Integrated eco-economic zoning and carbon neutrality zoning into the PLUS model to simulate land use change in the Guangdong–Hong Kong–Macao Greater Bay Area

Study area

The GBA (111°20′–115°24′E, 21°32′–24°26′N) lies in southern China with an area of nearly 56,000 km² (Fig. 2). Its terrain is high in the northwest and low in the southeast with altitudes of -165–1,559 m. The central part is plain, its surrounding is mountains and hills. The whole region is dominated by tropical and subtropical monsoon climate, with mean annual temperature of 21.8 °C and annual rainfall of 1,789.3 mm. The GBA comprises of nine inland cities and two special administrative regions. According to ‘Notice of the State Council on Adjusting the Standards for Urban Scale Classification (2014)’, cities with a permanent population of over 10 million in urban areas are considered as megacities; Cities with a permanent population of five million to 10 million in urban areas are considered as supercities; Cities with a permanent population of one million to five million in urban areas are considered as large cities. Therefore, the GBA contains eleven cities, including three megacities with a population of more than 10 million, namely Guangzhou, Shenzhen and Dongguan; three supercities with a population of five to 10 million, namely Hong Kong, Foshan and Huizhou; and five large cities with less than five million, namely Macao, Jiangmen, Zhongshan, Zhaoqing and Zhuhai.

With population growth and rising consumption levels, land use has changed significantly, accompanied by substantial farmland loss and urban sprawl (Zhang et al., 2022). As a result, the ecological risk is constantly increasing over time (Lu, Laffan & Pettit, 2022). In terms of carbon emissions, the amount increased by 184 Mt from 2000 to 2016, with a growth rate of 7% over 2000–2011 (Zhou et al., 2018).

Data sources

Land use data for 2010, 2015, and 2020 (spatial resolution: 30 m), which adopt a secondary classification system, were obtained from the Chinese Academy of Sciences (http://www.resdc.cn/) (Table 1) (Liu et al., 2010). We applied first class type, namely farmland, wetland, grassland, forestland, and construction land. The population density, GDP, elevation, and roads were collected from the Resources and Environmental Scientific Data Center (RESDC) (http://www.resdc.cn/). High-speed railways were obtained from the Baidu Map (https://map.baidu.com). Points of interest (POIs) were derived through network crawling, including restaurants, bus stops, metro stations, manufacturers, schools, hospitals, and residences. Validating by field points, the accuracy of POIs reached at 90%. County-level CO₂ emissions and sequestration data were downloaded from previous studies (https://www.ceads.net.cn/) (Chen et al., 2020). Those data were resampled to 500 m and unified coordinates to the Albers coordinate system.

Comprehensive partitions

Previous studies have verified the feasibility of the comprehensive partitions in other regions (Sun et al., 2020; Xu et al., 2022). Two comprehensive partitions were adopted, one partition was eco-economic zoning, and another was carbon neutrality zoning. In the view of governance philosophy, the core concept of two comprehensive partitions is to achieve harmonious coexistence between humans and nature, the former focuses on the coordinated development of ecological and economic systems, while the latter adjusts land use structure from the perspective of carbon neutrality. In the perspective of governance measures, the former focuses on both ecosystem protection and restoration as well as economic development, while the latter emphasizes emission reduction in high carbon emission areas. The indicators for partitioning were closely related to land use types, indicating that the comprehensive partitions have potential values, and another was carbon neutrality zoning

PLUS model

Compared with existing CA-based LUMs, the PLUS model has a higher accuracy through further understanding the relationships between land use types and driving factors by random forest classification algorithm, and could simulate land use change at patch level (Liang et al., 2021). The formulas could be seen in Appendix SB.

