Introduction

In February of 2019, the Chinese government issued the Outline of the Development Plan of the Guangdong–Hong Kong–Macao Greater Bay Area (GBA), which proposed transforming this region into an important support for the Belt and Road Initiative and another world-class bay area city cluster modeled after the Tokyo Bay Area in Japan, as well as the New York Bay Area and San Francisco Bay Area in the United States. The GBA urban agglomeration, located in the Pearl River Basin in China, currently holds an important strategic position in the overall development of the country and has attracted remarkable attention from the academic community (Hui et al., 2018). Specifically, many scholars have conducted studies on the urban agglomeration of the GBA (Liu et al., 2019; Weng et al., 2020), providing theoretical and decision-making support for planning and construction in this area. As the urbanization process in the GBA continues to accelerate, the proportion of the urban population continues to rise, and urban space continues to expand outward. However, while promoting the rapid development of the social economy, accelerated urbanization also contributes to a series of ecological and environmental problems, such as resource shortages, environmental pollution, ecosystem degradation, and a decline in ecosystem service functions, which have become key research areas in urban ecology and geology, as well as national and local planning (Chen et al., 2020; Sun et al., 2018). Considering the demand for harmonious development between humans and nature, it is becoming increasingly important to balance the conflict between cities and ecosystems and ensure sustainable development (Song et al., 2018).

Urban expansion refers to changes in the total amount and direction of urban construction land under the combined effect of economic development, population growth, urbanization, and other factors (Fang et al., 2018). In recent years, as the urbanization process has begun to move toward the urban agglomeration model of overall regional development, the expansion of urban agglomerations has become a key research focus. The synergy of urban agglomeration spatial expansion not only reflects the spatial growth pattern of urban agglomeration (Herold et al., 2002; Almeida et al., 2005) but also reflects the land-use problems in large-scale regions (Ramachandra et al., 2014). The process of urban agglomeration expansion is often associated with uncontrolled urban land demand and the disorderly development of urban fringe (Menhem et al., 2017; Giuliano et al., 2017), resulting in an increasingly prominent contradiction with the scarcity of land resources (Saadani et al., 2022). With the total amount of land resources unchanged, the key to solving these problems lies in optimizing the land-use structure of urban agglomeration and promoting coordinated development (Dutta and Das (2022)). At present, the urban agglomeration in the GBA is in a critical period of development and transformation, strengthening the radiative driving capacity of the central city and cultivating new growth poles. The coordinated and orderly development of urban agglomerations in the GBA is urgently required to weaken the differences in regional development scale and improve the overall level of urban agglomerations. Therefore, it is necessary to analyze and summarize the laws of spatial expansion and internal coordinated development of urban agglomeration in the GBA of Guangdong, Hong Kong, and Macao, enhance the synergy of land use within urban agglomeration, and implement the sustainable use of land resources.

