Introduction

As global climate change intensifies, rising sea levels, frequent extreme weather events, and ecosystem degradation continue to affect both human and natural systems1,2,3. Coastal urban agglomerations, which serve as powerful engines driving regional economic development, have experienced rapid urbanization, leading to disordered urban sprawl, chaotic spatial layouts, and large-scale reclamation projects. This uncontrolled expansion makes it difficult to effectively respond to the adverse effects of climate change4,5. Furthermore, due to the combined influence of geographical location and unique topography, coastal urban agglomerations have become highly sensitive and vulnerable to the impacts of climate change. According to recent projections by the World Bank, without adaptation, up to 300 million people could be affected annually by coastal flooding by 2050 under high emission scenarios6. This significantly increases the pressure on coastal urban areas to implement adaptive spatial strategies. A study funded by the European Investment Bank (EIB) estimates that global economic losses to coastal cities from sea level rise and climate change could exceed $1 trillion by 21007. Climate change is expected to expose coastal urban agglomerations to multiple risks, posing major challenges to the well-being and safety of urban residents8,9.

There are two main strategies for dealing with climate change: Mitigation and adaptation10,11. Mitigation refers to slowing down the growth rate of greenhouse gases through various emission reduction measures, emphasizing the control of sources in the short term. Adaptation is to adjust response measures in a timely manner based on the assessment of climate change risks, emphasizing the reduction of threats and potential impacts of climate change in the long term12,13. Given that coastal urban agglomerations play a key role in global socio-economic development, their formation, growth, and expansion have become primary subjects of urban spatial research. The spatial form and layout of coastal urban agglomerations directly influence their adaptability to climate change. However, the disorderly expansion of land use and the encroachment on ecological spaces driven by rapid urbanization increase the vulnerability of coastal areas to climate change14,15. As a sensitive area of ​​land-sea interaction and resource-economic agglomeration, coastal cities are subject to the impact of climate change-induced sea level rise, intensified storm surges, and red tide outbreaks, and the frequency, scope, and degree of disasters are much higher than those in inland areas. Therefore, in recent years, scholars have conducted extensive research on how coastal cities and urban agglomerations cope with climate risks16,17,18,19. Reasonable urban spatial form can not only reduce the frequency and intensity of extreme climate events, but also enhance the resilience of cities to climate change, thereby adapting to climate change20,21. Consequently, to address the increasingly frequent and extreme climate disasters, exploring how to effectively integrate climate adaptation strategies into the spatial planning of coastal urban agglomerations has become a crucial area of current research. Existing studies on urban planning in response to climate change mainly focus on global warming and urban heat islands, the interaction mechanisms between climate change and urban systems22,23, resilience and adaptation policy formulation24,25, urban vulnerability assessments26, and prediction and simulation under different scenarios27,28. However, there is currently a lack of studies on the climate-adaptive development of coastal urban agglomerations and on how to rationally allocate space within these areas to adapt to urban climates. This gap significantly limits the sustainable development of coastal urban agglomerations and the formulation of relevant policy recommendations.

In recent years, to better optimize urban space allocation, some scholars have begun using land use simulation models to conduct quantitative simulations of urban space under different development scenarios. Commonly used methods include cellular automata (CA)29, CLUE-S model30,31, FLUS model32,33, all of which demonstrated high prediction accuracy and reliability and have been widely applied in various research fields. However, these models have certain limitations in the simulation process, such as the inability to deeply analyze the internal mechanism of land use change within a specific time frame and difficulty in simulating the patch growth process of multiple land use types on a fine scale. To address these shortcomings, the PLUS (patch-generating land use simulation) model offers an optimized approach34,35. The PLUS model is a land use simulation tool based on a planning support system36. By conversion analysis and pattern analysis strategies, it can identify the conversion patterns of different land use types in a more detailed way, and explore the trigger factors driving land use change in depth37,38. This capability allows for more accurate predictions of, future land use trends and spatial pattern evolution, thereby improving the reliability and application value of simulation results. In addition, with the refinement of urban development management, the prediction research on land expansion is no longer limited to the goal of climate adaptability, but has gradually been refined into diversified land expansion trend predictions such as farmland protection, water ecological sensitive areas protection, carbon neutrality, and risk resilience according to different urban development conditions, backgrounds, functional positioning3,39,40. However, the climate change risk factors facing cities are not single and isolated. Current research mainly focuses on the assessment of single risk factors, ignoring the compound impact of multiple disaster factors on cities in time and space. At the same time, the research that has been carried out has mostly focused on assessing the relationship between urban form and disaster factors, lacking prospective research from the perspective of climate adaptability.

Most of the ecological pressure faced by marine systems originates from land systems, and the urban ecological environment is significantly affected by LUCC (Land Use and Land Cover Change), which alters the spatial pattern of regional ecosystem services41,42. Given the internal and spatial correlations of ecosystem elements, the pressure caused by urban agglomeration development on marine systems should be alleviated, and solutions to improve the service quality of both marine and land ecosystems should be explored43,44. On the other hand, the impact of coastal disasters on human well-being is complex; coastal disasters such as storm surges can lead to loss of life, damage to property and infrastructure, and adversely affect the quality of life and well-being of residents45,46. Therefore, it is inadvisable to pursue urban construction in areas with a high probability of marine disaster impact47,48. Human activities should avoid these areas as much as possible, strengthen the early warning and assessment of coastal disasters, and explore construction models that improve the disaster resilience of coastal urban agglomerations.

The Guangdong-Hong Kong-Macao Greater Bay Area (GBA) is one of the most economically developed and densely populated urban agglomerations in China, covering less than 0.6% of the country’s land area while contributing 12% of China’s total economic output. However, a series of issues, including impaired marine and land ecosystem services, an imbalance in land use structure, and frequent extreme weather events, are becoming increasingly prominent. For instance, Macau has experienced four severe storm surge disasters in the past decade, resulting in casualties and economic losses amounting to hundreds of millions of dollars. Against the dual backdrop of extensive ecological capital overdraft and rapid urban space expansion, it is crucial to study the pattern of land expansion in coastal urban agglomerations within the GBA from the perspective of climate adaptation. In response to the growing challenges of climate change, this study develops an ecological-risk constrained land expansion modeling framework to simulate future land use patterns in the Guangdong–Hong Kong–Macao Greater Bay Area under multiple development scenarios (Fig. 1). By integrating spatial ecological and risk constraints into the PLUS model, and combining geospatial data with advanced statistical methods, we assess how land expansion aligned with climate adaptation goals can effectively balance ecological protection, disaster mitigation, and development demands. Unlike traditional studies that focus solely on physical land changes, this research emphasizes the roles of spatial resilience and climate risk in shaping land use decisions. The proposed framework enhances the capacity for predicting land use trajectories and designing targeted policy interventions. Ultimately, the findings offer scientific support for coordinated spatial planning, land management, and ecological governance in coastal urban agglomerations pursuing sustainable and climate-resilient development.

