Scenario Simulation and Spatial Management Implications of Water Ecosystem Services in the Guangdong-Hong Kong-Macao Greater Bay Area (2035)

Abstract

Rapid urbanization threatens water ecosystem services (WESs) in China’s Greater Bay Area. This study employs a Markov-FLUS land-use simulation coupled with the InVEST model to project land-use patterns for 2035 under four scenarios—Natural Development (ND), Farmland Protection (FP), Economic Priority (EP), and Ecological Protection (EcoP)—and evaluates their impacts on water yield, soil retention, and total phosphorus (TP) export. Under ND and FP scenarios, modest gains in water yield (+32.25% and +32.13%) and soil retention (+46.16% and +45.91%) are achieved, but TP control remains limited (−0.05% and +4.82%). In contrast, the EP scenario drives severe declines in water yield (−13.39%) and soil retention (−2.11%) alongside a TP surge (+5.87%), evidencing ecological degradation under high-intensity development. Conversely, the EcoP scenario yields substantial improvements, water yield +50.67%, soil retention +70.94%, and TP export −8.17%, reflecting the synergistic “multiplier effect” of combined woodland and water-body restoration. Spatially, urban cores and agricultural margins exhibit divergent service responses, underscoring the need for differentiated management. We developed a spatial priority map by integrating the predicted WES changes under the Ecological Protection scenario with indicators of urban proximity and pollution risk. This map identifies critical intervention zones. We propose targeted spatial optimization—strict protection of sensitive ecological zones, green transformation in urban expansion areas, and diffuse pollution controls in agricultural peripheries—to reconcile development with ecosystem resilience.

1. Introduction

Water ecosystem services (WESs) have emerged as critical indicators for assessing the degradation of aquatic ecosystems, reflecting a growing recognition of their ecological and societal importance. WES refers to the diverse benefits and functions provided by aquatic ecosystems, including water supply, purification, flood regulation, and the maintenance of biodiversity. The development of WES research epitomizes the broader evolution in environmental science toward a deeper recognition of nature’s intrinsic and instrumental values. Initially, studies on water resources focused predominantly on their engineering and physical attributes—such as water allocation, hydropower generation, and flood control. However, over time, it became increasingly evident that water is not merely a physical commodity but also a provider of critical ecological services.

By the late 20th and early 21st centuries, the emergence and widespread adoption of the “ecosystem services” framework catalyzed a significant shift in water-related research. Scholars began to systematically evaluate and quantify WESs to better understand their contributions to human well-being. A landmark study by Costanza [1] quantified the global value of ecosystem services—including those provided by aquatic systems—offering a novel perspective on the economic importance of natural capital. In the 21st century, the intensifying impacts of climate change and anthropogenic disturbances have redirected the focus of WES research toward sustainability, ecosystem resilience, and the interdependence among various ecosystem services. Remote sensing and Geographic Information System (GIS) have become vital tools for monitoring ecological changes and assessing their implications for service provision at both local and global scales.

Recent studies have gone beyond the assessment of service provision to explore the linkages between WES and overall ecosystem health. For instance, researchers have examined the relationship between the loss of WES and declining water quality, as well as the impact of biodiversity reduction on water provisioning and regulatory functions [2,3]. A growing body of literature has focused on the valuation of WES by integrating ecological and economic approaches to quantify their benefits. Such evaluations enhance our understanding of how aquatic ecosystem degradation can adversely affect human well-being and socioeconomic stability [4]. The overarching aim of this line of research is to establish WES as a comprehensive and integrative indicator system—one that reflects not only the degree of ecological degradation but also its implications for society. This approach provides a scientific basis for designing effective protection and restoration strategies. To spatially quantify and map ecosystem service functions, the InVEST (Integrated Valuation of Ecosystem Services and Tradeoffs) model has been widely employed [5]. However, domestic applications of the InVEST model in China have often prioritized the visualization of spatial and temporal patterns of ecosystem services, rather than targeted and problem-oriented analyses. When multiple InVEST modules are applied, the selection of services is frequently arbitrary and lacks thematic focus.

Despite significant progress, WES research continues to face key challenges, such as data scarcity, model uncertainty, and the complexity of interdisciplinary integration. Future research must strive to deepen our understanding of ecosystem service mechanisms, enhance the precision of assessment and management tools, and strengthen the practical application of scientific insights. As this field evolves, continued contributions from the global research community will be essential to advancing sustainable development goals, safeguarding biodiversity, and ensuring long-term human well-being. In China, current research on ecosystem services predominantly focuses on specific geographic regions and often centers on analyzing spatiotemporal distribution patterns. For example, Yang [6] assessed four types of ecosystem service quality and their trade-offs in the Dongting Lake area. Wang [7] estimated four hydrological ecosystem services in the Dianchi Lake basin and analyzed their spatiotemporal dynamics. Zhao [8] evaluated ecosystem service levels in urban agglomerations and used a coupling coordination model to investigate the temporal and spatial evolution of interactions among economic, social, and ecological subsystems. Wu [9] conducted a quantitative assessment of the spatiotemporal variation in water yield across the Nanpan and Beipan River basins, highlighting regional heterogeneity and the impact of vegetation restoration on water provisioning.

Land use change has been identified as a critical driver of changes in ecosystem services, and increasing attention has been paid to the relationship between WES and urbanization. A growing body of literature has combined WES evaluation with predictive modeling in pursuit of more effective water-related ecosystem management strategies. For instance, Min [10] explored the trade-offs among multiple WESs and examined how these services change under different management scenarios. Shi [11] forecasted the impact of urbanization on WES by projecting future land use patterns. B [12] assessed the spatiotemporal variations in WESs and employed a Coupling Coordination Degree (CCD) model to evaluate the level of coordinated development between WES and urbanization. Raji [13] integrated the Future Land Use Simulation (FLUS) model with InVEST to simulate four development scenarios and assess habitat quality changes. Among the various WES, water yield, soil retention, and water purification have been the most frequently studied services in previous research [14]. The integrated ecosystem service trade-off and synergy assessment framework has become a mature method for evaluating these services and has been widely applied in empirical studies across diverse geographic regions [15,16,17]. Currently, understanding the mechanisms underlying the interaction between urbanization and ecosystems has become a major research focus across disciplines such as geography and ecology [18,19].

Given that urbanization is closely associated with the expansion of built-up land, most studies have approached the issue by examining the ecological impacts of urban expansion through the lens of land consumption. While many studies have analyzed the effects of land use and climate change on hydrological-related ecosystem services, the combined influence of socioeconomic factors has often been overlooked. However, socioeconomic factors play a pivotal role in linking ecosystems with human society. They not only affect the supply of ecosystem services but also shape the demand for them [20]. As key drivers of land use change and urban expansion, socioeconomic development and climate change must be jointly considered in future assessments of WES.

In this study, we focus on land-use change as a proxy for urbanization under different development pathways, using future climate projections from SSP1-1.9, SSP2-4.5, and SSP5-8.5 scenarios alongside static socio-economic variables as spatial drivers. By integrating the Markov model, FLUS model, and InVEST model, we simulate the impact of land use change on water ecosystem services at the regional scale under multiple future scenarios. This approach allows WES and urbanization policy to be adapted in response to future changes in climate and socioeconomic conditions. We construct four distinct land use scenarios: a Natural Development Scenario, an Economic Priority Scenario, an Ecological Conservation Scenario, and a Farmland Protection Scenario. The Markov model is employed to estimate the probability of land use transitions based on historical data [21], while the FLUS model is used to generate spatially explicit land use patterns [22]. FLUS integrates Artificial Neural Networks (ANN) with Cellular Automata (CA), enabling the simulation of complex land use dynamics characterized by nonlinear transition rules in a spatially explicit manner [23,24].