Nighttime light data, land-use data, socioeconomic statistics, and GIS spatial analysis methods are frequently used to analyze the spatiotemporal dynamics of urban agglomeration expansion according to indicators such as the area of urban expansion, the landscape expansion index, and the landscape pattern index (Fang and Yu (2017)). Furthermore, the mechanisms driving urban agglomeration expansion have been explored through multiple stepwise regression, spatiotemporal geographical weighted regression, and generalized additive models (Gao et al., 2017). Previous research combined the cellular automata (CA) and future land-use simulation (FLUS) models with machine learning and other methods to simulate urban agglomeration expansion and provide early warnings (Yang and Liu (2005); Chen et al., 2020). The ecological, environmental, and economic effects of urban agglomeration expansion can be measured by assessing the value of ecosystem services and building a comprehensive evaluation model of ecological and environmental effects using the Capello model and other methods (Kuang et al., 2013; Dietzel and Oguz (2005)). Existing land-use simulation research typically involves estimating the land-use change mode, process, and driving mechanisms and can be generally divided into two categories: simulation of the urban land-use scale and prediction of urban spatial expansion (Zhou et al., 2020). The expansion scale of a city can be determined from the speed, direction, and spatial differentiation characteristics of urban expansion (Zhou et al., 2019). Subsequently, urban spaces can be simulated and predicted from multiple perspectives, such as the population and environment, intensive and economic resource use, social and economic development, production, the standard of living, and ecological harmony (Zhao et al., 2021). Moreover, data related to the flow of information, logistics, people, and other elements within a city can be employed to explore the functional links between cities and the flow of elements (Beckmann, Mcpherson (2010)). With the rise of network science, scholars have also explored the evolution of the urban agglomeration spatial structure and functional center (Taylor et al., 2010) using social network analysis and multicenter measurement. For instance, the gravity model and technique for order of preference by similarity to ideal solution evaluation method were used to analyze the evolution of the location degree within an urban agglomeration (Jacobs et al., 2011). Another study analyzed the spatial evolution of urban agglomerations based on data such as nighttime light and Landsat remote sensing images, revealing the direction and intensity of urban expansion (Derudder et al., 2008). Similarly, models such as CA and spatial sentences have been used to explore historical changes in urban expansion and simulate future urban expansion (Van Vliet et al., 2017; Van Vliet (2019)). The study of urban agglomeration expansion is the basis of the study of spatial synergy. The existing research mainly focuses on the spatial form and mode of urban agglomeration, the analysis of element coupling and coordination, and the analysis of spatial association. Spatial expansion not only refers to the growth of boundaries but also reflects the diffusion of socioeconomic factors in space. Therefore, the distribution and flow of various development factors in urban agglomeration will impact the spatial development of urban agglomeration. At the same time, the radiative driving effect of the core city will promote the development of surrounding cities. The greater the gravity intensity of the core city, the greater the area and intensity of the regional spatial expansion, showing that the above three aspects of research are not independent but interconnected and mutually affected. The majority of previous research methods were initially based on the system and rational growth theories and then adopted the coupling coordination and urban gravity models to analyze the relationship between the spatial expansion process of urban agglomeration and population, economy, and social factors to demonstrate how to achieve rational, orderly, and coordinated growth of urban agglomeration. Over time, more studies have relied on nighttime light data to monitor the expansion dynamics of urban agglomeration based on data characteristics that are closely related to the regional economy and population distribution in a large range.

Generally, research on the spatiotemporal evolution of urban agglomerations has focused on the evolution of either economic or spatial patterns, ignoring the important interaction between these patterns. In the study of urban expansion, the traditional CA model offers considerable advantages over other models in simulating the progression of urban evolution. However, this model is mostly used to simulate the expansion of a single city, and it lacks consideration of the spatial interaction of cities within urban agglomerations. Furthermore, the CA model is not suitable for simulating the comprehensive expansion process of urban agglomeration at the macro scale. In contrast, the CA–Markov model is the most widely adopted simulation prediction model among researchers worldwide. This model can not only predict the total amount of transfer and the transfer probability matrix of various ecological space types but also simulate and predict the spatial distribution of ecological space types through the analysis of neighboring regions. However, due to the complex nonlinear relationship between ecological space and its impact factors, it is difficult to define this relationship and determine the conversion rules using the CA–Markov model. On the other hand, the artificial neural network (ANN) model is excellent for dealing with complex nonlinear relationships through machine learning (Ahmadizadeh et al., 2014; Saputra and Lee (2019)). Therefore, combining the ANN and CA–Markov models to construct the ANN–CA–Markov model compensates for the shortcomings of each individual model and has great research value for the simulation and prediction of ecological space (Singh et al., 2015).

Moreover, although the FLUS model can simulate the spatial dynamics of multiple land-use changes and their interactions, it does not currently simulate the spatial pattern of land use under various scenarios (Liu et al., 2017; Song et al., 2020). However, setting different scenarios with constrained conditions is an effective tool for geospatial simulation, spatial optimization, and decision-making (Saadani et al., 2020; Meng et al., 2021). In this study, we first analyzed the spatiotemporal evolution of urban expansion in the GBA urban agglomeration. Subsequently, based on the FLUS model, we integrated macroscale land-use demand predictions and local change distributions to simulate the interaction of land-use changes and their spatial dynamics in the GBA urban agglomeration under three different scenarios (natural development, ecological protection, and economic development) from 2025 to 2035. This study aimed to explore the basic trends of urban agglomeration expansion, reveal the impact of urban expansion on landscape patterns, provide a theoretical and practical basis for land-use planning, and produce a valuable reference for ecological protection and high-quality development in the Pearl River Basin. The ANN–CA–Markov model constructed in this study referring to the achievements of previous scholarshas with a high-simulation accuracy (Wei et al., 2022). With the CA model as the main framework, combined with the Markov and ANN models, it effectively solves the shortcomings of the CA model. It can not only predict the number and probability of the transfer of urban ecological space types but also account for the complex nonlinear relationship between the impact factors and urban ecological space changes to predict the distribution of urban ecological space, yielding a simulation accuracy of 83%.