Driving factors

Land use change is closely linked to population, economic, transportation, and natural factors. The sustained growth of urban population and economic development are key force driving the expansion of urban construction land. Transportation and natural factors define urban development trajectories and patterns. To eliminate multicollinearity among factors, we conducted a variance inflation factor (VIF) test. If the VIF value of two factors exceeded 5, we removed one of them. Eventually, a total of 12 driving factors were selected (Zhuang et al., 2022), including socioeconomic factors (GDP, population density, POIs density, distance to transportation line, distance to urban area, distance to villages), topographic factors (elevation, slope), and fragmentation factor (patch density of urban construction land) (Fig. 3). The PLUS model employs the random forest algorithm to individually analyze the driving factors of various types of land use expansion. It obtains the development probability for each land use type, as well as the contribution of driving factors to the expansion of each land use category during the period.

Land use dynamics during 2010–2020

Farmland and forestland were the dominant land use types during the study period, covering 23.1% and 54.4% of the entire region in 2010, respectively (Fig. 4; Table 6). However, by 2020, farmland and forestland had experienced a continuous decline of 4.7% and 1.3% in area, respectively. Construction land experienced a substantial expansion, with a increase of 11.4% (815.1 km²) during 2010–2015 and 5.2% (374.0 km²) during 2015–2020. The most widespread land use change in the GBA was the conversion of farmland to construction land, followed by the conversion of forestland to construction land during 2010–2020 (Figs. 4, 5).

Partition results

The ecological risk level presented a high trend in the southwest but a low trend in the east bank of Pearl River (Fig. 6). It suggested that land in the southwest was less disturbed. Owing to urbanization, the region in the east bank of Pearl River has suffered severe human disturbances (He et al., 2021). Therefore, the landscape tended to be fragmented, which resulted in the ecological risk index values increasing. The economic development in the central regions was higher than that in the surrounding regions, and the value of GDP of the east bank of Pearl River was higher than that in the west bank of the Pearl River. In terms of the carbon neutrality, the value of the central regions was higher than that in the surrounding regions.

The CDZ was concentrated in Foshan, Zhongshan, Dongguan, and Guangzhou rural areas (Fig. 6). The partition ERZ was mainly situated in Guangzhou core areas, Shenzhen, and Hongkong. The EPZ was concentrated in Jiangmen, Zhaoqing, and Zhuhai urban areas. The DPZ was scattered and mainly distributed in the surrounding regions. The partitioned LCS accounted for most of the area and was mainly distributed in the surrounding area. The partition MCS and partition HCS were concentrated in the central region and separated by the Pearl River (Fig. 6).

Simulation results analysis

The results of entire simulation in 2020, partition A simulation in 2020, and partition B simulation were shown in Fig. 7. The whole-region simulation results showed that the OA was 0.7720 and the FoM was 0.0829. The OA was 0.8454 and the FoM was 0.1146 under eco-economic zoning (Table 7). The OA was 0.7925, and the FoM was 0.1038 under carbon neutrality zoning. This indicated that the accuracy of the PLUS model under the two scenarios improved significantly, with increases of 38.23% for eco-economic zoning and 25.21% for carbon neutrality zoning, compared to the whole-region simulation.

The accuracy of the PLUS model varied with sub partition. The comparative change in OA of each partition was increased. FoM ranged from 0.0728 to 0.1947 under eco-economic zoning. FoM varied from 0.0768 to 0.1710 under carbon neutrality zoning (Table 7). Figure 7 showed the detailed simulated land use patterns of 2020 under different simulation strategies. The whole or partitioned simulation patterns all showed a high spatial consistency with the actual land use pattern.

Partitions applicability and its effects on land use simulation

The GBA lies in the Pearl River Delta, where is the most advanced agricultural region in China. Large-scale urbanization was inevitable occupied because its lower development costs. With the implementation of farmland requisition-compensation balance and farmland red line, farmland is more effectively protected. Simultaneously, forestland on gentle slopes was converted for urban development. Land use changed induced by urbanization alter regional ecological processes, eventually increased in ecological risk, particularly well-developed areas. Industrial urbanization has not only led to a surge in carbon emissions but also caused a decline in carbon sinks. Ultimately, this goal of carbon neutrality became more difficult achieve.