The Greater Bay Area (GBA) is selected as the study area due to its unique geographical location, rapid urbanization, and significant ecological and economic importance. The GBA encompasses a diverse range of ecosystems, including rivers, lakes, and coastal areas, making it an ideal region to investigate the complex interactions between urbanization and WES. This study aims to fill the gap in current literature by providing a comprehensive assessment of WES in the context of urbanization, considering both natural and anthropogenic factors. The findings are expected to offer valuable insights into sustainable urban planning and ecosystem management in the GBA and other rapidly urbanizing regions.

2. Methodology

2.1. Study Area

The study area is the Guangdong-Hong Kong-Macao Greater Bay Area (GBA), located in the southern coastal region of China (21°25′–24°30′ N, 111°12′–115°35′ E). It comprises two Special Administrative Regions—Hong Kong and Macao—as well as nine cities in Guangdong Province: Guangzhou, Shenzhen, Zhuhai, Foshan, Huizhou, Dongguan, Zhongshan, Jiangmen, and Zhaoqing Figure 1). The total area covers approximately 56,000 km².

Situated along both banks of the Pearl River Estuary, the region borders the South China Sea to the east, adjoins Hong Kong and Macao to the south, faces the western bank of the estuary to the west, and is backed by the Nanling Mountains to the north. The topography is characterized by plains, hills, and mountainous areas. The GBA has a subtropical monsoon climate, with hot and rainy summers and mild, drier winters. In 2023, the region’s average temperature reached 23.3 °C, which was 0.7 °C higher than the historical average and the second highest on record. The annual precipitation totaled 1865.7 mm, close to the long-term average, although significantly above normal in early autumn.

The area features diverse vegetation types, including coniferous forests, broad-leaved forests, shrublands, and grasslands. In 2023, the GBA’s permanent resident population increased by more than 440,000, and its total economic output exceeded 14 trillion yuan. It remains one of the most open and economically dynamic regions in China.

However, the region’s water ecosystems are increasingly facing severe challenges. Environmental degradation and water resource scarcity have become pressing issues. Therefore, understanding the spatiotemporal evolution and future projections of water ecosystem services in the Greater Bay Area holds significant theoretical and practical value for sustainable regional development and ecological management.

2.2. InVEST Model Principle and Data Processing

The Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model is a widely used tool for assessing the value of ecosystem services. It supports the analysis of how different land use and management strategies influence the provision of ecological services. Based on Geographic Information System (GIS) technologies, the model integrates principles from ecology and economics to quantitatively evaluate services such as water yield, water purification, and climate regulation.

InVEST enables the simulation of ecosystem service dynamics under various future development scenarios, offering scientific support for decision-making and policy formulation. The model’s core strength lies in its scenario-based analysis framework, which allows for the quantitative assessment of trade-offs and synergies among multiple ecosystem services under divergent socioeconomic pathways. This capability is particularly valuable for regional spatial planning, as it provides empirically grounded insights into the potential consequences of different land management strategies. By analyzing the effects of land cover and land use changes on ecosystem services, the model provides a basis for ecological spatial planning and natural resource management. Through its application, 23 researchers can gain a comprehensive understanding of the spatiotemporal evolution of water ecosystem services in the Guangdong-Hong Kong-Macao Greater Bay Area, thereby offering evidence-based guidance for future ecological conservation and sustainable development efforts.

The primary datasets used in this study were sourced from the Resource and Environment Science and Data Center of the Chinese Academy of Sciences, the China Meteorological Science Data Center, the National Tibetan Plateau Data Center, the FAO World Soil Database, the InVEST model user guide, and relevant published literature (

Table 1. Data sources.

Data Purpose	Required Data	Resolution	Time Range	Data Source
Soil Conservation	Digital Elevation Data	30 m	Static data (2020)	https://www.gscloud.cn/ (accessed on 22 May 2025)
	Land Use Data	1 km	2005–2020 (annual)	https://www.resdc.cn/ (accessed on 22 May 2025)
	Rainfall Erosion Factor	1 km	2005–2020 (annual)	http://www.geodata.cn (accessed on 22 May 2025)
	Soil Erodibility Factor	1 km	Static data (2020)	http://www.ncdc.ac.cn/ (accessed on 22 May 2025)
Water Yield	Precipitation	1 km	2005–2020 (monthly)	http://www.geodata.cn (accessed on 22 May 2025)
	Potential Evapotranspiration	1 km	2005–2020 (monthly)	http://www.geodata.cn
	Root Depth	1 km	Static data (2020)	http://www.ncdc.ac.cn/
	Plant Available Water Content	1 km	Static data (2020)	http://www.ncdc.ac.cn/
	Land Use Data	1 km	2005–2020 (annual)	https://www.resdc.cn/
Water Purification	Digital Elevation Data	30 m	Static data (2020)	https://www.gscloud.cn/
	Land Use Data	1 km	2005–2020 (annual)	https://www.resdc.cn/
	Nutrient Runoff Proxy (N/P)	1 km	2005–2020 (annual)	http://www.geodata.cn

). Data processing was conducted using ArcGIS 10.5 and Excel 2019, including formula calculations, regional clipping, accuracy sampling, raster computation, spatial interpolation, and coordinate system standardization, to ensure compatibility with the requirements of the ecosystem service assessment model applied in this study.

2.2.1. Water Production Module

The InVEST water yield model estimates the annual water yield for different land types based on the Budyko hydrological coupling equilibrium hypothesis (1974) and average annual precipitation data (Fick and Hijmans, 2017). The first step is to determine the annual water yield Y(x) for each grid cell x in the study area, which is represented by the following formula:

$$Y\left(x\right)=\left(1-{\displaystyle \frac{AET\left(x\right)}{P\left(x\right)}}\right)\cdot P\left(x\right)$$

(1)

In the formula, AET(x) represents the annual actual evapotranspiration of grid cell x, and P(x) represents the annual precipitation of that cell.

The calculation of vegetation evapotranspiration AET(x) and precipitation P(x) in the water balance equation is based on the Budyko hydrological-thermal coupling equilibrium hypothesis formula proposed by Fu and Zhang and others:

$${\displaystyle \frac{\mathit{AET}\left(x\right)}{P\left(x\right)}}=1+{\displaystyle \frac{PET\left(x\right)}{P\left(x\right)}}-{\left[1+{\left({\displaystyle \frac{PET\left(x\right)}{P\left(x\right)}}\right)}^{\omega }\right]}^{{\displaystyle \frac{1}{\omega }}}$$

(2)

In the formula, PET(x) represents potential evapotranspiration, and ω(x) is an empirical non-physical parameter that reflects climate and soil characteristics.

$$\omega \left(x\right)={\displaystyle \frac{Z\cdot AWC\left(x\right)}{P\left(x\right)}}+1.25$$

(3)

Z is the seasonal constant. Based on the InVEST model manual and relevant studies on the climate characteristics of the GBA [25,26], this study sets the baseline value of Z as 18.

The crop coefficient (K_c) represents the ratio of actual evapotranspiration to reference evapotranspiration (ET₀), reflecting the water demand of different land-cover types. Lower K_c values indicate lower evapotranspiration and thus higher potential water yield. K_c for vegetated pixels was estimated from the 1 km leaf-area index (LAI) product provided by the Chinese Academy of Sciences using the piece-wise relationship:

$$K_c=\left\{\begin{array}{c}{\displaystyle \frac{LAI}{3}}, LAI\le 3\\ 1, LAI>3\end{array}\right.$$

(4)

For non-vegetated covers (water, built-up, unused land), K_c was assigned from the InVEST global database. Root-depth values were likewise taken from the InVEST lookup tables. The resulting biophysical parameters are summarised in

Table 2. Biophysical coefficient table for water production modules.