Data and methods

Study area

According to the description in the Development Planning Outline of the Guangdong–Hong Kong–Macao GBA, the GBA urban agglomeration (21°50'–24°39' N, 112°02'–115°42' E) comprises nine cities of the former Pearl River Delta urban agglomeration, namely, Guangzhou, Shenzhen, Foshan, Dongguan, Huizhou, Zhongshan, Zhuhai, Zhaoqing, and Jiangmen, as well as the two special administrative regions of Hong Kong and Macao (Fig. 1). The overall urbanization level of the study area is relatively high, with an area of 5.6 × 105  km2. The study area lies south of the Nanling Mountains and adjoins the South China Sea. The terrain is predominantly occupied by hills and plains, with the central part forming the Pearl River Delta plain, and the elevation is typically less than 200 m. The urban agglomeration represents an area of concentrated population and construction land and has exhibited some of the strongest economic vitality of all regions in China over the past 30 years, with rapid economic development, rapid population growth, and extensive urban expansion. By the end of 2021, the permanent population in the GBA was 78.606 million, and the regional GDP was 12.6 trillion yuan. The rapid expansion of cities in the GBA has resulted in an increasingly prominent imbalance of economic structure, destruction of resources and the environment, and other issues, which have seriously affected the ecological environment and landscape pattern of the Pearl River Basin.

Fig. 1: The study area.
figure 1

In this study, the Guangdong–Hong Kong–Macao Greater Bay Area was selected as the study area. The GBA urban agglomeration (21°50'–24°39' N, 112°02'–115°42' E) comprises nine cities of the former Pearl River Delta urban agglomeration, as well as the two special administrative regions of Hong Kong and Macao.

Full size image

Data sources

The data used in this study included land-use data, basic geographic information data, digital elevation models (DEMs), and socioeconomic data of the GBA urban agglomeration (Table 1). Land-use data were obtained from the Resource and Environmental Science Data Center of the Chinese Academy of Sciences (http://www.resdc.cn), with a spatial resolution of 100 m × 100 m. The first-level classification includes forest land, cultivated land, grassland, water areas, construction land, and unused land. The overall classification accuracy is high, at more than 90%. Data were obtained for 1990, 1995, 2000, 2005, 2010, 2015, and 2020. Basic geographic information data were obtained from the National Basic Geographic Information Center (http://www.ngcc.cn/) and Open Street Map databases and included cities, administrative boundaries of cities, traffic line data, and river data. The DEM data consisted of 1-km elevation data from the NASA Shuttle Radar Topography Mission collection. The economic data included gridded population data with a spatial resolution of 100 m and GDP data with a 1-km grid, which were obtained from the Resource and Environmental Science Data Center of the Chinese Academy of Sciences (http://www.resdc.cn).

Table 1 Data list.
Full size table

Research methods

Landscape pattern index

Urban expansion directly changes the landscape pattern (Liao and Zhang (2021)). Scientific analysis of landscape patterns forms the basis of spatial change analysis and the evaluation of land-use changes. The landscape pattern index is typically evaluated on three levels: the patch level, patch-type level, and landscape level. Here, we measured and depicted the landscape pattern of the GBA urban agglomeration from the perspective of patch-type and landscape levels. Based on the results of previous research, the number of patches, largest patch index (LPI), patch aggregation index, landscape connectivity index, and other landscape indices with clear ecological importance were selected at the patch-type and landscape levels for further analysis (Liao and Zhang (2022)). We then quantitatively analyzed the land-use change status of the GBA urban agglomeration from the aspects of landscape fragmentation, aggregation, and diversity. Table 2 shows the calculation formulas and definitions of the key landscape indices.