Fig. 1
Fig. 1
Full size image

Ecological-risk constrained land expansion modeling framework. The figure was created by the authors.

The remainder of the paper is organized as follows: Sect."Material and methods" describes the multi-source data and detailed methods, including spatial identification of ecological and disaster resilience, land use simulation scenarios, and simulation methods. Section.“Results” presents the research results, while Sect.“discussion” discusses the significance of these results and the factors influencing land use patterns in coastal urban agglomerations under different ecological and risk constraints. Section.“Conclusion” summarizes the main findings, planning strategies, and research limitations.

Material and methods

Study area

The Guangdong-Hong Kong-Macao Greater Bay Area (GBA) has the potential to become a world-class urban agglomeration and represents a new effort to promote a comprehensive pattern of openness in the new era49. GBA is located in southern China (Fig. 2), it covers a total land area of approximately 57,000 km2, with a population of around 86.17 million in 2020 and a regional GDP reaching $1.6688 trillion. Population growth has been a key driver of the GBA’s rapid economic expansion, but it also presents multiple challenges in the face of climate change. From 1980 to 2019, the sea level along the GBA coast has shown a fluctuating upward trend, with an average annual rise of 3.6 mm/year—higher than the average rate of sea level rise along China’s coast. The GBA’s terrain is predominantly low-lying, particularly in the Pearl River Delta, where much of the area is less than 1 m above sea level, and approximately 13% of the land lies below sea level. This makes the GBA one of the regions in China most vulnerable to sea level rise. The region faces increasing challenges from climate change, including more extreme weather events and damage associated with rising sea levels. Additionally, rapid urbanization has led to the conversion of large areas of farmland, wetlands, and forests into impervious urban surfaces. In the Pearl River Delta, urban construction land accounts for around 30% of the total land area, and in some areas, this proportion approaches 50%, severely weakening the land’s ability to retain and absorb water, thereby increasing the risk of urban flooding. These issues continue to threaten human life, property, and overall well-being. To mitigate the impact of inconsistencies between datasets, the study employed standardization techniques, such as rescaling and interpolation, to harmonize the data at comparable spatial and temporal scales. Specifically, city-level data from the PRD were adjusted for aggregation to match the regional and national datasets. This approach helps to ensure that variations due to statistical methods do not introduce significant biases into the analysis.

Fig. 2
Fig. 2
Full size image

Geographical location and land use map. The map was generated under fair use for non-commercial academic purposes. Base map data was obtained from OpenStreetMap (https://www.openstreetmap.org) and is reused under the CC BY license (https://www.openstreetmap.org/copyright). The base layer reflects OpenStreetMap data accessed in March 2025. Study area were added using ArcGIS 10.8 (Esri) under a Windows 10 environment. The final figure was created by the authors.

Ecological resilience spatial identification

Ecological resilience is closely tied to the functional stability and diversity of ecosystem services (ESs), which reflect the capacity of ecological spaces to maintain functionality under environmental disturbances50,51. ESs are defined as the various benefits that humans obtain directly or indirectly from ecosystems52,53,54. Following the Millennium Ecosystem Assessment framework, this study classifies ESs into four categories: Provisioning services (e.g., food production and freshwater supply), regulating services (e.g., carbon sequestration, water purification, climate regulation), supporting services (e.g., habitat provision and biodiversity maintenance), and cultural services (e.g., recreational, aesthetic, and spiritual values).

These services are spatially quantified through multi-source datasets and modeling outputs. Specifically, regulating services are evaluated using the Carbon and Blue Carbon modules of the InVEST model to capture the climate mitigation potential of coastal ecosystems. Provisioning and supporting services are represented using biodiversity indices and agricultural productivity data. For cultural services, spatial proxies are derived from scenic spots, defined as publicly accessible recreational or tourism-oriented sites such as parks, heritage sites, and waterfront attractions. These data are obtained through Python-based web scraping of POI (Point of Interest) information from major online mapping platforms, geocoded, and aggregated into raster layers to reflect spatial density and service intensity.

By integrating both the ecological and social attributes of ESs within the unique land–sea interface of coastal urban agglomerations, this study constructs a spatial framework for identifying ecological resilience using ArcGIS 10.8 and InVEST. The resulting ESs-based ecological source areas and resistance surfaces are subsequently used to inform scenario-based land expansion simulations under climate adaptation goals.

In this study, the InVEST (Integrated Valuation of Ecosystem Services and Tradeoffs) model serves as a core tool for assessing ecosystem services and establishing spatial ecological constraints(https://naturalcapitalproject.stanford.edu/software/invest). Two modules are primarily used: The Blue Carbon Model, which quantifies the carbon storage capacity of coastal ecosystems by calculating carbon gains and losses across global carbon pools, and the Linkage Mapper, which evaluates ecological connectivity across terrestrial and marine ecosystems to delineate ecological corridors.

To run these models, we input multi-period raster datasets reflecting climate-related disturbances, shoreline type changes, and land use transitions. The Blue Carbon module outputs are used to map regulating services, particularly those linked to climate mitigation potential in coastal zones. The Linkage Mapper identifies least-cost ecological pathways, enabling the construction of a continuous ecological network. These outputs are further combined with spatial proxies of provisioning, supporting, and cultural services—including biodiversity indices and scenic POI data—to generate a composite ecological resilience map.

The resulting raster layers represent spatial gradients of ecological resilience and serve as constraint inputs in the PLUS land use simulation model. In this way, the ecological data produced by InVEST not only reflect ecosystem functionality and spatial stability but also provide critical spatial boundaries for scenario-based urban expansion planning under climate adaptation goals.