Land-Use	Root Depth (mm)	Kc
Farmland	1000	0.67
Woodland	3500	1
Grassland	2000	0.797
Built-up land	10	1
Water	0	0.3
Unused	100	0.5

.

2.2.2. Soil Conservation Module

The soil conservation module is used to simulate slope erosion and sediment transport processes at the watershed scale and is widely applied in the assessment of ecosystem services functions such as soil and sand conservation in catchment areas.

The module is based on the Revised Universal Soil Loss Equation (RUSLE) [27], which first estimates the annual soil erosion amount for grid cell i using the following formula:

$$usl{e}_i=R_i\cdot K_i\cdot L{S}_i\cdot C_i\cdot P_i$$

(5)

Under the condition of not introducing vegetation cover management factors and soil and water conservation measures, the annual potential soil erosion amount for grid cell i can be calculated:

$${rkls}_i=R_i\cdot K_i\cdot L{S}_i$$

(6)

Soil conservation can be calculated by subtracting the annual actual erosion from the annual potential soil erosion:

$$A_i={rkls}_i-usl{e}_i$$

(7)

In the formula, usle_i represents the annual actual soil erosion amount of grid i, rkls_i represents the annual potential soil erosion amount, A_i represents the soil conservation amount; R_i indicates the erosivity factor of precipitation, K_i indicates the erodibility factor of soil, LS_i indicates the slope length and aspect ratio factor, C_i indicates the vegetation coverage and crop management factor, and P_i indicates the soil and water conservation measure factor.

2.2.3. Water Purification Module

Water quality purification services are used to measure the ability of vegetation and soil to purify water. Since soil can intercept and transfer elements such as nitrogen (N) and phosphorus (P), thereby playing a role in water purification, the degree of water quality purification is usually measured by the output amount of nitrogen and phosphorus nutrients in the water quality purification module. The lower the output amount, the stronger the water purification function; conversely, the weaker it is. Ultimately, based on the calculation of annual average runoff, the amount of pollutants intercepted can be determined, and then the output amounts of nitrogen and phosphorus can be inferred.

The specific formula for calculating the output amounts of nitrogen and phosphorus nutrients is as follows:

$$Ex{p}_i=AL{V}_i\prod _{n=i+1}^i\left(1-E_n\right)$$

(8)

where Exp_i is the nitrogen and phosphorus nutrient salt output from upstream grid i to downstream water bodies; En represents the filtration efficiency of each downstream grid n; i represents the transport path of nutrients from generation to downstream water bodies; ALV_i is the corrected nutrient salt output value for grid i.

$$AL{V}_i=HS{S}_i\times PO{L}_i$$

(9)

where ALV_i is the corrected nutrient salt output value in grid i; HSS_i is the hydrological sensitivity of grid i; POL_i is the nutrient salt output coefficient of grid i.

$$HS{S}_i={\displaystyle \frac{{\lambda }_i}{{\lambda }_n}}$$

(10)

where λ_i is the runoff index of grid i; λ_n is the average runoff coefficient of grid i in the study area.

$${\lambda }_i=\mathit{log}\left(\sum _{u=1}^U{Y}_u\right)$$

(11)

where ${\displaystyle {\sum }_{u=1}^U}{Y}_u$ represents the sum of water yield in grid i, which can be obtained through the water yield module.

The biophysical attributes and nitrogen/phosphorus export loads were estimated based on the InVEST User’s Guide and relevant literature from regions adjacent to the study area [28]. The nitrogen- and phosphorus-export parameters adopted in this study are listed in

Table 3. N, P output related parameters.

Land Use Type	TN Load (kg/hm2/a)	TP Load (kg/hm2/a)	Retention Efficiency (%)
Farmland	26	3	25
Woodland	2.8	0.1	70
Grassland	8	1.5	48
Built-up land	0.01	0.01	5
Water	10	2.5	5
Unused	4	0.5	5

.

2.3. Simulate Land Use Methods Under Different Scenarios

2.3.1. Data Preparation and Preprocessing

The research first requires the collection of land use data from 2005 to 2020, sourced from the Chinese Academy of Sciences Resource Environment Science Data Platform: http://www.resdc.cn/ (accessed on 22 May 2025), with a resolution of 30 m; driving factor data includes natural factors (elevation, slope, aspect), socio-economic factors (GDP, population density), climatic factors (annual average precipitation, temperature), and transportation water system distance coefficients. Among them, elevation data is sourced from the Geographic Spatial Data Cloud: https://www.gscloud.cn/ (accessed on 22 May 2025), with a resolution of 1 km, slopes and aspects are derived based on DEMs; population, GDP, precipitation, and temperature data are sourced from the Chinese Academy of Sciences Resource Environment Science Data Platform: http://www.resdc.cn/ (accessed on 22 May 2025), with a resolution of 1 km; vector data such as railways, roads, and water systems are sourced from the OSM official website: https://www.openstreetmap.org/ (accessed on 22 May 2025), and the distances to railways, roads, and water systems are obtained through the ArcGIS Euclidean distance analysis tool. All driving factors need to be normalized to eliminate dimensional differences by unifying the data range to [0,1], as shown in Figure 2. Finally, all continuous-factor rasters were resampled to a common 100 m grid in ArcGIS using bilinear interpolation, while land use data were resampled with nearest-neighbour to preserve class integrity.

2.3.2. Markov Model Predicts Land Use Area

Referring to relevant literature [29,30,31], four scenarios (natural development, economic priority, ecological protection and farmland protection) were designed with the Markov model and aligned with the future land-use planning directives of the Guangdong-Hong Kong-Macao Greater Bay Area.

In the natural development scenario, the transition probabilities between land use types, derived from the historical period of 2005–2020, remain unchanged to project land use patterns to 2035.

In the economic priority scenario, the transition probabilities from Farmland, Woodland, Grassland, Water, and Unused land to Built-up land are increased by 30%, while the probability of Built-up land being converted to other types is significantly reduced. This adjustment aligns with the urban development boundary regulation mandating that “the urban expansion multiplier shall be controlled within 1.3 times” as stipulated in the Guangdong Provincial Territorial Spatial Plan (2021–2035), reflecting realistic and policy-constrained urban growth.

Under the ecological protection scenario, the transition probabilities from Farmland, Built-up land, and Unused land to ecological land (Woodland, Grassland, and Water) are increased by 30%, while the conversion out of these ecological lands is reduced by 30%. These settings align with the ecological conservation strategy emphasized in the Guangdong-Hong Kong-Macao Greater Bay Area Development Plan.

In the farmland protection scenario, the transition probabilities from other land types (Woodland, Grassland, Water, Built-up land, and Unused land) to Farmland are increased by 20%, while the probability of Farmland loss is reduced by 20%. This setting operationalizes the mandatory objective of maintaining “no less than 27.51 million mu of cultivated land” as a bottom-line requirement in the territorial spatial plan, aiming to prevent non-agriculturalization and non-grain utilization of farmland.

2.3.3. FLUS Model Space Allocation

The integration of the Markov chain for quantitative demand forecasting and the FLUS model for spatial explicit allocation forms a coherent modeling framework. This ensures the simulated 2035 land use patterns are both quantitatively aligned with scenario objectives and spatially realistic, providing the essential high-resolution input required to drive the subsequent InVEST model simulations.