Table 2 Calculation formula and meaning of main landscape indices.
Full size table

ANN–CA–Markov model

The ANN model is a complex, nonlinear, dynamic learning network system formed by simulating the basic features of human brain function through a large number of simple, interconnected neurons with strong parallel, distributed storage, processing, self-organization, adaptation, and self-learning capabilities. This backpropagation ANN mainly consists of a three-layer structure, including an input layer, a hidden layer, and an output layer. It was designed to obtain the minimum error network parameter values through continuous training of samples and then output the data with the smallest error by inputting data similar to the training samples. The CA model is a dynamic system that simulates evolution in space and time based on cellular space and specific rules. The CA model consists of four basic parts, cell, cell space, neighborhood, and rules, as well as two additional parts, state and discrete time. The Markov model is a stochastic mathematical method used to predict the prior probability and conditional probability of events. The Markov model consists of four parts: state, state-transition process, state-transition probability, and state-transition matrix.

The combined ANN–CA–Markov model simulates land-use changes and future land-use scenarios under natural and human influences (Liu et al., 2017). This model is based on the traditional CA model and ANN algorithm, and when combined with the Markov chain model, the resulting ANN–CA–Markov model calculates the future quantity demand of different land uses according to the requirements of different scenarios. By combining historical land-use data and multiple driving factors, including human activities and natural elements such as temperature, precipitation, soil, terrain, transportation, location, and policies, we obtained the development suitability probability of various land types and then calculated the overall conversion probability of each cell (defined below). Finally, future land-use changes were simulated through the roulette competition mechanism.

The components of the CA model (cell, cell space, state, neighborhood, rule, and cycle number) form the underlying structure of the combined ANN–CA–Markov model. In addition, the Markov and ANN models are used to calculate the most important part of the rule (defined below). The specific construction is as follows:

(1) Definition of cell: the size of the ecological spatial structure type map is a 30 m × 30 m grid.

(2) Definition of cellular space: all grids of the ecological space structure type map.

(3) Definition of state: the state of a cell is an attribute of the grid, which is divided into cultivated land, forest land, grassland, water area, unused land, and construction land.

(4) Definition of the neighborhood: a 5 × 5 grid of connected cells; each cell has 24 cells around it that have an impact on its properties.

(5) Definition of rules: the transformation rules of cells mainly include the transformation rules for space and quantity. The transformation rules for quantity are mainly calculated using the Markov model, and the transformation rules for space are calculated using the ANN model.

(6) Determination of the number of cycles: as the interval between the basic data of each period is 5a, the age interval for determining the number of cycles is set to 5a.

Scenario setting

Future land-use simulations must consider different constraints in order to control the future development trend and expansion direction of the study area in the model prediction. Thus, based on the land area predicted by the Markov chain model, combined with the relevant local planning documents as constraints, we implemented three models to simulate the impact of future land-use demands and policy controls on the spatial evolution of future land-use changes in the GBA urban agglomeration: a natural development scenario, an ecological protection scenario, and an economic development scenario (Deng et al., 2016). The Chinese government issued the Outline of the Development Plan for the GBA in 2019 to strengthen the protection of mountains, hills, and forest ecosystems around the Pearl River Delta and to build the forest ecological barrier in the northern mountain ranges. These factors were demarcated, restricted, and input into the ANN–CA–Markov model as constraints. Under the natural development scenario, land-use change was not affected by human plans and policies and only evolved according to the characteristics of land-use transformation under historical conditions. We assumed that the historical growth model would continue to exist in the coming decades; therefore, no restrictions were set in the simulation experiment. The demand for various land-use types in 2025 and 2035 was calculated using the Markov chain model. For example, the construction land area was predicted to increase to 8247.8 km2 by 2025 and 8473.92 km2 by 2035. The ecological protection scenario emphasized the co-construction and co-protection of the ecological environment. Incorporating the land-use development of the GBA urban agglomeration, the natural development scenario predicted the area and spatial distribution of each landscape type in 2030 based on the current data of landscape types in 2020. Therefore, based on the land-use status in 2020 and the future development plan, the urban development scenario predicted that the transfer probability of cultivated land and forest land to artificial land surface will be reduced by 30%. Furthermore, under this scenario, the probability of conversion from shrublands and water bodies to artificial surfaces was reduced by 20%, while the probability of conversion from artificial surfaces to forest land was increased by 10%. Moreover, habitat degradation was strictly prohibited, and various reservoirs and lakes in the region were used as constraints to restrict their arbitrary conversion (Yin et al., 2023). The economic development scenario emphasizes the economic development goals of the United Nations Sustainable Development Goals (SDGs), without restrictions on urban expansion. Based on the land-use stochastic matrix from 2010 to 2015, the transfer probability from cultivated land to construction land increased by 60%, the probability of grassland transferring to construction land increased by 50%, the probability of transferring unused land to construction land increased by 30%, and the probability of mutual transfer between other land types remained unchanged.