Ecological patches identification

Ecological sources are landscape types that facilitate ecological processes and are crucial for the stable development of the ecological environment. We extract ecological sources by quantifying ESs, which include provisioning services, regulating services, cultural services, and supporting services. Regulating services are calculated using the Carbon and Blue Carbon modules of the InVEST model, the formula is as follows:

$${\text{CS}}_{\text{i}}={\text{C}}_{\text{above}}+{\text{C}}_{\text{below}}{+\text{C}}_{\text{soil}}{+\text{C}}_{\text{dead}}$$
(1)

\({\text{CS}}_{\text{i}}\) represents the total carbon storage, while \({\text{C}}_{\text{above}}, {\text{C}}_{\text{below}}{,\text{ C}}_{\text{soil}}{, \text{C}}_{\text{dead}}\) represent aboveground carbon storage, belowground carbon storage, soil carbon storage, and dead biomass carbon storage, respectively.

Supporting and provisioning services are represented by raster data on biodiversity and food production functions. Cultural services are derived by using Python to scrape POI data for “scenic spots” and converting it into spatial locations. After normalizing the four basic ESs in the GBA, a comprehensive ESs evaluation spatial map is generated through overlay analysis. Finally, ecological patches with high ESs importance and an area larger than 50 km2 are selected as ecological sources. Ecosystem services are categorized into four levels using the natural breaks method, with the highest category representing areas with the greatest ecological service potential, defined as high ecosystem services.

Resistance surface construction

The ecological resistance surface simulates the difficulty of species migration and quantifies the spatial distribution of resistance within the ecological security pattern55,56,57. Based on the characteristics of the GBA, we assign different resistance values to various land use types to quantify the ecological cost of material movement across different landscapes. The resistance value for water bodies, which are closest to natural habitats, is set at 5; forests, which minimally obstruct ecological material flow, have a resistance value of 20; grasslands, as artificially planted areas with limited vegetation diversity and subject to human activity, are assigned a resistance value of 70; unused land and construction land, being artificial land use types that obstruct biological information flow, are assigned resistance values of 180 and 200, respectively58,59. Considering the dual ecological and social attributes emphasized in the ES concept, this study uses nighttime light data to modify the basic ecological resistance surface. Finally, the comprehensive ecological resistance surface is constructed using the Minimal Cumulative Resistance (MCR) model60. The calculation formula is as follows:

$$\text{MCR}={\text{f}}_{\text{min}}\sum_{\text{j}=\text{n}}^{\text{i}=\text{m}}{\text{D}}_{\text{ij}}\times {\text{R}}_{\text{i}}$$
(2)

where, \(\text{MCR}\) represents the minimal cumulative resistance value for diffusion from ecological source patch j to a certain point; \({\text{D}}_{\text{ij}}\) is the spatial distance that the species crosses from ecological patch j to a certain point through base surface i; \({\text{R}}_{\text{i}}\) is the ecological resistance coefficient modified based on the nighttime light index.

Ecological corridor construction

We use the Linkage Mapper plugin on the ArcGIS platform to calculate the weighted cost distance between patches. Ecological corridors are constructed by extracting the raster connections that meet the requirements based on the least-cost path.

Landscape fragmentation

High-intensity development and construction activities by human societies continuously erode landscape patterns, particularly in coastal urban areas where landscape forms shift from continuous linear distributions to fragmented patchy distributions61,62. This transformation leads to reduced biodiversity, habitat fragmentation, and damage to ecological spaces. We further refine and revise the spatial pattern of ecological resilience using landscape fragmentation. Landscape fragmentation \(\left({F}_{i}\right)\)is usually represented by patch density \((\text{PD})\), reflecting the overall heterogeneity and degree of fragmentation of the landscape. The formula is as follows:

$${F}_{i}=PD=\frac{{NP}_{i}}{{A}_{i}}$$
(3)

where, \({NP}_{i}\) represents the number of patches of a landscape type, \({A}_{i}\) represents the total area of the landscape or patches.

Risk resilience spatial identification

We selected the Coastal Vulnerability Index (CVI) to assess the risk adaptability of coastal zones to sea level rise and storm surges (Table 1). The CVI comprises the exposure index, sensitivity index, and adaptability index63,64. Exposure and sensitivity indicate the likelihood of coastal populations and properties being adversely affected by natural disasters, while adaptability reflects the ability of coastal communities to mitigate the impact of such disasters65,66. The coastline vector layer, sea level digital elevation model, and POI data collected using Python are imported into the Coastal Vulnerability module of InVEST software to identify high-risk coastal areas with significant vulnerability. These areas with significant vulnerability are more difficult to cope with climate risks. When faced with marine disasters, they are prone to tight buffer space, land vulnerability, and potential risks of significantly increased disaster losses. Finally, the raster image of the coastal risk resilience avoidance zone is generated according to the simulation requirements of different scenarios.

Table 1 Data source.
Full size table

Simulation models

The integration of InVEST, GIS, and PLUS models is essential for addressing the multi-dimensional aspects of land use change, ecosystem services, and disaster resilience. InVEST provides a comprehensive assessment of ecosystem services, including carbon sequestration and water regulation, which are critical for understanding ecological impacts. However, InVEST alone cannot simulate land use transitions, which is where the PLUS (Predictive Land Use Simulation) model complements it (https://github.com/HPSCIL/Patch-generating_Land_Use_Simulation_Model). PLUS simulates land use changes under various policy scenarios, allowing for predictions on future spatial development. GIS supports both models by offering a spatial framework to integrate and analyze data at various scales, making it possible to assess both ecological and socio-economic factors within a unified system.

Each model serves a distinct purpose aligned with the study’s objectives: InVEST quantifies ecosystem services, PLUS simulates land use changes, and GIS provides the spatial infrastructure for integrating data from various sources. The combination of these models enables a comprehensive analysis of land use dynamics, ecosystem services, and disaster risk within a consistent spatial framework, enhancing decision-making in urban planning.

Compared to other land use simulation models like CLUE-S and CELLS, PLUS offers greater flexibility in incorporating diverse policy scenarios and spatially explicit ecological data. While CLUE-S is useful for regional land use projections, it tends to aggregate land use categories and may not capture fine-scale dynamics. Similarly, CELLS focuses on urban growth but lacks the integration of ecological and disaster risk factors, which PLUS provides. This makes PLUS particularly suitable for this study’s objectives of modeling land use changes in relation to ecological services and risk management.

