (D_k^{t-1}/D_k^{t-2})\hfill & D_k^{t-1}>D_k^{t-2}>0\hfill \end{array}\right.$$

(4)

Among them, $Inerti{a}_k^t$ represents the inertia coefficient of land-use type $k$ at iteration time $t$, while $D_k^{t-1}$ and $D_k^{t-2}$ respectively indicate the differences between the number of pixels and the required number of land-use type $k$ in the previous iteration and the iteration before that.

The formula for calculating the total probability of land-use transformation is as follows:

$$TPro{b}_{p,k}^t=s{p}_{p,k}\times {\Omega }_{p,k}^t\times Inerti{a}_k^t\times (1-s{c}_{c\to k})$$

(5)

Among them, $TPro{b}_{p,k}^t$ is the total probability of pixel $p$ converting to land-use type $k$ at iteration $t$, $s{p}_{p,k}$ is the suitability probability output by the neural network, $s{c}_{c\to k}$ is the cost of land-use type $c$ converting to type $k$, and $1-s{c}_{c\to k}$ indicates the difficulty level of the conversion occurring.

3.3.3. Model Accuracy Validation

The coefficient can effectively validate the accuracy of predictive models. The coefficient is commonly used for model accuracy validation. A higher coefficient indicates greater precision in the predicted results, while a lower coefficient suggests higher model accuracy. The GeoSOS-FLUS model integrates both coefficients to enhance the scientific basis of simulation accuracy [17].

The $Kappa$ coefficient can effectively assess the accuracy of the prediction model, while the $FoM$ coefficient is commonly used for model accuracy validation. A higher value of the $Kappa$ coefficient indicates higher prediction accuracy, whereas a lower value of the $FoM$ coefficient signifies greater model accuracy. The GeoSOS-FLUS model integrates both coefficients to enhance the scientific basis of simulation accuracy [17].

The formula for the $Kappa$ coefficient is as follows:

$$Kappa=(p-p_e)/(1-p_e)$$

(6)

Here, $p$ represents the overall accuracy, $p_e=(a_1\times b_1+a_2\times b_2+\cdots a_n\times b_n)/(S\times S)$ and $n$ represent the number of categories, $a_1,a_2\cdots a_n$ represents the area of each land-use type in the actual results, $b_1,b_2\cdots b_n$ represents the area of each land-use type in the simulated results, and $S$ indicates the number of samples.

The formula for the $FoM$ coefficient is as follows:

$$FoM=B/(A+B+C+D)$$

(7)

Among them, $A$ represents the area where conversion actually occurred but was not simulated; $B$ represents the area where conversion occurred both in reality in and the simulation; $C$ represents the area where conversion occurred in reality, but the simulated conversion result differs from the actual one; $D$ represents the area where no change occurred in reality, but conversion was simulated.

The simulation diagram of the land landscape pattern in the Greater Bay Area based on the model is shown in Figure 2.

4. Results and Analysis

4.1. Identification of Ecological Source Areas

Ecological source areas are the “source” of ecological land protection, serving as habitats for species and the starting points for species interaction and dispersal. They are also the core components of ecological networks. Based on the Habitat Quality module in the InVEST model, habitat quality data were generated. To visually present spatial habitat quality, the natural break classification method in the Reclassify tool of ArcGIS 10.8.1 software was used to divide the data into five levels: low, moderate–low, moderate, good, and excellent (see Figure 3,

Table 3. Classification of habitat quality in the Guangdong-Hong Kong-Macao Greater Bay Area.

Rank	Range of Values	Description of Rank Characteristics
low	0	Habitat quality is low, and biodiversity is low.
low to moderate	>0–0.49	Habitat quality is relatively low, and biodiversity is relatively low.
moderate	>0.49–0.79	Habitat quality is moderate, and biodiversity is moderate.
good	>0.79–0.94	Habitat quality is moderate, and biodiversity is moderate.
excellent	>0.94–1	Habitat quality is excellent, and biodiversity is excellent.