Markov chain theoretical basis

The traditional Markov chain is extracted from the theory of stochastic processes by Russian mathematician A.A. Markov (Fingleton, 1997). The theory measures events and their development trends by constructing a state transfer probability matrix. Given the current knowledge or information, the past (the historical state) is irrelevant for predicting the future—“no posteriority,” also known as “Markovianity.” The Markov probability transfer matrix can be used to find the smooth distribution of this stochastic process, which can predict the dynamic evolution trend of an economic phenomenon (urban agglomeration expansion in this study). For instance, in the traditional Markov chain {Xn, n = 0, 1, 2, ⋯}, state space, pij is the one-step transfer probability, and {πi, i∈S} is the probability distribution of each state. If {πi, i∈S} satisfies the equation system \(\pi _i = \mathop {\sum}\nolimits_{j = 0}^\infty {\pi _jp_{ij}}\)(i = 0, 1, 2, ⋯), the probability distribution {πi, i∈S} of the traditional Markov chain is the smooth distribution. Extending it to spatial Markov chains, the smooth spatial distribution under each spatial lag type is obtained according to the similarity principle. The maximum value of the transfer probability is considered the possible evolution trend of the corresponding state.

Results

Land-use characteristics of the GBA urban agglomeration

The proportion of various land-use types in the GBA urban agglomeration from 1990 to 2020 reflects the structural characteristics of land resources in the study area (Fig. 2). From 1990 to 2020, the dominant land-use type was consistently forest land, accounting for more than 50% of the total land area, followed by cultivated land and construction land, with average proportions of 30.56% and 7.92%, respectively. The water area accounted for 6.19% of the total area, and shrubs, unused land, and grassland accounted for 0.02%, 0.18%, and 0.02%, respectively. With the continuous development of the GBA urban agglomeration, the construction land area increased significantly, from 1573.78 km2 in 1990 (2.16% of the total land area) to 6865.93 km2 in 2020 (12.55% of the total land area), representing a growth rate of 336.27%. However, the cultivated land area decreased every year, with a total reduction of 11,363.55 km2 from 1990 to 2020, which was more than twice the growth area of construction land. The trend of construction land changes was opposite to that of cultivated land changes. From 1990 to 2020, the construction land area changed according to the following pattern: sharp increase (growth rate of 121.57%), steady increase (61.61%), and slow increase. In contrast, the pattern of cultivated land area change was as follows: rapid decrease, steady decrease, decrease, and slow decrease. The trends of unused land and cultivated land change were generally consistent over the study period and were characterized by a gradual reduction in land area. Moreover, the conversion rate of unused land and cultivated land to grassland increased every year, rising sharply from 2015 to 2020.

Fig. 2: Changes in the area of land-use transfer from 1990 to 2020.
figure 2

The proportion of various land-use types in the GBA urban agglomeration from 1990 to 2020 reflects the structural characteristics of land resources in the study area.

Full size image

Spatial and temporal evolution pattern of urban agglomeration expansion

We analyzed the spatial development pattern and spatiotemporal evolution characteristics of the GBA urban agglomeration using the standard deviation ellipse combined with the landscape pattern and expansion area. From 2010 to 2020, the GBA urban agglomeration exhibited significant urban land expansion, with a growth rate of 192.32%. From 1990 to 2020, the standard deviation ellipse of construction land and the spatiotemporal change trend of the urban agglomeration gravity center showed a migration trend in the southeast (Fig. 3). The center of gravity moved slowly from 1990 to 1995 and 2005 to 2015 but moved rapidly toward Guangzhou from 2000 to 2005 and 2015 to 2020. This changing trend was closely related to the policies issued in 2009 by the Outline of the Reform and Development Plan of the Pearl River Delta Region and in 2019 by the Outline of the Development Plan of the GBA. The implementation of these policies greatly promoted the development of urban agglomerations. Therefore, the movement trend of the center of gravity of construction land was relatively stable between 2010 and 2020. The oblateness direction angle of the standard deviation ellipse of the GBA urban agglomeration was consistently between 90° and 105° and was stable at ~96.3°. Notably, urban expansion consistently occurred in the northwest–southeast direction, indicating that Guangzhou and Shenzhen formed the core of the urban agglomeration, with Foshan, Dongguan, and Zhongshan as the three growth poles. The development axis of the GBA is characterized by a vertical direction and rapid development, which promote urbanization and ecological civilization. Guangzhou played the role of a central regional city, and Shenzhen represented the national comprehensive pilot project of new urbanization. Construction land was gradually distributed from the northwest to the southeast in the study area.