LEAS module setting

The data required for input into the LEAS module include land use raster maps from two periods and land use driving factors. The former consists of land use raster maps with a resolution of 30 m for the years 2010 and 2020, derived from global 30-m surface coverage data (http://www.globallandcover.com). The latter involves the selection of 19 driving factors from five categories: Coastal zone conditions, socio-economics, physical geography, transportation location, and neighborhood factors, tailored to the regional characteristics of the Guangdong-Hong Kong-Macao Greater Bay Area (Table 2 and Fig. 3). The 19 driving factors were imported into SPSS (Statistical Package for the Social Sciences) software for multicollinearity testing. The tolerance values for all driving factors ranged from 0.120 to 0.788, and the variance inflation factor (VIF) values ranged from 1.303 to 8.316, all of which were within acceptable limits. Next, the table was imported into SPSS software to construct a binary logistic regression model to analyze the extent to which the driving factors influence changes in different land use types. The total factor method was used for variable selection during the statistical process, and all driving factors passed the p-value test, indicating significant contributions to urban spatial expansion (see appendix a). Finally, the map and driving factor raster data were imported into the LEAS module for random forest calculation. In this study, the uniform sampling method was selected, the number of decision trees was set to 20, the sampling rate was set at 0.01, and the number of driving factors and the characteristics of the training random forest were set to 19. This process yielded raster diagrams showing the development probability of each land use type, as well as the contribution values of each driving factor to changes in different land use types.

Table 2 Driving factor list.
Full size table
Fig. 3
Fig. 3
Full size image

Driving factors Layers. The final figure was created by the authors.

CARS module setting

The CARS module integrates"top-down"(land demand) and"bottom-up"(land expansion suitability) approaches and is a cellular automaton model that generates simulated land types using random seeds. Building on the traditional model, a simulation process based on dynamic driving factors is proposed to guide land changes and expansion to meet future land scale requirements. In the CARS module, we define restricted development areas and incorporate planning constraints (ecological constraint areas defined in Sect. 2.2, risk constraint areas defined in Sect. 2.3) into the spatial expansion simulation, excluding these areas from the process of urban spatial expansion simulation. By setting different constraints under multiple scenarios, we can compare the simulation results in terms of climate change response, ecological benefits, and economic benefits. The specific calculation process of the CARS module is as follows:

First, the feedback between land demand and spatial distribution suitability is measured. The basic formula for calculating the overall probability of land use type k can be expressed as:

$$O{P}_{i,k}^{d=1,t}={P}_{i,k}^{d=1}\times {\Omega }_{i,k}^{t}\times {D}_{k}^{t}$$
(5)

where, \({P}_{i,k}^{d=1}\) represents the growth probability of land use type k in grid cell i; \({D}_{k}^{t}\) is the adaptive driving coefficient, which represents the future demand for land use type k, which depends on the gap between the land use scale and land use demand before and after iteration; \({\Omega }_{i,k}^{t}\) represents the neighborhood effect of cell i, which is the coverage ratio of land use type k in the neighborhood, and the calculation formula is as follows:

$${\Omega }_{i,k}^{t}=\frac{\text{con}\left({c}_{i}^{t-1}=k\right)}{n\times n-1}\times {w}_{k}$$
(6)

where, \(\text{con}\left({c}_{i}^{t-1}=k\right)\) represents the total number of grid cells occupied by land use type k at the last iteration in the \(n\times n\) operation window, and \({w}_{k}\) the weight coefficient of different land use types.

Land use simulation scenario setting

The selection of the five land expansion scenarios—Business As Usual (BAU), Economic Development (ED), Ecology Protection (EP), Risk Aversion (RA), and Ecology and Risk Integrated Coordination (ERIC)—was guided by a combination of historical trends, policy goals, and ecological considerations. These scenarios were selected to encompass a range of potential futures based on current policy trajectories and regional development trends. The rationale behind selecting these specific scenarios lies in their ability to explore critical trade-offs between economic development, ecological protection, and disaster resilience. Unlike the widely used Representative Concentration Pathways (RCPs), which model climate change through greenhouse gas trajectories, our scenarios are tailored to regional policy orientations and simulate land use changes under spatial and governance constraints. Climate impacts are captured indirectly through indicators such as coastal vulnerability, blue carbon capacity, and landscape fragmentation, which reflect the spatial implications of climate change on land resilience. This framework enables a more practical and policy-relevant analysis of sustainable development strategies at the regional scale.

Business As Usual (BAU) simulates future land use based on the continuation of current development trends without introducing additional constraints. Land demand is projected using a Markov chain model to estimate patterns by 2035. Economic Development (ED) aims to maximize regional economic growth while satisfying basic ecological and spatial constraints. Ecology Protection (EP) prioritizes ecological conservation, focusing on maximizing ecosystem preservation under the same constraint conditions. Risk Aversion (RA) seeks to minimize regional disaster risk, particularly from storm surges, sea level rise, and coastal flooding, which pose serious threats to low-lying coastal areas. Regions with low resilience to these hazards, identified through the Coastal Vulnerability Index (CVI), are excluded from land expansion. Ecology and Risk Integrated Coordination (ERIC) balances economic development and ecological protection by integrating both ecological and disaster risk constraints. It aims to optimize land use layout while ensuring climate adaptation and sustainable spatial development.

For specific details on the calculation process and parameter settings, please refer to Appendix b.

Simulation accuracy verification

The PLUS model has been widely used to simulate different regions at different scales (regional, continental and global), and can obtain higher simulation accuracy than other traditional models such as CLUE-S, ANN CA and Logistic-CA67,68. The accuracy of PLUS simulation results is usually verified by using the Kappa coefficient and FOM (Figure of Merit) value. The calculation formula is as follows:

$${\text{Kappa}}=\frac{\left({P}_{0}-{P}_{c}\right)}{\left(1-{P}_{c}\right)}$$
(7)
$$\text{FoM}=B/(A+B+C+D)$$
(8)

where, \({P}_{0}\) represents the ratio of the number of correctly simulated grid cells to the total number of grid cells; \({P}_{c}\) is the expected consistency, that is, the proportion of grid cells simulated correctly under random conditions; \(A\) represents the wrong grid cells whose actual land use type has changed but not in the simulation map; \(B\) represents the correct grid cells whose actual land use type has changed in both the actual land use type and the simulation map; \(C\) represents the grid cells whose actual land use type has changed but whose prediction is wrong in the simulation map; \(D\) represents the wrong grid cells whose actual land use type remains unchanged but changes in the simulation map. Generally speaking, when the Kappa coefficient is greater than 0.80 and the FOM coefficient is greater than 0.2, the overall model fit is considered good and the simulation results are credible.