). Overall, the habitat quality in the Guangdong-Hong Kong-Macao Greater Bay Area is relatively high, with the excellent level accounting for over 60% of the total area (

Table 4. Habitat quality grades, area and proportion.

Rank	Area (km2)	Proportion (%)
low	5346.52	8.16
low to moderate	483.25	0.74
moderate	18,517.07	28.26
good	1690.13	2.58
excellent	39,478.73	60.26
total	65,515.70	1

). These areas are mainly distributed in the western and northeastern regions. Low-quality habitats account for 8.16% of the total area and are concentrated in the central region and the Pearl River estuary area.

Using the Mask Extraction tool, the ecological source areas in the study region were divided and extracted according to administrative regions, with the results shown in

Table 5. The area and proportion of ecological source areas in each administrative division of the Guangdong-Hong Kong-Macao Greater Bay Area.

Administrative Division	Ecological Source Area (km2)	Administrative Region Area (km2)	Proportion
Guangzhou	4107.92	8542.92	48.09%
Jiangmen	6521.47	10,933.84	59.64%
Zhaoqing	14,051.86	17,707.39	79.36%
Huizhou	8961.13	13,392.70	66.91%
Hong Kong	778.43	1319.49	58.99%
Macau	9.94	37.46	26.53%
Shenzhen	871.36	2282.38	38.18%
Zhuhai	968.18	1826.96	52.99%
Foshan	1492.47	4475.57	33.35%
Dongguan	804.46	2908.84	27.66%
Zhongshan	751.82	2088.14	36.00%
Total	39,319.04	65,515.70

. Zhaoqing City has the highest proportion of source areas, followed by Huizhou, while the Macao Special Administrative Region has the lowest. In this study, areas with habitat quality greater than 0.94, classified as excellent, were extracted as the source areas for the ecological network (Figure 4). These areas cover 39,478.75 square kilometers, accounting for 60.25% of the total area of the study region.

4.2. Sea Level Rise Simulation Results

The simulated inundation areas due to sea level rise are primarily concentrated in the southeastern coastal regions of the study area. As shown in Figure 5, the inundation ranges of the simulated sea level rise are mainly located in the southeastern coastal areas of the study region, including Guangzhou, Dongguan, Zhongshan, Shenzhen, Jiangmen, and Zhuhai. The Hong Kong Special Administrative Region has a relatively smaller inundation area. Under the scenarios of 0.5 m, 1.0 m, and 2.0 m sea level rise, the inundated areas are 556.55 km², 600.36 km², and 822.54 km², respectively. Guangzhou is the most affected city, with inundated areas of 155.95 km², 160.89 km², and 202.27 km² under the three scenarios, respectively (

Table 6. Area table of administrative divisions in the Guangdong-Hong Kong-Macao Greater Bay Area affected by different rising heights at sea level.

	SLR0.5 (m2)	SLR1 (m2)	SLR2 (m2)
Guangzhou	155.95	160.89	202.27
Jiangmen	89.43	111.16	177.09
Zhaoqing	7.98	9.49	16.31
Huizhou	10.47	10.49	10.69
Hong Kong	21.00	21.00	21.13
Macau	1.56	1.56	1.72
Shenzhen	25.73	25.78	26.40
Zhuhai	121.46	127.07	168.91
Foshan	0.56	3.69	26.34
Dongguan	59.74	62.54	74.51
Zhongshan	62.67	66.69	97.17
Total	556.55	600.36	822.54

).