Fig. 3: The standard deviation ellipse of construction land and the spatiotemporal trends in the center of gravity of the GBA Urban Agglomeration.
figure 3

From 1990 to 2020, the standard deviation ellipse of construction land and the spatiotemporal change trend of the urban agglomeration gravity center showed a migration trend in the southeast.

Full size image

Scientific analysis of urban landscape patterns forms the basis of spatial change analysis and the evaluation of land-use changes (Table 3). The number of patches for water areas, construction land, and unused land was lowest in 2000 and generally increased from 2000 to 2020. Conversely, the number of patches of cultivated land, forest land, and grassland fluctuated and declined from 2000 to 2020, indicating that construction land gradually replaced cultivated land, forest land, and grassland. The LPI value was highest for forest land, followed by cultivated land and water areas, indicating that these were the main land-use types in the GBA urban agglomeration. In addition, the LPIs of different land-use types fluctuated to varying degrees from 1990 to 2020; the LPI of cultivated land decreased significantly, the LPI of construction land increased significantly, the LPI of grassland, water areas, and unused land decreased slightly, and the LPI of forest land remained essentially unchanged, with slight fluctuations. This indicates that the rapid expansion of construction land primarily encroached on cultivated land, followed by grassland, water areas, and unused land. Before 2015, the patch aggregation index of all land-use types except construction land showed a downward trend, indicating that the continuous expansion and aggregation of construction land reduced the aggregation of other types of land and that ecological land was eroded and the patches were more fragmented. After 2015, the patch aggregation index of construction land increased slightly, reflecting sprawling urban expansion, whereas that of the other five types of land began to either show an upward trend or decrease more slowly. The upward trend was most notable for grassland, cultivated land, and forest land, indicating that policy implementation effectively controlled the destruction of ecological land. The landscape connectivity index of construction land increased every year, indicating that the continuous expansion of construction land occurred in contiguous areas, whereas that of other ecological land and unused land fluctuated and decreased, revealing a more fragmented distribution.

Table 3 Landscape index from 1990 to 2020.
Full size table

Land-use simulation and future scenario prediction

The land-use simulation results for the GBA urban agglomeration for 2015 exhibited suitable accuracy and reliability. The ANN–CA–Markov model simulations of urban expansion in the GBA urban agglomeration according to land-use changes had a Kappa coefficient of 0.72 and an overall accuracy of 83%. These metrics indicate that the selected driving factors have a strong ability to explain land use in the study area and can reliably predict the distribution of future land use. Compared with the actual land-use pattern in 2015, the simulated land-use structure for various land-use types was relatively consistent; differences between actual and simulated results mainly occurred in local areas (Fig. 4).

Fig. 4: Comparison of actual and simulated results of land use in 2015.
figure 4

Compared with the actual land-use pattern in 2015, the simulated land-use structure for various land-use types was relatively consistent; differences between actual and simulated results mainly occurred in local areas.

Full size image

To determine the impact of future land-use demands and policy controls on the spatial evolution pattern of urban expansion under different scenarios, we set up three scenarios to simulate land-use changes in the GBA urban agglomeration: natural development, ecological protection, and economic development (Fig. 5).

Fig. 5: Results of multi-scenario land-use simulations for 2025 and 2035.
figure 5

R To determine the impact of future land-use demands and policy controls on the spatial evolution pattern of urban expansion under different scenarios, this study set up three scenarios to simulate land-use changes in the GBA urban agglomeration: natural development, ecological protection, and economic development.

Full size image

Under the natural development scenario, the construction land area was predicted to increase by 8247.8 km2 in 2025 and 8473.92 km2 in 2035, with a large amount of unused land and cultivated land being replaced as the dominant land-use type, mainly by grassland and construction land.