This study uses the FLUS model coupled with a Markov model to simulate and predict the spatial distribution pattern of land use in the Guangdong-Hong Kong-Macao Greater Bay Area under four different scenarios for the year 2035. The FLUS model was developed by Liu [22] and others based on the principles of traditional cellular automata, which estimates the development probabilities of various land types within an area through the operation of artificial neural network algorithms on baseline land use data and driving factor data. After a round-robin competition mechanism, the simulation results are obtained. First, drive factor data from 2005 to 2020 is used to train the neural network, generating a land type suitability probability map (validation phase and prediction phase). In the validation phase, the simulated results for 2020 were compared with the actual 2020 land use map, and the Kappa coefficient was calculated as 0.88 (see

Table 4. Confusion Matrix for the Accuracy Assessment of the FLUS Model Simulation for the Year 2020.

Actual\Predicted	Farmland	Wood Land	Grassland	Water	Built-Up Land	Unused	Total
Farmland	248,822	31,282	3735	24,544	27,010	6	335,399
Wood land	15,429	905,298	11,205	18,708	18,455	78	969,173
Grassland	1257	7768	82,308	1344	1235	6	93,918
Water	6523	8802	1113	455,712	8670	0	480,820
Built-up land	5343	7986	975	6712	580,920	12	601,948
Unused	3	40	15	116	20	666	860
Total	277,377	961,176	99,351	507,136	636,310	768	2,482,118
Producer’s Accuracy	89.71%	94.19%	82.85%	89.86%	91.30%	86.72%
User’s Accuracy	74.19%	93.41%	87.64%	94.78%	96.51%	77.44%
Overall Accuracy		0.92	Kappa		0.88

), indicating that the predicted results are highly consistent with reality and can be used for predictions.

Based on scenario settings, neighborhood factor weights and cost transfer matrices are determined. Neighborhood impact factors reflect the interactions between different land types as well as between different land type units within the neighborhood range, with values ranging from 0 to 1. The closer to 1, the stronger the expansion capability of the land type. A 3 × 3 Moore neighborhood is adopted as the neighborhood range. Referring to relevant literature [32], the final determination of the neighborhood impact factor parameters for each land type is shown in

Table 5. Neighborhood Factor Parameter.

Land Type	Farmland	Woodland	Grassland	Water Area	Built-Up Land	Unused
Verification stage	0.00	0.05	0.42	0.33	1.00	0.33
2035 Natural Development Scenario	0.01	0.00	0.42	0.32	1.00	0.35
2035 Economic Priority Scenario	0.00	0.07	0.34	0.25	1.00	0.33
2035 Ecological Protection Scenario	0.00	0.98	0.85	1.00	0.79	0.72
2035 Farmland Protection Scenario	1.00	0.00	0.53	0.36	0.58	0.49

. Conversion costs are used to represent the difficulty level of converting from the current land type to the desired one. This study designed four different conversion costs based on the four scenarios set up. Finally, the Markov predicted area results were input into the FLUS model, combined with suitability probabilities and spatial rules, to generate LUCC raster data for the four scenarios in 2035.

2.3.4. Parameter Sensitivity and Contribution Decomposition Analysis

To evaluate the uncertainty of model parameters and quantify the individual contribution of different land-use changes under the ecological protection scenario, we conducted two supplementary analyses:

(1)

Parameter Sensitivity Analysis: Based on the 2020 baseline land use data, we performed a one-factor-at-a-time sensitivity test on the key parameters of the water yield module. The seasonal constant Z (baseline value = 18) and the crop coefficient (Kc) values for all land types were increased and decreased by 20%, respectively. Additionally, the Kc value for water bodies was individually adjusted by ±20%. The InVEST water yield module was run for each perturbation, and the change rate of the basin-total water yield was recorded.

(2)

Contribution Decomposition Analysis: To clarify the drivers behind the increased water yield under the Ecological Protection (EcoP) scenario, we designed a set of counterfactual experiments. Based on the simulated 2035 EcoP land use map, we created two counterfactual scenarios— “No Forest Gain” and “No Water Gain”—using raster calculator operations to isolate the independent contributions of forest and water expansion to the change in water yield. All scenarios were run through the water yield module using identical meteorological and soil parameters.

2.4. Construction of Spatial Priority Index

We developed a multi-criteria spatial priority index depending on the scenario simulation results. This index identifies areas where conservation or restoration interventions, aligned with the Ecological Protection (EcoP) scenario, are predicted to yield the greatest synergistic benefits for WES enhancement. The analysis was conducted at a 100 m resolution. The priority index (PI) was constructed based on three core criteria:

Potential for WES Improvement (WES_Improve): This criterion captures the magnitude of positive change expected under the EcoP scenario. For each pixel, we calculated the normalized (Z-score) sum of improvements in water yield (WY) and soil retention (SC) and the reduction in total phosphorus export (TP):

$${WI}_i= Z\left(\Delta {WY}_i\right)+ Z\left(\Delta {SC}_i\right)+ Z\left(-\Delta {TP}_i\right)$$

(12)

where Δ represents the change from the 2020 baseline to the 2035 EcoP scenario.

Pressure from Urban Expansion (UP): Proximity to existing urban areas is a proxy for future development pressure and the need for protective ecological buffers. The Euclidean distance to 2020 built-up land (DB) was calculated and inverted so that higher values indicate higher pressure:

$${UP}_i=1-{\displaystyle \frac{DB-{DB}_{min}}{{DB}_{max}-{DB}_{min}}}$$

(13)

Current Pollution Risk (TR): Areas that are currently significant sources of nutrient pollution are high priorities for intervention to achieve quick wins in water quality improvement. The 2020 TP export value was used directly for this criterion.

Each of these three criteria layers was normalized to a 0–1 scale. The overall Priority Index was then calculated as a weighted linear combination:

$${PI}_i=\left(\alpha \ast {WI}_i \right)+ \left(\beta \ast {UP}_i\right)+ \left(\gamma \ast {TR}_i \right)$$

(14)

where α = 0.5, β = 0.25, and γ = 0.25 to emphasize the primary goal of maximizing ecosystem service gains while still accounting for pressure and current risk.

To assess the impact of weight selection on identifying the core priority areas, we conducted a systematic weight sensitivity test. We designed 8 alternative weight combinations (as shown in

Table 6. Weight Sensitivity Analysis Plan.

Scheme	WES Improve (α)	Urban Pressure (β)	Pollution Risk (γ)
Baseline	0.50	0.25	0.25
1	0.60	0.20	0.20
2	0.70	0.15	0.15
3	0.80	0.10	0.10
4	0.40	0.30	0.30
5	0.40	0.40	0.20
6	0.40	0.20	0.40
7	0.33	0.33	0.33
8	0.60	0.30	0.10

), which together with the baseline scheme (α = 0.5, β = 0.25, γ = 0.25), form a spectrum covering various policy preferences. For each weight combination, we calculated the PI value. To quantitatively compare the differences between the results and the baseline scheme, we extracted the pixels ranked in the top 20% of the PI values for each scheme (representing the highest priority areas) and used the Jaccard Similarity Index (JSI) to calculate the spatial overlap degree with the top 20% areas of the baseline scheme. The JSI ranges from 0 to 1, with a higher value indicating a higher degree of spatial consistency.

3. Result

3.1. Evolution of Land Use and Multi-Scenario Simulation

3.1.1. LUCC from 2005 to 2020

Between 2005 and 2020, the Guangdong-Hong Kong-Macao Greater Bay Area (GBA) underwent significant and complex land use transformations, characterized primarily by rapid urban expansion. Throughout this period, forest land remained the dominant land use type, consistently exceeding 53% of the total area, while farmland exceeded 22%. Collectively, these two categories comprised over 75% of the region’s land area, providing a foundation for regional ecological stability and environmental carrying capacity. Nevertheless, sustained urban development triggered notable structural shifts. Farmland, woodland, water bodies, and unused land all experienced reductions: farmland decreased by 8.28%, woodland by 2.15%, water bodies by 4.7%, and unused land declined sharply by 60.30%. Grassland was the sole ecological land use category exhibiting expansion, increasing by 3.27%. In contrast, built-up areas expanded significantly by 28.81%, adding approximately 1936 km².