Results

Spatial distribution of ecological resilience

Regulating services are largely concentrated in the ecological barrier areas on both flanks of the GBA, corresponding to Zhaoqing, Huizhou and Jiangmen city, where forest land is concentrated. Supporting services are concentrated in Guangzhou, Shenzhen and Hong Kong, mainly benefiting from coastal wetland ecosystems such as mangroves and coral reefs. Supply services present a spatial pattern of"high in the east and low in the west", and cultural services are concentrated in Guangzhou, Shenzhen, Hong Kong and Macau (Fig. 4).

Fig. 4
Fig. 4
Full size image

Secondary ESs and spatial pattern of integrated ESs.

Based on the high-value areas of comprehensive ecosystem service efficiency, a total of 117 ecological source areas are identified, with a total area of approximately 11,600 km2, accounting for about 21.1% of the total area of the GBA. The resistance surface shows an obvious pattern of"high in the middle and low around". High-resistance value areas are mainly concentrated in the southwest of Guangzhou, Shunde District of Foshan, and the central and western parts of Dongguan. In these areas, construction land is relatively concentrated and far from large-scale ecological source areas (Figs. 5-a and 5-b). The ecological corridors with less resistance in GBA amount to 37, and the core corridors are distributed in a circular pattern in the areas surrounding Guangzhou, such as Zhaoqing, Huizhou, and Jiangmen. Forest land and grassland jointly constitute a corridor landscape type featuring high ecological connectivity. The spatial extension direction is essentially in accordance with the distribution of ecosystem service efficiency (Fig. 5-c). Besides, the landscape fragmentation degree in central cities is generally high. Specifically, there are extensive areas of landscape fragmentation in the center of Guangzhou and Foshan City, which extend outward along major traffic arteries. This implies that the interaction between human activities and the ecological environment is intense. Traffic arteries impede the natural activity range of organisms, and the overall habitat quality and ecosystem service functions are under significant pressure (Fig. 5-d).

Fig. 5
Fig. 5
Full size image

Spatial distribution of ecological resilience. Ecological sources and corridors, nighttime light intensity, resistance surface and landscape fragmentation are shown in (a), (b), (c), (d), respectively, and the final superimposed and generated resilience spatial division is shown in (e).

From the perspective of the overall spatial distribution pattern, ecological spaces with low resilience radiate in a cross shape along the Pearl River Hongqili Waterway and Xijiang River The urban development clusters in Foshan, Guangzhou, and Dongguan, which largely overlap with areas experiencing significant loss of ecosystem services and high landscape fragmentation, are notably affected. Regarding spatial morphology, non-resilient ecological spaces are most concentrated, primarily in Jiangmen City, Bao’an District of Shenzhen City, Baiyun District of Guangzhou City and Shunde District of Foshan City, and are more likely to appear at the junctions of urban divisions (Fig. 5-e). Low-resilience ecological spaces are predominantly found in urban–rural fringe areas, spreading outward along major roads and both sides of significant rivers. Medium-resilience ecological spaces account for the largest proportion and are fragmented, spanning across various ecological units. Highly resilient ecological spaces are relatively dispersed within ecological barrier areas such as Zhaoqing and Huizhou, where land use types are mainly concentrated in unconverted forests and waters.

From the perspe2ctive of urban agglomerations, there is evident spatial heterogeneity in ecological spatial resilience among cities. Among the four central cities, Guangzhou faces the most severe ecological space resilience issues, with the proportion of low-resilience and non-resilient ecological spaces exceeding 80%. Hong Kong has the lowest proportion of low-resilience ecological spaces. Among other node cities, Huizhou and Zhaoqing exhibit relatively low proportions of low-resilience ecological spaces, both below 50%, and no large areas of non-resilient ecological spaces have emerged yet. This indicates that the mountain forest ecological barrier in the northern part of the urban agglomeration has provided a significant ecological function.

Spatial distribution of risk resilience

The results of the exposure index, sensitivity index, and adaptability index are shown in Figs. 6-a and 6-b. The natural habitat of the GBA is primarily mangroves, with large areas distributed in the Mangrove Nature Reserve in Futian District, Shenzhen; the west coast of Tin Shui Wai in the Hong Kong Special Administrative Region; Qi’ao Island; and the east coast of Jiangmen City. The overall spatial morphology is fragmented into patches along the coastal zone. Based on the distribution of GBA coastal transportation and medical facilities, it is evident that transportation facilities in the Macao Special Administrative Region, the Hong Kong Special Administrative Region, and Shenzhen are relatively concentrated in coastal areas. In the event of a typhoon storm surge, transportation facilities in these vulnerable coastal areas face a high risk of paralysis.

Fig. 6
Fig. 6
Full size image

Spatial distribution of risk resilience. (a) and (b) respectively show some layers for calculating the Sensitivity Index and Adaptability Index, (c) shows the risk resilience classification results, (d) and (e) respectively show the risk resilience limiting factors of the RA scenario and the ERIC scenario. The difference between them is that the land expansion in the RA scenario will avoid spaces with no resilience and low resilience, while the land expansion in the ERIC scenario will only avoid spaces with no resilience. The map was generated under fair use for non-commercial academic purposes. Base map data was obtained from OpenStreetMap (https://www.openstreetmap.org) and is reused under the CC BY license (https://www.openstreetmap.org/copyright). The base layer reflects OpenStreetMap data accessed in March 2025. Study area were added using ArcGIS 10.8 (Esri) under a Windows 10 environment. The final figure was created by the authors.

According to the final superimposed risk resilience spatial identification results, the resilience space of medium and high-risk accounts for about 19.4% (10864km2), coastal areas such as Tolo Harbour in Pak Shek Kok, Hong Kong, and Ma On Shan New Town face potential marine disaster risks (Fig. 6-c). Their high-density residential and recreational functions exacerbate the impact of marine disasters, likely leading to significant economic losses. In contrast, the coastlines along the south bank of Zhuhai’s Jinwan District and Gaolan Island are less vulnerable to marine disasters due to ongoing large-scale construction activities and the lack of complete infrastructure and living conditions, making these areas less sensitive to marine risks. However, during future urban development, it is crucial to consider the threat of marine disasters to these coastal zones and adjust land expansion plans proactively to avoid areas with low-risk resilience.