Under the ecological protection scenario, construction land expansion was generally restricted and expansion of the GBA urban agglomeration was considerably reduced compared with that under the natural development scenario. By 2025, the construction land area was predicted to increase by 7247.8 km2, accounting for 8.31% of the total area, and relatively little unused land was predicted to remain. Cultivated land is classified as ecological land, and the simulated encroachment of urban construction land on ecological land was limited. The stability of cultivated land, forest land, and grassland was somewhat enhanced, indicating less conversion of forest land under the ecological protection scenario than under the natural development scenario; these land-use types were spatially concentrated with a contiguous distribution.

Under the economic development scenario, the amount of construction land transformation was predicted to increase substantially in 2025 by 2103.43 km2, accounting for 10.31% of the total area. Thus, restrictions on construction land increased the stability of cultivated land. The transformation between other types of land use did not include basic farmland, and the cultivated land area decreased slightly by 144.56 km2. Unused land occupied a larger area under the economic development scenario than under the natural development scenario, but its area was reduced by 20.59%. Forest land exhibited the second-largest reduction, decreasing by 6.12%, whereas grassland was somewhat protected; thus, its area increased by 3.65%.

Discussion

The ANN–CA–Markov model is advantageous for simulating future urban expansion because it comprehensively considers the impacts of climate change and human activities, exhibits high-simulation accuracy, and generates a realistic land-use pattern. In this study, the ANN–CA–Markov model, based on urban flow, was proposed to simulate the urban expansion of the GBA urban agglomeration from 1990 to 2020. This model combines the traditional logical CA model with the gravity model of urban agglomeration, regards the intensity of urban flow as the driving force that affects the transformation of non-urban cells into urban cells, and applies logical regression to calculate the coefficients of each driving force. Additionally, this study implemented the traditional CA model as a comparison. The Kappa coefficient and urban expansion proportion index were used to evaluate the simulation results. The experiment showed that the overall simulation accuracy of the model proposed in this study was higher when simulating the growth of urban agglomeration. In the expansion simulation of urban agglomeration that did not consider the spatial interaction within the urban agglomeration, the urban growth of each city was relatively independent. Furthermore, the growth of central cities showed a fluctuating and irregular distribution, and the growth of marginal cities showed an over-growth and centralization trend.

Due to the continuous development of the transportation network, the development of the urban agglomeration in the GBA of Guangdong, Hong Kong, and Macao has been adopted as an overall plan. The development of all parts should be coordinated, prioritizing the development of the central city and driving the common development of the surrounding areas along with the center. Therefore, the urban growth patch should first appear at the junction of Guangzhou or Shenzhen and nearby cities. This change trend is closely related to the policies issued in 2009 by the Outline of the Reform and Development Plan of the Pearl River Delta Region and in 2019 by the Outline of the Development Plan of the GBA. The plan outline proposes to optimize the spatial layout of the Pearl River Delta in accordance with the positioning of the main functional areas. With Guangzhou and Shenzhen as the center and the east and west banks of the Pearl River Estuary as the focus, it will promote the regional economic integration of the Pearl River Delta and drive the more rapid development of this region, forming a new pattern of coordinated development that optimizes the allocation of resource elements and highlights regional advantages. Following this plan, Guangzhou would develop into the leading area in Guangdong in terms of livability and become an international metropolis that would serve as a model for the whole country and worldwide. Meanwhile, it is recommended that Shenzhen continue to serve as an experimental field and demonstration area of the special economic zone, strengthen the functions of scientific and technological research and development and high-end services, strengthen the status of the national economical center city and the innovative national city, and build an international demonstration city of socialism with Chinese.

However, the spatial expansion of the GBA urban agglomeration was restricted and affected by multiple factors, such as topography, the ecological environment, and the social economy. The speed and mode of urban spatial expansion were particularly restricted by local and regional policies. Due to the data sources and data quality in this study, we only considered the impact of key influencing factors, such as topography, transportation networks, GDP, and population; the constraints of natural conditions, such as climate, precipitation, and soil were not considered. Therefore, the selection of influencing factors represents a limitation of this study. In addition, our research only analyzed urban expansion in the horizontal direction. However, as the concept of smart urban growth gradually penetrates urban planning practices, urban development will become more intensive, compact, three-dimensional, environmentally focused, efficient, and sustainable. Thus, urban expansion should also be considered in the vertical direction through the height of urban buildings. As such, future research should analyze three-dimensional changes in land-use behavior by including the type of buildings (high-rise or low-rise) and evaluate three-dimensional development in combination with the ecological environment.