Spatially, these transformations were most pronounced in central GBA cities including Guangzhou, Shenzhen, Dongguan, and Foshan. The region exhibited a distinct pattern of urban sprawl radiating from Guangzhou and Shenzhen (Figure 3), where built-up areas encroached upon peripheral zones, progressively compressing ecological space and intensifying pressure on ecological carrying capacity. Notably, built-up land emerged as the most dynamic land use type, concentrating within the Pearl River Delta core. Analysis of land use maps reveals its concentrated growth, particularly in Guangzhou, Shenzhen, Foshan, and Dongguan, where urban peripheries advanced, converting substantial farmland and woodland into residential, industrial, and transport uses.

3.1.2. Spatiotemporal Dynamics of Ecosystem Services

The differences in ecosystem services across the four time periods are illustrated in Figure 4. Water yield exhibited significant spatiotemporal heterogeneity across the Guangdong-Hong Kong-Macao Greater Bay Area during 2005–2020. Total annual yields measured 5.31 × 10¹⁰, 5.91 × 10¹⁰, 4.54 × 10¹⁰, and 4.87 × 10¹⁰ m³ in 2005, 2010, 2015, and 2020, respectively, representing a net decrease of 4.4 × 10⁹ m³. The water-yield results for the 2020 baseline exhibited sensitivity to spatial resolution: basin-total values differed by less than 0.7% among 30 m, 100 m, and 1 km grids (48.86 × 10⁹, 48.76 × 10⁹, and 48.53 × 10⁹ m³, respectively). The discrepancy between 30 m and 100 m was below 0.3%, and that between 100 m and 1 km was 0.24%—both within typical floating-point uncertainty. Based on these results, the 100 m resolution was selected for all subsequent scenario analyses as it provides an optimal balance between computational efficiency and process representation. Water yield exhibited pronounced spatial heterogeneity across the GBA. Higher yields characterized central and southwestern regions, where abundant precipitation and extensive impervious surfaces suppressed evapotranspiration. In contrast, forested northwestern and eastern areas showed lower yields due to reduced rainfall and elevated evapotranspiration rates.

Soil conservation exhibited an overall declining trend, with total amounts of 9.05 × 10⁹, 9.83 × 10⁹, 7.74 × 10⁹, and 8.48 × 10⁹ t in the respective study years, representing a net reduction of 5.7 × 10⁸ t. Spatially, marked declines occurred in mountainous and hilly areas of the northern and western regions. These reductions were primarily driven by vegetation degradation, resulting not only from climate stress or disease but also significantly from direct human activities such as agricultural expansion, infrastructure development, and mining, which led to reduced vegetation cover and increased habitat fragmentation. The loss of forest and grassland exposed the soil, weakening root stabilization and surface protection, and initiating a cycle of escalating erosion. Central urban zones maintained relatively stable but consistently low conservation levels due to dense impervious surfaces and limited green infrastructure. Conversely, soil conservation capacity improved in southern and partial western areas, with particularly significant enhancement observed in western Zhaoqing.

TP output increased from 1.22 × 10⁶ kg (2005) to 1.29 × 10⁶ kg (2020), a net rise of 6.3 × 10⁴ kg (+5.2%). Output increases primarily concentrated in agricultural zones, industrial clusters, and rapidly urbanizing areas, driven by intensive farming practices and substantial industrial wastewater discharges that elevated phosphorus concentrations in aquatic systems.

Due to the lack of long-term water quality monitoring data in the study area, the model results were validated using a spatial pattern rationality check. A comparison between the spatial distribution of the simulated total phosphorus (TP) load (Figure 4) and the land use map of the study area (Figure 3) reveals a strong and logical correspondence.

The simulated TP load shows significant spatial heterogeneity, with the highest values concentrated along the main river channels. This pattern aligns with the process of pollutant migration, whereby runoff-transported phosphorus is conveyed through tributary networks and accumulates in the main stems, making riparian zones critical pathways for pollutant generation. Secondary high-load areas exhibit a high spatial coincidence with urban construction land and cultivated land. These high-intensity human activity zones are known primary sources of phosphorus: urban areas generate substantial domestic sewage and surface runoff, while croplands are associated with fertilizer application and soil erosion. Conversely, the model simulated generally low TP loads in areas dominated by forest and grassland. This is consistent with the ecological expectation that these land cover types have a strong capacity for nutrient retention and interception. In conclusion, the high-value TP areas identified by the model spatially correspond with theoretically recognized non-point source pollution hotspots (riparian zones, urban areas, and croplands), while low-value areas align well with regions of ecological conservation. This consistency indicates that, despite the absence of direct quantitative validation, the model’s results are reasonable and reliable in their spatial pattern.

3.1.3. Ecosystem Service Interrelationships

As shown in Table 7, three key ecosystem service relationships emerged in the GBA during 2005–2020:

(1)

A weak trade-off between water production and soil conservation was observed, indicated by consistently negative correlations. After controlling annual precipitation and slope, this trade-off intensified (Δr ≈ 0.11), implying that climatic and topographic factors previously diluted the land-use-mediated competition between the two services. This relationship reflects hydrological dynamics where increased water production involves greater surface runoff, elevating erosion potential—particularly in steep, low-vegetation topography. Urban expansion exacerbated this trade-off through impervious surface growth, accelerating runoff while reducing soil water retention capacity.

(2)

Water production and water purification exhibited a moderate trade-off, demonstrated by strong positive correlations with TP output. The positive correlation became even stronger when precipitation and slope effects were removed (Δr ≈ 0.08), revealing a clearer water-quantity vs. water-quality trade-off driven by land-use change. Elevated water production intensified surface flow velocity, transporting larger nutrients and sediment loads that reduce purification capacity.

(3)

Soil conservation and water purification maintained weak synergistic relationships, evidenced by stable negative correlations with TP output. Partial correlations were almost identical to simple correlations (|Δr| < 0.01), indicating that the synergy is dominated by local land-use rather than climatic or topographic controls. This co-benefit relationship was enhanced through improved contaminant interception by high-coverage vegetation and organic-rich soils, coupled with increased infiltration that retains water-pollutant complexes and reduces runoff-mediated transport.

Table 7. Correlation Matrix of Ecosystem Services from 2005 to 2020.

Ecosystem Services	Year	Simple Correlation (r)	Partial Correlation (r)	Change (Δr)
Water production vs. Soil Retention	2005	−0.265	−0.411	0.146
	2010	−0.253	−0.417	0.164
	2015	−0.316	−0.422	0.106
	2020	−0.306	−0.407	0.101
Water production vs. TP output	2005	0.495	0.606	0.111
	2010	0.484	0.594	0.11
	2015	0.525	0.604	0.079
	2020	0.504	0.58	0.076
Soil Retention vs. TP output	2005	−0.363	−0.369	0.006
	2010	−0.369	−0.366	−0.003
	2015	−0.372	−0.375	0.003
	2020	−0.378	−0.377	−0.001

Note: Partial correlation coefficients control the effects of annual precipitation and slope.

Table 7. Correlation Matrix of Ecosystem Services from 2005 to 2020.