To align with the scenario-setting parameters of land expansion simulation and to explore the impact mechanism of disaster resilience on urban spatial expansion, risk areas with medium and low resilience are further delineated as land expansion avoidance areas in the RA scenario (Fig. 6-d). These include coastal areas with low elevation and relatively flat terrain, such as the eastern part of Hong Kong, the western part of Bao’an District, Shenzhen, and Hengqin Island. Additionally, the development of coastal urban agglomerations should ensure a safe and stable coastal spatial pattern while actively promoting the blue economy. Consequently, only the low-resilience risk areas are delineated as land expansion avoidance areas in the ERIC scenario, and the impact of comprehensive coordination of ecological-risk restricted areas on land expansion is explored based on the simulation results (Fig. 6-e). Compared with the RA scenario, the risk avoidance area of ​​ERIC excludes areas such as Tanzhou and Shazhou coasts in Dongguan City and the southern bank of Jinwan District in Zhuhai City.

Analysis of land expansion simulation results

Analysis of simulation results of multiple scenarios

The model validation process involved simulating land use changes in the GBA for 2020 using 2010 as the baseline year, and comparing the simulated results with the actual 2020 land use map. Validation was performed using the Kappa coefficient and overall accuracy, with the model achieving a Kappa coefficient of 0.78 and an overall accuracy of 84.6%, and a FoM of 0.28, indicating good predictive performance. Initially, the Kappa coefficient was 0.62, and the overall accuracy was 72.8%, suggesting moderate model performance. Three main issues were identified and corrected: First, the conversion of built-up land and water bodies, where the initial model predicted expansion in agricultural and grassland areas, while actual expansion in 2020 occurred largely in water bodies; this was addressed by redefining coastal reclamation areas as agricultural or grassland zones. Second, the model overestimated the expansion of built-up land in Hong Kong’s Tuen Mun and Fanling areas, which was corrected by restricting the conversion of grassland to built-up land using ArcGIS. Third, the model initially predicted concentrated urban expansion, while actual expansion was more dispersed and occurred over a wider area; this was adjusted by fine-tuning the neighborhood size and expansion coefficient parameters. After these corrections, the Kappa coefficient improved to 0.78, and the overall accuracy increased to 84.6%, demonstrating that the revised model accurately simulates land use expansion in the GBA urban agglomeration and provides a reliable basis for future spatial planning decisions69.

The distance to mangrove areas, proximity to urban centers, and nighttime light intensity are critical drivers of urban land expansion (Fig. 7). The distance to mangroves reflects ecological constraints, limiting expansion in sensitive areas. Proximity to urban centers indicates the influence of concentrated economic activity and infrastructure, facilitating expansion in adjacent areas. Nighttime light intensity, as an indicator of urbanization, highlights regions with high human activity and population density. Together, these factors shape the spatial patterns of land expansion, particularly in scenarios where ecological protection is not prioritized, with economic growth and urbanization playing a significant role in driving land use change.

Fig. 7
Fig. 7
Full size image

Impact of driving factors on construction land change. The final figure was created by the authors.

Figures 8 and 9 show the simulation results and the land areas of various types under different scenarios, respectively. Under the BAU scenario, the expansion of construction land exhibits a distinct"core–edge"pattern, with the original construction land serving as the core and expanding outward along the main urban roads. This results in a generally disorderly and fragmented expansion. The conversion of construction land is characterized by large-scale encroachment on surrounding cultivated land and water areas, with reductions in forests and grasslands as well. This type of expansion scenario exacerbates the fragmentation of the urban agglomeration landscape, potentially leading to issues such as resource scarcity and diminished ecosystem service functions.

Fig. 8
Fig. 8
Full size image

Simulation results of land expansion under multiple scenarios. The final figure was created by the authors.

Fig. 9
Fig. 9
Full size image

Area of different types of land use in each scenario. The final figure was created by the authors.

The ED scenario represents the largest scale of construction land expansion. Cities with high development potential, such as Shenzhen, Foshan, and Dongguan, are experiencing rapid growth. Notably, the construction of land in national-level development zones such as Dongguan Songshan Lake High-tech Industrial Development Zone, Foshan High-tech Industrial Development Zone, and Qianhai Shenzhen-Hong Kong Modern Service Industry Cooperation Zone has increased significantly. Land use is inefficient, and the existing stock of construction land is not fully utilized. Nearly all water areas and cultivated land within the development zones are repurposed for urban development and construction.

Under the EP scenario, the expansion pattern of construction land involves inward infill. Constrained by boundary conditions such as nature reserves and ecologically damaged areas, the main forest and grassland areas, water bodies, and other ecological spaces within urban agglomerations are strictly protected. This approach helps alleviate the conflict between habitat protection and urban development.

Under the RA scenario, avoiding large areas of low-resilience risk space leads to the expansion of land use inland. Qiashui County, Lengkeng Town, and Deqing County in the northwest suburbs of Zhaoqing City have experienced a significant increase in construction land. Strictly controlling the direction of land expansion to avoid marine disasters may pose a threat to inland urban agricultural production and ecological protection.

Under the ERIC scenario, the distribution of newly added construction land is more balanced, with infill growth being the primary type. A small amount of construction land is concentrated along the main roads of urban agglomerations. The overall land use structure and layout of urban agglomerations are more compact, which, to some extent, balances the needs for ecological protection and risk adaptation.

Analysis of expansion hot spots and expansion differences

Identifying land expansion hotspots can help better balance the relationship between development and protection. Figure 10-a shows the hotspots of urban construction land expansion under different scenarios. The areas where expansion occurs in all five scenarios can be considered hotspots of future urban spatial expansion, covering approximately 10,462.15 km2. Excluding the existing construction land of 9,088.19 km2 in 2020, the area of urban spatial expansion hotspots will be approximately 1,373.96 km2 by 2035. The newly added construction land is distributed around the existing construction land in a marginal expansion mode, primarily in the central area of the urban agglomeration, such as Baiyun District of Guangzhou, Nanhai District of Foshan, Longhua District of Shenzhen and the junction of Dongguan City. The hotspots of construction land distribution demonstrate an intensive development model. The overlapping area of newly added construction land in the four scenarios is about 1,656.36 km2, while the overlapping area of the three scenarios and two scenarios are 1,395.89 km2 and 2,046.43 km2, respectively. According to the spatial distribution characteristics of construction land, the overlapping area of multiple scenarios shows a trend of expanding toward the southeast, distributed in Longhua District, Luohu District, Bao’an District of Shenzhen and Zhangmutou Town, Songmushan Reservoir and its surrounding areas of Dongguan, with the transformation of construction land primarily involving the erosion of cultivated land and water areas.