In addition, coordinating the relationship between urbanization and the ecological environment is essential for promoting future urbanization of the GBA urban agglomeration. In future research, we plan to set development scenarios and corresponding parameters according to the SDGs, further optimize the setting of conversion rules, and optimize urban expansion from the perspective of interaction and collaborative development of the social–economic–ecological composite system. The development and utilization of basic ecological land types, such as natural forests, natural grasslands, and wetlands, must be strictly controlled. Furthermore, the development of swamps, tidal flats, and other land types must be strictly planned in accordance with policies to protect and improve their ecological functions. This research has important implications for promoting reasonable urban planning to effectively optimize the urban spatial structure. Moreover, our simulations of future urban expansion in the GBA provide important guidance for spatial planning decision-making, sustainable urban development, and ecological balance maintenance.

The expansion of urban agglomeration over a long time series is a complex systematic process. This study only examined the development of urban agglomeration in the GBA from the perspective of land-use expansion, which has certain limitations. In addition, the statistical caliber of the socioeconomic data for Hong Kong and Macao is inconsistent with that of data for the mainland. Furthermore, individual deficiencies exist in the socioeconomic statistical indicators of each city in a long time series, such as that used in analyzing the main factors driving the expansion of the GBA urban agglomeration. Therefore, this study of the driving mechanism of land-use expansion in the GBA is still relatively unsophisticated and warrants further improvement.

Conclusions

In this study, we used land-use coverage data from 1990 to 2020 to analyze the spatiotemporal evolution of regional expansion in the GBA urban agglomeration and employed the ANN–CA–Markov model to simulate future urban expansion in the study area under multiple scenarios. The major findings are as follows:

(1) From 1990 to 2020, the construction land area increased, whereas that of cultivated land decreased. The proportion of construction land in the study area increased from 2.16% (1573.78 km2) in 1990 to 12.55% (6865.93 km2) in 2020, indicating a total growth rate of 336.27%, with the most rapid growth occurring from 1990 to 2000 (121.57%) and more steady growth from 2000 to 2010 (61.61%).

(2) The Kappa coefficient and overall accuracy of the ANN–CA–Markov model (0.72 and 83%, respectively) indicate high-simulation accuracy and suitability for predicting the land-use dynamics of the study area. During the study period, urban expansion mainly occurred in Guangzhou, Shenzhen, Foshan, and Dongguan via extensional expansion, typically involving the encroachment of ecological land, such as cultivated land and water areas. Newly added grassland occupied a portion of the cultivated land. The expansion of cultivated land mainly encroached on forest land near cities. In the short term, the replacement of cultivated land with construction land represented the main source of future urban expansion, followed by the occupation of unused land, which was predicted to increase in the future.

(3) The three future scenarios (natural development, ecological protection, and economic development) showed marked differences in the spatial structure and characteristics of future land-use changes. Under the natural development scenario, urban expansion was not constrained, and rapid growth led to the transformation of a large amount of ecological land. After adding ecological constraints (ecological protection scenario), the transformation of forest land and cultivated land was effectively controlled, which was beneficial for the protection of ecological space and the restoration and improvement of natural ecological functions. Under the economic development scenario, urban expansion was predicted to occupy more unused and cultivated land.

(4) The expansion process of the land use of the cities within the GBA urban agglomeration was essentially consistent with their respective socioeconomic development levels. However, the expansion differences of the cities also reflected the reality that the GBA urban agglomeration has not yet fully achieved coordinated development within the region and that large cities have more development space and opportunities than small cities. In the guidance of future urban agglomeration development, the coordination between cities should be further improved, the division of labor and cooperation within the process of urban agglomeration clarified, and the degree of coordinated development within urban agglomerations increased.

(5) The expansion pattern of land use in the GBA urban agglomeration has gradually shifted from diffusion to agglomeration, fully demonstrating that the urban expansion theory is applicable in guiding the development research of the GBA urban agglomeration. The functional orientation and development planning of cities within the urban agglomeration are expected to affect the formation of the expansion pattern of urban agglomeration land use. The expansion pattern of land use in the GBA was affected by the main traffic trunk lines. The relatively complete traffic facilities system has led to the formation of expansion clusters such as “Guangfo,” “Guangzhou–Shenzhen–Hong Kong,” “Macao–Pearl River,” and “Shenguan–Hui,” which made the expansion of land use in the GBA more concentrated.