Revealing Multiscale Characteristics of Ecosystem Service Flows: Application to the Yangtze River Economic Belt

Abstract

Assessing ecosystem service (ES) supply–demand relationships and identifying their driving forces are essential for ecological security and sustainable ecosystem development. Using ES supply–demand mismatches as a basis, this study characterized the spatiotemporal evolution of ES supply and demand from 2000 to 2023. Additionally, a SHAP-informed Stacking Bayesian optimization model was employed to identify key drivers of supply–demand imbalances. Building on this, threshold-aware spatial optimization of ecosystem service flows was performed using an improved minimum-cost algorithm within an NSGA-II multi-objective framework. The results showed that: (1) The YREB’s supply–demand balance (SDB) exhibited significant spatial heterogeneity. Water SDB declined with fluctuations, decreasing from 5.343 × 10¹¹ m³ to 4.433 × 10¹¹ m³, whereas carbon SDB shifted from a surplus (+1.514 × 10⁹ t) to a deficit (−1.673 × 10⁹ t) during the study period. Crop SDB rose from 1.361 × 10⁸ to 1.450 × 10⁸ t across the study period. (2) Nighttime light intensity (NLI) was the dominant factor for water SDB and carbon SDB, while cropland area was the key driver for crop SDB. (3) Over 2000–2023, water SDB flow increased from 8.5 × 10⁹ m³ to 1.43 × 10¹⁰ m³. Carbon SDB flows more than tripled from 9.576 × 10⁷ tons to 2.89 × 10⁸ tons. Crop SDB flow increased nearly twelvefold over 2000–2023, from 3.3 × 10⁵ t to 3.93 × 10⁶ t. The findings provide scientific support for coordinating ecological conservation and high-quality development across the Yangtze River Economic Belt.

1. Introduction

Ecosystem services (ESs), the vital interface between human and natural systems [1], include essential functions such as food production, water provision, carbon sequestration, and soil retention [2,3,4]. These services play indispensable roles in maintaining climate stability, water security, and food security [5,6]. However, escalating anthropogenic pressures, including rapid population growth, land use intensification, and the increasing frequency of extreme climate events, have led to widespread mismatches between ecosystem service supply and human demand [7,8], thereby threatening both ecological sustainability and human well-being [9].

The spatial scales of ES studies have expanded significantly, evolving from site-specific or watershed-level assessments to national and even global investigations, thus enabling a broader understanding of regional and systemic interactions [10,11,12]. Amid accelerating globalization and regional interdependence, ecosystem services are increasingly redistributed across space through multiple pathways, including commodity trade, inter-basin water transfers and the outsourcing of carbon emissions, a phenomenon conceptualized as ecosystem service flow [13,14]. This shift toward recognizing the spatial decoupling of ES supply from benefit has offered new perspectives on cross-boundary ecological interactions [15,16], revealing complex socio-ecological linkages and the need for integrated, transregional approaches to ecosystem governance.

Since the concept of ecosystem service flow (ES flow) was introduced, accurately quantifying the spatial redistribution of services from supply areas to beneficiary regions has become a central focus in ecosystem service research. A growing body of studies has developed diverse methodological approaches to map and measure the spatial trajectories and intensities of ES flows at varying scales and from different perspectives. For instance, the breakpoint model has been used to quantify the magnitude of service transfers between regions [17,18], while the supply–demand ratio has served as a diagnostic metric for identifying the directionality and intensity of ES flows [19,20]. Additionally, the ARIES (Artificial Intelligence for Ecosystem Services) platform has enabled simulation of probabilistic service pathways from supply to demand zones [21]. Recent studies have sought to integrate physical flow (PF) concepts with ES flows within unified assessment frameworks, thereby improving the realism and policy relevance of ES flow analysis [22]. Yet, most approaches are implemented on a per-service basis, either analyzing one service at a time or stacking layers without endogenous coupling. So cross-service co-movement, competition, and trade-offs are rarely modeled within a unified allocative framework. In complex socio-ecological systems, ecosystem services such as carbon sequestration and water yield often exhibit spatial interdependencies, including synergistic interactions and trade-offs [23,24,25]. These services may follow shared pathways, exhibit functional synergy, or compete for limited ecological resources [26,27]. Thus, there is a pressing need to develop integrated modeling frameworks that capture the interdependence of multiple ES flows, incorporating multisource service layers, multiscale transfer paths, and multi-beneficiary objectives. Such frameworks are essential for revealing the nonlinear spatial dynamics that govern ecosystem–society interactions and for characterizing the complex redistribution patterns of composite ES flows. Addressing these gaps, source–sink landscape theory is coupled with a multilayer network and a threshold-aware minimum-cost optimizer operating on a pressure basis; exports from stressed nodes are curtailed while deficit relief receives priority. The framework simulated the spatiotemporal evolution of water, crop, and carbon services, enabling systematic tracking of redistribution under explicit supply–demand thresholds.