Fig. 10
Fig. 10
Full size image

Expansion hot spots and expansion differences. (a) shows the overlapped areas of new construction land under multiple scenarios, and (b) shows the difference and overlapped areas of construction land expansion under different scenarios. The map was generated under fair use for non-commercial academic purposes. Base map data was obtained from OpenStreetMap (https://www.openstreetmap.org) and is reused under the CC BY license (https://www.openstreetmap.org/copyright). The base layer reflects OpenStreetMap data accessed in March 2025. Study area were added using ArcGIS 10.8 (Esri) under a Windows 10 environment. The final figure was created by the authors.

Further analysis of the expansion difference areas of typical scenarios will help identify the layout of construction land under different constraints (Fig. 10-b). Compared with other scenarios, the expansion difference of the ED plan is the most significant, with an area of approximately 1,433.02 km2, mainly distributed in small suburban areas with high accessibility, such as Huaiji County in Zhaoqing, Conghua District in Guangzhou City, and Longmen County in Huizhou City. The expansion is occurring outward along the original urban construction land, with the conversion of construction land primarily affecting cultivated land. The expansion difference area of the ERIC plan is 562.79 km2, mainly distributed on both sides of the Xijiang River in Fengkai County, Zhaoqing City, on the south bank of the Dongjiang River in Boluo County, Huizhou City, and in Enping City, Kaiping City and other county-level cities along the G15 Shenhai Expressway.

Comparison of ecological and economic benefits in multiple scenarios

To further quantify the differences in ecological and economic benefits under various scenarios, we calculated two types of indicators: Ecosystem Service Value (ESV) and land use economic benefits. Using the BAU scenario as the baseline, we assessed the changes in ecological and economic benefits under other scenarios (Fig. 11). Compared with the BAU scenario, the ED scenario yielded the highest economic gain (25.91%) but resulted in a notable ecological loss (− 2.32%), reflecting a growth-oriented strategy with limited environmental safeguards. Conversely, the EP scenario prioritized ecological preservation, achieving a 3.17% gain in ecological value at the cost of a 6.47% reduction in economic output. The RA scenario produced moderate improvements in economic benefits and slight ecological degradation. Among all scenarios, the ERIC scenario demonstrated the most balanced outcome, combining a substantial economic benefit (15.89%) with the smallest ecological loss (− 1.21%). To quantify this trade-off, we calculated a balance index, defined as the arithmetic mean of the relative changes in ecological and economic values. While the ED scenario showed the highest numerical index (11.80%), its ecological degradation undermines its suitability as a sustainable option. In contrast, the ERIC scenario achieved a balance index of 7.34%, the highest among scenarios that maintained both ecological and economic viability, followed by RA (0.60%) and EP (− 1.65%). This suggests that dual-constraint spatial strategies can effectively harmonize development demands with ecological sustainability in climate-sensitive coastal urban regions.

Fig. 11
Fig. 11
Full size image

Comparison of ecological and economic benefits. The final figure was created by the authors. *All values are percentage changes relative to the BAU scenario.

Discussion

A framework for land expansion simulation under ecological-risk constraints

Formulating an optimal land expansion plan from the perspective of climate adaptation is a critical task in planning for the GBA coastal urban agglomeration. This task typically involves balancing the multiple needs of urban development with climate adaptation. However, many land expansion simulation and prediction studies rely on existing land use development models and often fail to comprehensively consider urban development policies and climate change goals. As a result, their predictions may not achieve optimal adaptation effects. This paper proposes a framework for land use simulation under ecological-risk constraints, which can assist decision-makers in developing urban planning strategies that effectively address complex climate change challenges. The study employs InVEST and GIS coupling technology to evaluate terrestrial ecosystem services and coastal blue carbon effects, identify ecological resilience spaces, and analyze intensive land expansion patterns. Subsequently, based on the areas identified as vulnerable to coastal hazards—including natural threats such as storm surges, sea level rise, and coastal flooding, as well as exposure-related social factors—by the Coastal Vulnerability module of the InVEST model, the risk resilience space is delineated to enhance the safety and climate adaptability of land expansion. Finally, the identification results of these two components are used as constraints (i.e., avoidance areas for land expansion) and incorporated into the PLUS model to conduct multi-scenario land expansion simulations and characterize future land development under varying climate adaptability conditions. This framework provides a technical basis for cross-studying climate adaptability and land expansion in coastal urban agglomerations and aids in guiding the symbiotic integration of complex coastal urban systems and the compact and safe development of urban space.

Ecological and economic benefits of future land expansion patterns

The simulation results indicate that construction land under different scenarios expands outward from the center of the urban agglomeration. However, there are notable differences in land expansion under various constraints. Comparing the construction land area between the ED scenario and the ERIC scenario, it is evident that when ecological protection and disaster adaptation are insufficiently considered, cultivated land and coastal areas in Zhaoqing City, Huizhou City, and Jiangmen City become hotspots for land expansion. As climate adaptability constraints increase, the expansion of construction land in coastal cities such as Jiangmen, Zhuhai, Shenzhen, and Huizhou significantly decreases, shifting towards inland areas with relatively low ecological protection intensity70. From the perspective of balancing ecological and economic benefits, the ERIC scenario provides a better equilibrium between the two, with ecological benefits being second only to the EP scenario and economic benefits second only to the ED scenario. In other words, while the ERIC scenario sacrifices some economic benefits, it compensates by achieving certain ecological benefits and optimizing urban agglomeration land use. Nonetheless, this scenario still requires further exploration of intensive and efficient land use models.