Ecosystem Services	Year	Simple Correlation (r)	Partial Correlation (r)	Change (Δr)
Water production vs. Soil Retention	2005	−0.265	−0.411	0.146
	2010	−0.253	−0.417	0.164
	2015	−0.316	−0.422	0.106
	2020	−0.306	−0.407	0.101
Water production vs. TP output	2005	0.495	0.606	0.111
	2010	0.484	0.594	0.11
	2015	0.525	0.604	0.079
	2020	0.504	0.58	0.076
Soil Retention vs. TP output	2005	−0.363	−0.369	0.006
	2010	−0.369	−0.366	−0.003
	2015	−0.372	−0.375	0.003
	2020	−0.378	−0.377	−0.001

Note: Partial correlation coefficients control the effects of annual precipitation and slope.

3.2. Land Use Change in 2035

This study is based on simulations of different development scenarios in the Guangdong Hong Kong Macao Greater Bay Area in 2035, and analyzes the process and spatial distribution characteristics of land use change under four paths: natural development, farmland protection, economic priority, and ecological protection, as shown in

Table 8. Degree of change in six land use types in four scenarios.

Land Use Type	Natural Development Scenario	Land Conservation Scenario	Ecological protection Scenario	Economic Priority Scenario
Farmland	−4.90%	+10.07%	−12.09%	−12.52%
Wood land	−2.07%	−3.98%	+1.76%	−4.05%
Grassland	+10.62%	+7.13%	+20.88%	+3.71%
Water area	−1.10%	−7.91%	+13.98%	−9.33%
Built-up land	+13.66%	+2.55%	+1.69%%	+36.77%
Unused land	−18.79%	−29.56%	−25.44%	−41.92%

and Figure 5. The Natural Development scenario, which assumes a continuation of current policies and trends, shows significant expansion of built-up land, especially in core urban areas such as Guangzhou, Shenzhen, Dongguan, and Foshan. Agricultural land—particularly farmland and forest—is rapidly replaced, and in peripheral areas like Zhaoqing and Qingyuan, some degraded farmland naturally transforms into grassland due to lack of regulation.

The Farmland Protection scenario emphasizes strict land-use regulations, especially the enforcement of the farmland redline and land consolidation policies. Farmland increases in peripheral cities like Zhaoqing, Qingyuan, and Heyuan. Urban development is constrained and primarily occurs around existing city boundaries, reflecting a more compact and efficient urbanization process.

The Economic Priority scenario focuses on rapid urbanization and industrial development to drive economic growth. Under this scenario, built-up land expands extensively, especially in cities like Guangzhou, Dongguan, Huizhou, and Zhuhai. This comes at the cost of large-scale loss of farmland and forest, with some ecological redline zones also facing development pressure. The result is a dramatic restructuring of land use types and spatial patterns.

In contrast, the Ecological Protection scenario emphasizes ecological functionality and landscape integrity. Through the designation of ecological redlines, the restoration of ecological corridors, and the rehabilitation of key ecological nodes, this scenario strictly limits urban expansion and prioritizes the restoration of forests and wetlands. Built-up land growth is minimized, while forest and grassland areas increase significantly. This results in a spatial pattern characterized by concentrated urban development and expansive ecological protection, simultaneously improving land-use efficiency and ecological stability.

Presentation of Sensitivity Analysis

To assess the uncertainty of key parameters in the water production model, we conducted a single-factor sensitivity analysis on the seasonal constant (Z) and crop coefficient (K_c). Based on the baseline land use in 2020, the results are as follows:

(1)

When the Z value decreases by 20% from the baseline of 18 to 14.4, the total water yield of the basin increases by 2.7% (from 4.87 × 10¹⁰ m³ to 5.00 × 10¹⁰ m³). When the Z value increases by 20% to 21.6, the water yield decreases by 1.8% (to 4.78 × 10¹⁰ m³). This indicates that the model responds robustly to changes in the Z parameter and the direction is consistent with theoretical expectations.

(2)

The model shows high sensitivity to the K_c parameter. When the K_c values of all land types are simultaneously reduced by 20%, the total water yield of the basin increases significantly by 17.0% (to 5.70 × 10¹⁰ m³). Conversely, when all K_c values are increased by 20%, the water yield decreases sharply by 16.6% (to 4.06 × 10¹⁰ m³).

(3)

Increasing the K_c value of water bodies alone by 20% leads to a decrease in water yield of 2.1%, while decreasing it by 20% results in an increase of 0.6%. The model shows the highest sensitivity to the K_c value of forests. Decreasing the K_c value of forests alone by 20% can increase the water yield by 10.7%, while increasing it by 20% leads to a decrease of 10.3%.

3.3. Ecosystem Service Changes

Land use changes under the four scenarios have significant and complex effects on ecosystem services, including water yield, soil conservation, and phosphorus output, as shown in

Table 9. Ecosystem service supply and its percentage change under four different scenarios in 2020 and 2035.

Indicator	Water Production (109/m3)	Change Rate (%)	Soil Conservation (108 t)	Change Rate (%)	Phosphorus Output (104 kg)	Change Rate (%)
The current situation in 2020	48.77		84.82		128.74
Natural Development scenario	64.50	32.25	123.97	46.16	128.67	−0.05
Farmland Protection scenario	64.44	32.13	123.76	45.91	134.95	4.82
Economic Priority scenario	42.24	−13.39	83.03	−2.11	136.77	5.87
Ecological Protection scenario	73.48	50.67	144.99	70.94	118.22	−8.17

. In the Natural Development scenario, ecosystem services exhibit spatial differentiation. Some marginal and mountainous areas experience increases in water yield and soil conservation due to grassland and forest expansion. However, in highly urbanized areas, green space reduction and impervious surfaces lead to reduced water regulation capacity and increased phosphorus runoff.

The Farmland Protection scenario enhances ecosystem services by stabilizing farmland area and creating clearer spatial boundaries between agricultural and ecological land. This configuration improves soil retention and rainwater regulation, especially in the transitional zones of hills and plains in western and northern GBA. Water yield and soil conservation capacity increase significantly in critical ecological function areas.

The Economic Priority scenario results in the most severe ecological degradation. As ecological land is widely converted to built-up land, forest and wetland ecosystems shrink, significantly reducing regional water yield and soil retention capacity. Impervious surfaces increase runoff and exacerbate urban flooding risks. In addition, widespread agricultural intensification and industrial expansion lead to a sharp rise in phosphorus output, increasing the risk of water pollution.

In the Ecological Protection scenario, ecosystem services are significantly enhanced. Reforestation, wetland restoration, and water source conservation efforts lead to a substantial increase in water yield, particularly in the northwest mountains and upper Pearl River. Increased vegetation cover reduces erosion, improving soil conservation. Reduced human disturbance results in a significant decrease in phosphorus output, enhancing the region’s capacity for pollutant filtration and overall ecosystem resilience.

3.3.1. Sensitivity Analysis of Policy Constraints

To test the sensitivity of the model to spatial constraints, we compared the simulation results under the ecological protection scenario using current method with those using policy constraint method. The policy constraint method sets hard constraints in the FLUS model, prohibiting any development and construction within the ecological red line and floodplain areas.

As shown in

Table 10. The impact of constraints on ecosystem services under ecological protection scenarios.

Ecosystem Service Indicators	Current Method	Policy Constraint Method	Change
Water Production (109/m3)	73.48	73.96	+0.48
Soil Conservation (108 t)	144.99	144.86	−0.13
Phosphorus Output (104 kg)	118.22	125.12	+6.9

, the introduction of spatial hard constraints has produced complex marginal effects on ecosystem services. Policy constraints effectively protected key ecological spaces and slightly increased water yield (+0.65%). However, it also led to a significant increase in total phosphorus output of 5.84%. This seemingly contradictory result reveals a potential ‘crowding-out effect’ of policies: construction activities that were prohibited from developing in sensitive areas have shifted and intensified non-point source pollution pressures in regions such as agriculture or urban-rural fringe zones.