This study focuses on the Yangtze River Economic Belt (YREB), one of the most prominent regions in China where tensions between economic development, natural resource utilization, and biodiversity conservation are highly pronounced. The YREB exemplifies a nature–economy–society coupled socio-ecological system, encompassing diverse ecosystems, including mountains, rivers, forests, farmlands, lakes, and grasslands. It hosts 41.3% of China’s forest cover, around 20% of national surface water bodies, 39.7% of the country’s rare and endangered plant species, and 33% of its freshwater fish [28,29]. However, escalating tensions between socio-economic development and ES sustainability have degraded ecosystem stability and biodiversity, significantly weakening the region’s ES supply capacity [30,31]. Against this backdrop, this study aims to address the following questions:

(1)

How can a regional-scale evaluation framework be constructed to accurately identify supply–demand balance of key ESs, such as water yield, crop production, and carbon sequestration, across the YREB?

(2)

How do ESs flow within the region, and what spatial flow characteristics do critical ESs exhibit?

(3)

How can the spatial configuration and supply–demand regulation of multiple ESs be optimized at the regional level to achieve coordinated multifunctionality and ecological sustainability?

2. Materials and Methods

2.1. Study Area

The YREB is a strategically designated region aligned with the Yangtze River Basin and national development priorities (Figure 1). Spanning 11 provincial-level administrative units, it stretches across eastern, central, and western China. Covering 2.05 million km² (21% of China’s territory), the YREB is home to over 595 million people and generates more than 40% of the national GDP. Its terrain is highly diverse, ranging from mountains and plateaus to basins and plains, and supports a wide array of ecosystems. The YREB hosts extensive river–lake wetlands, subtropical evergreen forests and large cropland plains, delivering high-value ecosystem services. Given its ecological complexity and economic prominence, the YREB functions as a national backbone of ecosystem service supply and redistribution and thereby anchors regional integration and sustainable development.

2.2. Data Collection

The data included land use, meteorology, cropland area, population, economic output, urbanization, and industrial structure, detailed data descriptions were presented in

Table 1. Data description.

Data	Resolution	Description	Source
LULC	1 km	classified into six land cover types: cropland, forest, grassland, water, construction land, and unused land.	https://www.resdc.cn (assessed on 1 July 2025) [32]
precipitation	1 km	summing the monthly data	https://www.resdc.cn (assessed on 1 July 2025) [32]
evapotranspiration	1 km	summing the monthly data	https://www.resdc.cn (assessed on 1 July 2025) [32]
Normalized difference vegetation index (NDVI)	1 km	generated using the maximum value composites	https://www.resdc.cn (assessed on 1 July 2025) [32]
Digital elevation model (DEM)	1 km	projected onto the Yangtze River Economic Belt	https://www.gscloud.cn (assessed on 1 July 2025)
crop consumption		calculated by equation	China Statistical Yearbook
water consumption		calculated by intensity of water use	China Statistical Yearbook
carbon emission		calculated by equation	China Statistical Yearbook
cropland area	1 km	statistics by prefecture-level city	Yang et al., 2025 [33]
nighttime light intensity	1 km	statistics by prefecture-level city	Wu et al., 2021 [34]
population		statistics by prefecture-level city	China Statistical Yearbook

. All datasets were resampled to a 1 km spatial resolution.

2.3. Methodology

This study adopts an integrated framework of ecosystem service assessment–supply–demand balance analysis–flow optimization (Figure 2). First, it assesses the temporal trajectories of ecosystem service dynamics across the YREB from 2000 to 2023 using the InVEST 3.14.2 software combined with the nonparametric Theil–Sen slope estimator. Subsequently, the ecosystem service supply–demand balance (SDB) and its dominant drivers were evaluated using a combination of mathematical formulations and a SHAP-enhanced Stacking–Bayesian optimization Model. Finally, an improved minimum-cost algorithm based on the NSGA-II multi-objective evolutionary framework was employed to simulate the spatial flow patterns of ecosystem services within the YREB during the study period. By quantifying the spatial dynamics of ecosystem service flows and identifying mismatches between supply and demand, this study provides a scientific basis for delineating ecological surplus and deficit zones.

2.3.1. Ecosystem Services Assessment

The selection of ecosystem services in this study integrates economic, social, and environmental dimensions while ensuring data availability and statistical reliability (

Table 2. Indicator of ecosystem services.