As the urban expansion slowed down, natural land has been protected to a certain extent (Zhang et al., 2022). Previous researches have studied the ecological security patterns across the GBA by identifying the key ecosystem services (Bi, Xie & Yao, 2020; Gao et al., 2022; Jiang et al., 2021; Wang et al., 2021; Zhang et al., 2023), and our results were consistent with these findings. The economic data have been validated by statistical data, with a high accuracy. Therefore, eco-economic zoning was reliable. In terms of the pattern of carbon neutrality, the results were similar to those from other studies (Li et al., 2023; Lin & Li, 2020; Luo et al., 2021). It suggests that the foundation data for partition based on carbon neutrality was applicable. Ecological risk, economic development and carbon neutrality are strongly related to land use, which guarantees that space zoning based on these indicators can be feasible. Additionally, the two partitions are realistic and have a practical significance.

Theoretically, reasonable space partitioning can improve the performance of land use models. The land use conversion rules of the GBA vary between the whole region and the sub regions, both with a strong spatial heterogeneity. Under partition A, the main reason was that the land use dynamics still varied greatly in sub partitions. Additionally, partitioning increased the complexity and uncertainty of the model. Therefore, eco-economic zoning need to be further studied in other regions. Under carbon neutrality zoning, the ecological resources were mainly distributed in the LCS partition with fewer human activities; thus, the land use changed slowly. The differences of industrial structures in MCS and HCS cause the changes of land use in different ways. For example, the leading industry of Shenzhen is the service industry, which commonly requires less land. In contrast, the leading industry of Foshan is manufacturing, which requires more land use. This indicates that carbon emissions are closely related to land use types, and carbon zoning is more suitable for the partitioned simulation of land use change. In our study, we considered the spatial heterogeneity in view of ecological risk, economic development, and carbon neutrality. Thus, both the accuracy of eco-economic zoning simulation and carbon neutrality zoning simulation improved significantly.

Zoning method

Spatial heterogeneity exists in land use simulation. The dominant driving factors and their strength vary in different sub partitions. Partitioning the region can be based on land use attributes, administration, and main function (Yin et al., 2018). At present, partitioning the region faces some uncertainties, such as how to determine the thresholds and the number of sub partitions (Xia et al., 2019). A series of classification methods can be used to generate sub partitions, including k-means cluster algorithms, dual-constrained spatial clustering, and self-organizing map (Ke, Qi & Zeng, 2016; Qian et al., 2020; Xia & Zhang, 2021; Xu et al., 2021). But the k-means cluster algorithm has a strong subjectivity, and the number of partitions needs to be determined in advance. Additionally, it treats each pixel as discrete, which neglects spatial autocorrelation. In contrast, self-organizing maps preserve the properties of the whole region well when considering the constraint of neighborhood distance. Therefore, more methods need to test, such as Geodetector, Geo-self-organizing map, and decision trees. Transition rules are also important for the accuracy of the simulation. Previous studies have adopted logistic regression, artificial neural network, and random forest to capture the transition rules (Liao et al., 2016). More methods need to be applied and compared, such as deep learning methods (He et al., 2018).

Limitations and future research

Although two partitioning strategies have been applied to land use simulation, there are still some limitations that need to be addressed. Firstly, driving factors are stable in the PLUS model, which decreases the reliability of the long term simulation based on the model (Feng et al., 2019). Additionally, the main driving factors vary in different zones. However, we used the same driving factors in different sub-partitions. In future, the main driving factors should be screened out for model simulation. Secondly, the land use data had a coarse spatial resolution, which possibly misses the information at finer resolution. This limits the generalizability of the conclusion. Thirdly, there is a need to put more efforts to investigate the meaning of eco-economic zoning, such as the threshold method and number of the sub-region. Nevertheless, we proposed two zoning strategies separately based on ecological risk and carbon neutrality at urban agglomerations levels and integrated them into partitioned land use simulation, which has a great significance for “current-future” ecosystem management to achieve the balance of economic development and ecological protection. The practical applications contain territorial spatial planning, environmental zoning control, regional coordinated development strategy.