3.3.2. Driver Decomposition Analysis

To clarify the driving mechanisms of water yield changes under the EcoP scenario for 2035, we have quantitatively decomposed the contributions of climate change and land use chang. The analysis results show that future changes in water yield in the study area are mainly dominated by climate factors. As shown in

Table 11. Results of the driver decomposition analysis for water yield change under the EcoP scenario.

Driver	Contribution/109 m3	Percentage of Total Change
Total Change(ΔY_Total)	+24.71	100.0%
Climate Change (ΔY_Climate)	+25.12	+101.7%
Total LULC Effect (ΔY_LULC)	−0.41	−1.7%
Water Expansion	−0.19	−0.8%
Forest Expansion	−0.30	−1.2%
Interaction and Others	+0.08	+0.3%

, the total increment in water yield from 2020 to 2035 (ΔY_Total = 24.71 × 10⁹ m³) is dominated by the contribution of climate change (ΔY_Climate) at 25.12 × 10⁹ m³, accounting for 101.7% of the total increment. In contrast, land use change under the EcoP scenario generally has a slight negative effect (ΔY_LULC = −0.41 × 10⁹ m³). Further decomposition of the land use change effect reveals its complex internal trade-offs: The expansion of forests, due to its high evapotranspiration water consumption, had an effect of −0.30 × 10⁹ m³ on regional water yield (i.e., reducing water production). Similarly, the expansion of water bodies also had a negative effect of −0.19 × 10⁹ m³ due to water surface evaporation. The combined effect of other land use changes contributed +0.08 × 10⁹ m³.

3.4. Spatial Distribution of Ecosystem Services

The spatial distribution of ecosystem services varies significantly across the four scenarios, revealing distinct regional impacts of different development strategies, as shown in Figure 6. Under the Natural Development scenario, core urban clusters such as Guangzhou, Shenzhen, and Foshan exhibit pronounced degradation of ecosystem services. In contrast, mountainous areas in Zhaoqing and Qingyuan show improvement in water yield and soil conservation, resulting in a sharp urban-rural ecological disparity.

The Farmland Protection scenario presents a more balanced spatial distribution of services. The establishment of ecological buffers around urban fringes enhances transition zones between cities and rural areas. In eastern and northern agricultural areas, the mosaic of farmland and ecological land creates a resilient landscape pattern, with improved regulation of water and soil-related services.

The Economic Priority scenario results in a fragmented ecological service pattern. Ecosystem functions are severely diminished in urbanized and coastal areas, while only some high-altitude, undeveloped mountainous zones retain functional ecosystems. The spatial pattern reflects an “urban ecological collapse with marginal ecological support.”

In contrast, the Ecological Protection scenario forms a cohesive ecological network built on restored forests, wetlands, and water bodies. Ecosystem services shift from point-based to corridor- and patch-based distributions, with greatly enhanced ecological connectivity. Urban green rings, mountain corridors, and water networks collectively construct a spatially continuous and functionally stable ecological structure. This integrated system effectively mitigates the ecological pressure of urban expansion and promotes regional ecological security.

3.5. Spatial Prioritization for Targeted Management

The spatial priority map (Figure 7) synthesizes the scenario outcomes to provide a clear, actionable geography for land-use management and investment. The distribution of priority levels reveals distinct spatial patterns:

(1)

Priority 1 (Highest) zones are highly concentrated and strategically positioned, particularly at the junction of Huizhou, Dongguan, and Shenzhen.

(2)

Priority 2 (High) zones form core ecological patches around the major metropolitan areas of Guangzhou and Shenzhen. These regions face significant urban pressures and represent critical areas for enhancing urban ecological quality through the development of green infrastructure aimed at mitigating diffuse pollution.

(3)

Priority 3 (Medium) zones are widely distributed across the peri-urban regions and township clusters of Guangzhou, Zhongshan, Foshan, and Dongguan. These areas serve as vital transitional buffers between high- and low-priority zones and are essential for implementing structural ecological controls to curb urban sprawl.

(4)

Priority 4 (Low) zones are mainly located in Zhaoqing and Huizhou. These regions have a sound ecological foundation and are subject to relatively low development pressure. The primary strategy here should emphasize preventive protection to maintain existing ecosystem service functions.

(5)

Priority 5 (Lowest) zones form expansive, contiguous tracts along the western coast, particularly within Jiangmen, Zhuhai, and Zhongshan. These areas include key natural reserves, which are ecologically sensitive and recognized for their high conservation value. This priority map effectively moves beyond narrative guidance, providing a spatially explicit decision-support tool for targeting conservation efforts where they are needed most and will be most effective.

Figure 7. Spatial priority map for water ecosystem service management.

Figure 7. Spatial priority map for water ecosystem service management.

Weight Sensitivity Results

Weight sensitivity analysis demonstrates that the identification of the highest priority areas (top 20%) in this study exhibits high robustness. As shown in

Table 12. Sensitivity Analysis of Weighting Schemes Using Jaccard Similarity Index (JSI).

Scheme	α: β: γ	Jaccard Similarity Index (JSI)	Schema	α: β: γ	Jaccard Similarity Index (JSI)
1	0.60: 0.20: 0.20	0.897	5	0.40: 0.40: 0.20	1.000
2	0.70: 0.15: 0.15	0.765	6	0.40: 0.20: 0.40	0.900
3	0.80: 0.10: 0.10	0.620	7	0.33: 0.33: 0.33	0.900
4	0.40: 0.30: 0.30	0.932	8	0.60: 0.30: 0.10	0.693

Note: JSI ranges from 0 to 1, with higher values indicating greater spatial consistency. The baseline scheme uses weights of 0.5: 0.25: 0.25.

, the Jaccard Similarity Index (JSI) between all alternative weight schemes and the baseline scheme exceeds 0.61, with an average value of 0.84. Specifically, Scheme 5 (0.4, 0.4, 0.2) shows perfect consistency with the baseline scheme (JSI = 1.000), while Scheme 4 (0.4, 0.3, 0.3) achieves an extremely high similarity (JSI = 0.932). Even under the extremely ecology-biased Scheme 3 (0.8, 0.1, 0.1), the JSI remains at 0.620, indicating that the core areas are still well identified.

4. Discussion

4.1. Impacts of Land Use Change on Ecosystem Services

The analysis of land use change reveals significant spatial heterogeneity in ecosystem service provision under different scenarios. In the “Natural Development” scenario, continued urbanization in the absence of effective regulatory controls leads to pronounced negative effects from land use transformation. Particularly in major urban centers such as Guangzhou, Shenzhen, and Foshan, the extensive expansion of built-up land has resulted in a substantial reduction in forest and farmland areas. This trend reduces water yield and soil retention capacity, increases surface runoff, and intensifies nutrient loading, thereby exacerbating local water quality degradation. The pattern observed under this scenario aligns with common issues in rapidly urbanizing regions, where urban expansion often outpaces ecological conservation, resulting in a continual decline in ecosystem functionality.

In contrast, the “Farmland Protection” scenario emphasizes the stability of agricultural land and the implementation of corresponding protection policies. The results indicate more favorable outcomes for water-related ecosystem services, particularly in terms of soil conservation and water retention. Limiting urban sprawl and preserving farmland effectively safeguards the region’s ecological infrastructure, especially in peripheral cities such as Zhaoqing, where both farmland and forested areas are maintained. These findings underscore the critical role of agricultural land protection in mitigating the ecological impacts of urbanization and maintaining essential ecosystem services provided by natural landscapes.