Recommendations for optimizing land use development patterns

The simulation results indicate that GBA ecological land will continue to experience certain losses in the future. Therefore, future land development should prioritize the optimal allocation of ecological service-carrying capacity alongside economic development goals. For instance, in urban areas prone to inefficient expansion, such as the Shunde District of Foshan City and Baiyun District of Guangzhou City, high-resilience ecological spaces identified through ecological space resilience evaluations should be incorporated into the ecological protection red line, serving as a rigid boundary for urban growth. The“Blue Bay”remediation actions should be sustained in coastal areas such as the Pearl River estuary, Bao’an District of Shenzhen City, and Taishan City of Jiangmen City71. Mangroves should be used as natural barriers to prevent coastline erosion, and efforts should be made to promote ecological restoration along the urban agglomeration’s coastline. Additionally, Shenzhen Bay is a key coastal area where the GBA city merges with the sea and is a crucial migratory bird corridor72. Public spaces facing the sea should be carefully planned to protect natural habitats and migratory corridors. Coastal public space design should prioritize maintaining adequate setbacks and enhancing the ecological quality of coastal zones. While natural resources like the Futian mangroves and Dasha River corridor provide valuable ecological services, future studies could explore how such natural assets can be integrated into urban planning to enhance the harmony between human settlements and nature. For the restoration of ecological spaces in highly urbanized areas, it would be beneficial to further investigate strategies such as creating ecological corridors to improve connectivity between green spaces and enhance ecosystem services. Approaches like those used in Hong Kong’s ecological space management could offer valuable insights for improving urban livability, although more localized research is needed to adapt these methods to specific urban contexts.

In addition, disaster resilience in the coastal zone should be continuously evaluated and optimized. While engineering measures such as coastal embankments, drainage systems, and sluices are commonly employed in coastal protection, future research should explore how such measures can be integrated into land-use planning and spatial development strategies to enhance long-term resilience against coastal hazards like storm surges. Further studies could focus on assessing the effectiveness and environmental impacts of these interventions in specific coastal areas. However, the coastlines along the GBA are diverse, and large-scale projects may pose certain risks to the ecological environment. Therefore, adaptive countermeasures should be considered based on the goals of ecological civilization, a concept that emphasizes sustainable development by integrating ecological protection with economic growth. These measures should align with the characteristics and resilience of different types of coastlines. For example, for low-resilience tourist coastlines, we can develop an ecologically resilient sea boardwalk that serves both recreational and disaster relief purposes. When there is no storm surge warning, residents and tourists can enjoy the boardwalk. Additionally, wind-resistant and water-resistant aquatic plants can be planted in the nearshore area. During a storm surge, tide-proof seawalls can help mitigate flooding and protect the lives and property of nearshore residents. For agricultural and fishing coastlines, we should actively promote the protection and restoration of oyster reefs, which can attract marine species such as shellfish and fish, enhance marine biodiversity, and provide sustainable fishery resources for coastal residents.

Problems and prospects

First, the research scope covers nine cities and two special administrative regions (Hong Kong and Macao) in the GBA but lacks a complete set of unified basic data and planning data. The basic data for the nine cities in the Pearl River Delta are collected from relevant yearbook materials, social and economic development bulletins, master plans, and other materials of Guangdong Province. However, some provincial-level data are not subdivided into the Pearl River Delta city cluster level or city level. Consequently, some data are estimated based on historical development trends and the proportion of Pearl River Delta data. Additionally, there are differences in the statistical methods used for economic and social data between mainland Hong Kong and Macao, which may lead to deviations in research results during the data cleaning process. The use of basic data with unified standards and methods would help produce more accurate research results and is an important direction for achieving regional coordinated development in the GBA. To achieve refined management of coastal urban agglomerations, future efforts should focus on deepening quantitative data and technical methods to develop a multi-spatial scale, multi-disciplinary integrated decision-making toolbox. Furthermore, although this study explores land expansion scenarios and associated flood risks, the consideration of future climate scenarios was not included in the flood risk analysis. Future climate models, such as Representative Concentration Pathways (RCPs), could provide more comprehensive insights into potential flood risks under various climate change scenarios. This would help in refining risk management strategies for urban development in the face of climate change.

Conclusion

This study developed an ecological-risk constrained land expansion modeling framework to analyze the land expansion patterns of the coastal urban agglomeration in the Guangdong-Hong Kong-Macao Greater Bay Area from the perspective of climate adaptation. The land use simulation results under various scenarios were evaluated from multiple dimensions, including economic benefits, ecological benefits, and land expansion hotspots. The main conclusions are as follows:

Construction land shows a trend of expanding outward from the center of the urban agglomeration across different scenarios, but the expansion patterns vary depending on the constraints applied. In scenarios that do not consider ecological protection and disaster adaptation, such as the ED scenario, cultivated land and coastal areas in node cities like Zhaoqing, Huizhou, and Jiangmen become hotspots for land expansion. However, as climate adaptability constraints increase, the expansion of construction land in coastal cities such as Jiangmen, Zhuhai, Shenzhen, and Huizhou has significantly decreased, with the expansion area gradually shifting to inland regions with relatively lower ecological protection intensity.

While the ED scenario yields the highest economic benefits, it falls short in ecological performance. In contrast, the ERIC scenario strikes a balance, delivering superior ecological benefits compared to both the ED and RA scenarios, though it sacrifices some economic benefits. This highlights the trade-off between ecological protection and economic growth, especially in climate adaptation planning.

The urban spatial layout analysis under multiple scenarios reveals significant spatial heterogeneity. For instance, the EP scenario favors inward filling, while the RA scenario drives expansion inland to avoid low-resilience risk zones. These findings emphasize the necessity of coordinating ecological protection and risk management strategies to ensure compact, resilient, and sustainable urban growth.

The simulation model is subject to multiple sources of uncertainty, including input data errors, simplified assumptions, and parameter selection. For instance, variations in land cover data or future policy assumptions may significantly alter land use predictions. Future work should conduct sensitivity analyses to identify key drivers of model outcomes and enhance result robustness. Additionally, incorporating high-resolution data, modeling regional dynamics, and considering socio-economic factors—such as migration, economic shifts, and governance—will help reduce uncertainties and improve the model’s applicability under climate change scenarios.

In summary, this study provides a comprehensive land use simulation and assessment framework for coastal urban agglomerations facing climate change and rapid urbanization. This framework not only helps optimize urban spatial layout but also assists in finding a balance between economic and ecological benefits, thereby offering strong support for future policymaking and urban planning.