4.3. Land-Use Change

4.3.1. Simulation and Result Validation

This study used the actual land-use data from 2000 as the initial state and randomly selected 2% of sample points for model training to obtain different classification results. Combined with the land-use policies issued in the Guangdong-Hong Kong-Macao Greater Bay Area in recent years, after repeated adjustments, the neighborhood density was set as follows: cultivated land 0.7, forest land 0.4, grassland 0.4, waters 0.2, urban and rural construction land 0.9, and others 0.2. Additionally, the number of iterations, neighborhood range size, model acceleration factor, and the quantity target for simulated land-use transformation were set to the default parameters, which are 3700, 3, 0.1, and 8, respectively. The Markov model was used to predict the land-use types for 2020, and the simulated results were compared with the actual land-use data for 2020. Using the Precision Validation module, the Kappa coefficient was calculated to be 0.883 and the figure of merit (FoM) coefficient to be 0.119, indicating a strong consistency between the model predictions and the actual situation. Under the same parameters, the FLUS model was used to generate the land-use type map for 2040 (Figure 6c). Similarly, based on the simulated land-use data for 2040, 2060, and 2080, the land-use data for 2100 were ultimately predicted. The conversion cost matrix is shown in

Table 7. Conversion cost matrix.

	Cultivated Land	Forest Land	Grassland	Waters	Urban land	Unused Land
cultivated Land	1	0	1	0	1	1
forest land	0	1	0	0	0	0
grassland	0	0	1	0	1	1
waters	0	0	0	1	0	0
urban land	0	0	0	0	1	0
unused land	0	0	1	0	0	1

, and the predicted land-use results are shown in Figure 7.

4.3.2. Characteristics of Land-Use-Type Changes

Based on the proportion of each land-use type across different years, a line chart was created as shown in Figure 8. Figure 8 shows that during the period from 2000 to 2100, the relative dominance of land-use types in the Greater Bay Area remained relatively stable, with forest land consistently occupying the dominant position. Its proportion remained relatively stable throughout the period, accounting for approximately 50%. The changes in urban and rural construction land and cultivated land were the most significant. The proportion of urban and rural construction land steadily increased from 7.69% in 2000 to 26.83% in 2100. The growth was particularly pronounced between 2000 and 2040; although the rate of increase gradually slowed down between 2040 and 2100, it still remained in a state of continuous expansion. In contrast, the proportion of cultivated land decreased from 24.97% to 10.54%. Further analysis revealed that urban and rural construction land expanded outward, encroaching on surrounding arable and grasslands, although this trend of expansion has been somewhat mitigated. It was found that the main source of the increase in urban construction land was cultivated land, a process that led to the fragmentation of cultivated land.

4.4. Overlay Analysis

This study conducted an overlay analysis of the ecological source areas in the Guangdong-Hong Kong-Macao Greater Bay Area, the simulated inundation ranges of three different sea level rise scenarios, and the projected urban construction areas for the year 2100. The aim was to identify the extent of ecological source areas affected by the combined impacts of sea level rise and rapid urbanization by 2100. The specific affected areas are illustrated in Figure 8. After excluding waters, the land-use type most severely affected is urban construction land, followed by cultivated land. These areas face extremely high risks of seawater inundation. To avoid potential disaster losses and resource wastage, development and construction activities in these regions should be strictly restricted.

Further quantitative analysis shows that by 2100, the area of ecological source areas affected by the dual effects of sea level and urbanization at three different heights will be 109.88 square kilometers, 125.05 square kilometers and 255.10 square kilometers, respectively, accounting for 0.17%, 0.19% and 0.34% of the total ecological source areas, and the coupling is very limited (see

Table 8. The impact area and proportion of highly coupled urbanization on ecological source areas of three types of sea level rise in 2100.

Sea Level Rise Height (m)	Ecological Source Areas Affected by Dual Influences (km2)	Area Proportion Affected by Dual Influences in Total Area (%)
0.5	109.88	0.17
1	125.05	0.19
2	255.10	0.34

for details).