4.2. Balancing Economic Development and Ecological Health

The “Economic Priority” scenario clearly illustrates the trade-off between economic growth and ecological sustainability. In this scenario, large-scale expansion of urban and industrial land significantly undermines ecosystem service functions, as evidenced by decreased water yield, reduced soil retention capacity, and increased phosphorus export. These outcomes highlight a central challenge in rapidly developing regions: prioritizing economic growth often accelerates ecological degradation, threatening the long-term sustainability of regional development. The findings suggest that without prudent planning and management, aggressive development may lead to irreversible ecological damage, thereby weakening ecosystem services such as water purification and flood regulation.

In contrast, the “Ecological Protection” scenario produces the most promising results in terms of ecological restoration and sustainable land use practices. The restoration of forests, wetlands, and other natural habitats significantly enhances water yield, improves soil retention, and effectively reduces phosphorus export. This scenario underscores the synergistic effects of integrating ecological protection with strategic land-use planning. The recovery of natural systems not only strengthens ecosystem service functions but also lays a robust foundation for sustainable urban development. The optimized spatial distribution of services, especially the improved connectivity of ecological networks and the restoration of natural corridors, further emphasizes the importance of maintaining ecological integrity amidst urban expansion. These outcomes strongly align with the concept of Nature-based Solutions (NbS), as the proactive restoration of forests and wetlands exemplifies a core NbS principle—working with nature to address societal challenges. This demonstrates that ecological measures can serve as effective pathways for achieving sustainable watershed management.

4.3. Spatial Optimization for Sustainable Development

A key contribution of this study lies in its emphasis on spatial optimization as an effective strategy for reconciling land use change with ecosystem resilience. Spatial analysis results indicate that a uniform approach to land management is inadequate; instead, differentiated strategies should be formulated based on the specific characteristics of each region. For instance, in highly urbanized areas, a green transformation strategy is essential—enhancing urban permeability can mitigate the ecological impacts of impervious surfaces. In peripheral agricultural zones, efforts should focus on controlling diffuse pollution and promoting the integration of ecological and agricultural land use.

Moreover, the study highlights the importance of maintaining and restoring ecological corridors and buffer zones on the urban fringe. These measures enhance the connectivity of fragmented ecosystems, support biodiversity conservation, and strengthen the resilience of hydrological systems. In regions such as the Greater Bay Area, where urbanization pressure is particularly high, spatial optimization strategies must balance ecological conservation with the spatial demands of urban and economic development, minimizing ecological degradation while accommodating population growth.

4.4. Policy Implications and Future Research Directions

The findings of this study underscore the necessity of developing integrated policies that simultaneously promote development and environmental protection. Policymakers must recognize the close interconnection between land use change and ecosystem services to avoid sacrificing ecological health for urban and industrial expansion. The results indicate that policies supporting sustainable urban planning, ecological restoration, and farmland protection can significantly enhance the capacity of regions to sustain critical ecosystem functions. Furthermore, innovative policy instruments like the design of blockchain-based ecological compensation mechanisms could be explored to enhance transparency and efficiency in cross-jurisdictional environmental management, using our quantified ES values as a potential basis for transactions.

Furthermore, this study identifies several promising directions for future research. While the InVEST model provides valuable support for analyzing the impact of land use changes on ecosystem services, future studies could incorporate additional ecological components, such as biodiversity and carbon sequestration. In addition, investigating socioeconomic factors—such as income growth and population dynamics—and their influence on land-use decisions and ecological outcomes would provide a more comprehensive understanding of the mechanisms at play. The quantitative findings of this study are inherently shaped by the unique geographical, climatic, and socio-economic context of the Greater Bay Area. Therefore, the specific numerical outcomes should not be directly extrapolated to other regions without careful consideration of local conditions. Nevertheless, the methodological framework combining land-use simulation with ecosystem-service assessment under multiple scenarios remains transferable. Future research should apply this approach in comparative studies across different urban agglomerations to identify both universal patterns and context-specific dynamics, while also exploring innovative approaches such as blockchain-based ecological compensation mechanisms.

4.5. Uncertainty and Limitations

Although this study reveals main robustness conclusions through a multi-scenario framework and sensitivity analysis, there are still several sources of uncertainty. First, scenario uncertainty exists in the setting of future land use conversion probabilities. Although we have set fixed disturbances based on historical trends and policy goals, future socio-economic pathways inherently lack predictability. Second, biophysical parameter uncertainty affects the absolute values of ecosystem service amounts, such as the parameters Z and crop coefficient Kc in the Budyko formula. Third, regarding scale effects, although we have verified that a resolution of 100 m can effectively balance precision and efficiency for the baseline year of 2020, the sensitivity of change magnitudes to spatial scales under future scenarios still needs further examination, for example, running all model modules at a resolution of 30 m. These uncertainties suggest that the quantitative results of this study should be understood as best estimates under specific assumptions, with their core value lying in revealing the relative trends and trade-offs across different development paths, rather than providing precise predictions.

5. Conclusions

This study provides a comprehensive assessment of the impacts of land use change on water-related ecosystem services (WESs) in the Guangdong-Hong Kong-Macao Greater Bay Area (GBA) under four development scenarios projected to 2035, utilizing historical climate data and future climate projections from SSP1-1.9, SSP2-4.5, and SSP5-8.5, alongside static socio-economic inputs. By integrating spatial simulation and ecosystem service modeling, the analysis reveals how different land-use policies shape ecological outcomes and offers actionable insights for sustainable land management in one of China’s most densely populated and economically dynamic regions.

(1)

Spatial Heterogeneity in Ecosystem Service Responses: This study demonstrates substantial spatial heterogeneity in water-related ecosystem service (WES) responses—including water yield, soil retention, and nutrient regulation—under four land-use development scenarios in the Guangdong-Hong Kong-Macao Greater Bay Area (GBA). These findings underscore the necessity of spatially explicit and context-sensitive land management strategies to achieve ecological sustainability in rapidly urbanizing regions.

(2)

Superior Effectiveness of the Ecological Protection Scenario: Among all scenarios analyzed, the Ecological Protection scenario yields the most favorable ecological outcomes. It significantly enhances water yield and soil conservation functions while effectively reducing phosphorus export. These results affirm that land-use strategies centered on ecological restoration and environmental conservation can produce synergistic benefits, supporting both resource sustainability and environmental quality.

(3)

Ecological Trade-offs under Economic Development Prioritization: The Economic Priority scenario, characterized by accelerated urban expansion and industrialization, leads to pronounced degradation of ecosystem services. The proliferation of impervious surfaces weakens hydrological regulation and exacerbates pollution risks, highlighting the ecological trade-offs and long-term environmental costs associated with development-driven land-use approaches.

(4)

Necessity for Differentiated Land Management Strategies: The spatial distribution of WES indicates a pressing need for region-specific management strategies. In urban areas, implementing green infrastructure and reducing impervious surface coverage are critical to restoring hydrological functionality. In agricultural zones, mitigating diffuse pollution and integrating ecological land uses can enhance regulatory services. Additionally, restoring ecological corridors and buffer zones is essential for improving landscape connectivity, supporting biodiversity, and strengthening the resilience of water-related ecosystem functions.

(5)

Policy Implications and Directions for Future Research: The results highlight the importance of incorporating ecosystem service indicators into territorial spatial planning frameworks to guide sustainable regional development. Policymakers should prioritize adaptive, spatially optimized land-use planning approaches that balance economic growth with long-term ecological integrity. Future research should integrate additional ecosystem functions such as biodiversity conservation and carbon sequestration, along with socioeconomic drivers, to improve our understanding of ecosystem dynamics and support evidence-based land-use decision-making. Furthermore, future research should strive to quantify the uncertainties of this research framework more comprehensively, further enhancing the model’s decision support capabilities in supporting refined and adaptive spatial planning.