5. Discussion

1. The purpose of this study was to find the minimum risk protection baseline and select the simple “bathtub model”. Although this method is widely used in the preliminary assessment of large-scale long-term inundation risks (such as in IPCC reports) [32,33], it has the following three limitations. One is hydrological simplification: Without considering surface runoff, groundwater infiltration or drainage systems, the inland inundation range may be overestimated. Second, dynamic factors lacking short-term events such as storm surges and astronomical tides have not been included, and the actual risk of inundation is much higher than the static water level [34]. The third issue is the neglect of protective facilities: The disaster reduction role of existing or planned engineering measures such as DAMS and gates has not been considered, which may lead to an overestimation of the inundation area.

However, a simple “Bathtub model” was chosen for prediction, and its core value lies in revealing the minimum protection that ecological source areas need to avoid being submerged by seawater. Without considering extreme weather, astronomical tides, earthquakes and the failure of protective facilities, the simulation results show that under the three scenarios of sea level rise heights of 0.5 m, 1 m and 2 m, the coupling effect of urbanization and sea level rise will still lead to irreversible losses of the ecological source area ranging from 109.88 to 255.10 km² (accounting for 0.17% to 0.34% of the total area of the ecological source).

2. In the specific research of the Guangdong-Hong Kong-Macao Greater Bay Area, the setting of threat factor weights needs to be combined with the regional ecological environment characteristics, the intensity of human activities and relevant academic research results and form a quantitative basis through multi-dimensional argumentation. The specific logic is as follows: First, refer to the “Blue Book of the Guangdong-Hong Kong-Macao Greater Bay Area: The key data of “urbanization rate reaching 89.0%” in the “Report on the Construction of the Guangdong-Hong Kong-Macao Greater Bay Area (2024)” [35], which indicates that the high-intensity urbanization process makes the expansion of urban land use most directly and significantly occupy and interfere with the surface habitat. Therefore, in the weight distribution, urban land use should be given a higher priority. Second, there are problems of soil and water pollution caused by excessive application of pesticides and chemical fertilizers in the cultivated land of this area. Thirdly, water areas are confronted with ecological function degradation caused by human activities. Fourth, the types of unused land (such as abandoned industrial and mining land, bare rock, sandy land, saline–alkali land, etc.) usually represent extreme or degraded environmental conditions, exerting direct or indirect pressure on the survival of most organisms. Therefore, this study takes these four types of land use as important carriers of ecological stress.

Moreover, this type of habitat quality research in this study usually depends on the parameters of the InVEST model dataset and the researcher’s understanding of the study area, because the InVEST habitat quality model requires the weights of threat factors, the maximum impact distance, and suitability. Therefore, the InVEST habitat quality model may be significantly influenced by the researchers’ judgments, which is one of the limitations and potential for future improvement of this model [36,37].

3. This study directly equates the areas with high InVEST habitat quality (>0.94) to ecological source areas. This is because the high-quality habitat areas (HQ > 0.94) are significantly superior to other areas in terms of biodiversity maintenance capacity (Table 3). Moreover, these high-quality habitat areas are mainly concentrated in the continuous mountainous areas in the west (Zhaoqing-Jiangmen) and the hilly areas in the northeast (Huizhou) and are less affected by human interference and destruction. Their spatial distribution characteristics show (Figure 4) that the large and small patches naturally have high connectivity, which largely meets the requirements of landscape connectivity for ecological source areas. Moreover, this study focuses on the coupled stress effects of sea level rise and urbanization, rather than the optimization of ecological networks. When identifying “highly vulnerable ecological spaces”, habitat quality is used as a direct indicator reflecting the ecosystem’s sensitivity to stress factors, and its simplified application is in line with the research orientation of the minimum risk bottom line. However, this simplified treatment has certain potential limitations to some extent. For instance, the lack of comprehensive, in-depth and detailed investigation and confirmation of connectivity may lead to deviations in the identification of ecological source areas. There may be a few areas with poor connectivity within the high-quality habitat areas, which may have certain impacts on the exchange of species, nutrients and energy within this area.

4. This study selects the Digital Elevation Model (DEM), slope, aspect, distance from railway, distance from main road and distance from expressway as driving factors. The reason is that these factors have stronger stability over a longer time dimension compared with economic and policy factors. The selection method of such driving factors can meet the research requirements with the core goal of finding the minimum risk protection baseline. The stability of natural geographical factors (elevation, slope and direction), which are not affected by policy or economic fluctuations. The layout of major transportation arteries forms the framework of the spatial structure of the Guangdong-Hong Kong-Macao Greater Bay Area and is stable over a long-time scale.

5. It is estimated that by 2100, the areas of ecological sources affected by the coupling of three different levels of sea level rise and urbanization will account for 0.17%, 0.19% and 0.34% of the total area, respectively, and the coupling impact will be limited. There are three reasons for this: First, geographical isolation. The main ecological source area of the study area is in the western mountains (Zhaoqing, Jiangmen) far from the coast and the northeastern hills (Huizhou), with a DEM exceeding 50 m. The second is urban expansion avoidance. Urban expansion prioritizes the occupation of cultivated land in plain areas (reducing cultivated land by 14.41%), while ecological sources are mostly located in steep slopes with high development costs (forest land only decreases by 2.3%). The third issue is risk mismatch. The spatial overlap between the areas inundated by sea level rise (estuarine lowlands) and high-value ecological sources is low, and high-quality habitats are concentrated in mountain forests rather than coastal wetlands. However, the areas affected by the coupling of sea level rise and urbanization are mostly the coastal wetlands at the Pearl River estuary (such as Nansha in Guangzhou and Qi ‘ao Island in Zhuhai). As key nodes for migratory birds and important restoration areas for mangrove resources, their loss will directly weaken the ecological network connectivity of the Greater Bay Area. Future planning should be based on the minimum risk protection baseline proposed in this study, and the ecological red lines that inundate sensitive areas should be delineated first.

6. The ecological source loss (109.88–255.10 km²) revealed in this study is the minimum estimated loss after excluding extreme weather, astronomical tides, earthquakes and the failure of protective facilities, representing the baseline space that must be adhered to for ecological security in the Greater Bay Area. If extreme weather conditions are superimposed or protection fails, the actual impact may increase by several orders of magnitude. Planning suggestions need to go beyond engineering thinking and incorporate the minimum risk protection baseline into the rigid constraints of territorial space.

6. Conclusions

This study takes the Guangdong-Hong Kong-Macao Greater Bay Area as the research object and employs the integrated method of “InVEST-Bathtub-GeoSOS-FLUS” to predict the changes in ecological source areas under the combined impacts of urban construction and sea level rise by 2100. The results indicate the following:

• Areas with high habitat quality are mainly distributed in Zhaoqing, Jiangmen, Zhuhai, Huizhou, Hong Kong, the southern part of Zhongshan, and the northeastern part of Guangzhou. In contrast, the central region, including Foshan, Dongguan, southern Guangzhou, and the coastal areas around the Pearl River estuary, has relatively poor habitat quality.
• The impact of sea level rise is primarily concentrated in the southeastern coastal areas of the study region. The inundated areas under the scenarios of 0.5 m, 1.0 m, and 2.0 m sea level rise are 556.55 km², 600.36 km², and 822.54 km², respectively. Guangzhou is the most affected administrative region, followed by Zhuhai.
• The land-use prediction results for 2100 based on the FLUS model show that the proportion of urban construction land increases from 7.69% to 26.8%, while the proportion of cultivated land decreases from 24.95% to 10.55%. The rate of change is slowing down, indicating that the newly added urban construction land has encroached on most of the cultivated land areas.
• By 2100, urban construction land is projected to be the most affected land-use type (excluding waters) under the three sea level rise scenarios. The overall proportion of ecological source areas affected by the combined impacts of the three sea level rise scenarios and urbanization is relatively small, accounting for 0.17%, 0.19%, and 0.34%, respectively. The coupling impact range is limited.

Overall, the impacts of sea level rise and urban development on ecological source areas are complex and urgent. This study provides a key reference for regional ecological protection and sustainable urban development.