Indicators	Explanation
carbon sequestration	capacity of ecosystems absorbing carbon dioxide from the atmosphere
water yield	capacity of ecosystems capturing and storing water
soil retention	capacity of ecosystems preventing soil erosion and degradation
water purification	capacity of ecosystems assimilating nutrient loads (e.g., nitrogen and phosphorus)
crop yield	capacity of ecosystems providing crop

). Five representative ecosystem services, carbon sequestration, water yield, soil retention, water purification, and crop yield, were evaluated using the Integrated Valuation of Ecosystem Services and Trade-offs (InVEST 3.14.2 software) model and the Carnegie–Ames–Stanford Approach (CASA), with validation based on annual net primary productivity (NPP) data (MOD17A3HGF.006).The InVEST 3.14.2 software, supported by the Natural Capital Project, is widely used for evaluating the spatiotemporal dynamics of ecosystem services at multiple scales due to its efficiency and capacity to simulate diverse service types [35]. Further methodological details are provided in the Supplementary Materials. Given the substantial variation in land area across prefecture-level cities within the YREB, direct comparison of total ecosystem service values may introduce significant bias. To enhance horizontal comparability, all ecosystem services were normalized on a per-unit-area basis. This standardization enables more accurate assessments of ecosystem service provision intensity across spatial units. The per-area ecosystem service intensity (ESI) was calculated as follows:

$$ESI={\displaystyle \frac{E{S}_{ij}}{A_i}}$$

(1)

where $E{S}_{ij}$ denotes the value of the j-th ecosystem service in the i-th prefecture-level city, and $A_i$ represents the area of the i-th city.

2.3.2. Theil–Sen Trend Estimator

Monotonic trends in long-term ecosystem service dynamics were assessed using the Theil–Sen estimator, a robust nonparametric method widely used in environmental and ecological time-series analysis. Unlike ordinary least squares regression, the Theil–Sen approach is less sensitive to outliers and does not assume normality or homoscedasticity of residuals [36]. It computes the median of all pairwise slopes between temporal observations, providing a consistent estimate of the direction and magnitude of the linear trend. The slope β is calculated as the median of the pairwise slopes among all observations:

$$\beta =\mathrm{median}\left({\displaystyle \frac{x_j-x_i}{j-i}}\right), \mathrm{for} \mathrm{all}1\le i<j\le n$$

(2)

where $x_i$ and $x_j$ are the observed values at time steps $i$ and $j$, respectively, and $n$ is the total number of observations. Trends were identified as follows:

$$Trend=\left\{\begin{array}{l}”\mathrm{S}\mathrm{i}\mathrm{g}\mathrm{n}\mathrm{i}\mathrm{f}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{n}\mathrm{t} \mathrm{i}\mathrm{n}\mathrm{c}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{e}”, \mathrm{i}\mathrm{f} \beta >0 \mathrm{a}\mathrm{n}\mathrm{d} p<0.05\\ ”\mathrm{S}\mathrm{l}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t} \mathrm{i}\mathrm{n}\mathrm{c}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{e}”, \mathrm{i}\mathrm{f} \beta >0 \mathrm{a}\mathrm{n}\mathrm{d} 0.05\le p<0.1\\ ”\mathrm{N}\mathrm{o} \mathrm{s}\mathrm{i}\mathrm{g}\mathrm{n}\mathrm{i}\mathrm{f}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{n}\mathrm{t} \mathrm{t}\mathrm{r}\mathrm{e}\mathrm{n}\mathrm{d}”, \mathrm{i}\mathrm{f} p\ge 0.1\\ ”\mathrm{S}\mathrm{l}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t} \mathrm{d}\mathrm{e}\mathrm{c}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{e}”, \mathrm{i}\mathrm{f} \beta <0 \mathrm{a}\mathrm{n}\mathrm{d} 0.05\le p<0.1\\ ”\mathrm{S}\mathrm{i}\mathrm{g}\mathrm{n}\mathrm{i}\mathrm{f}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{n}\mathrm{t} \mathrm{d}\mathrm{e}\mathrm{c}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{e}”, \mathrm{i}\mathrm{f} \beta <0 \mathrm{a}\mathrm{n}\mathrm{d} p<0.05\end{array}\right\}$$

(3)

2.3.3. Quantitative Estimation Methods of Ecosystem Service Supply and Demand

Water, food and carbon-related ecosystem services are central to human well-being and climate policy. Their supply and demand can also be operationalized using established, spatially explicit equations and routinely available datasets. Thus, these three ecosystem services were selected to represent resource supply functions: crop yield for crop supply, carbon sequestration for carbon supply, and water yield for water supply. The corresponding demand was measured by water consumption, crop consumption, and carbon emissions, respectively. The supply–demand balance (SDB) for water, crop, and carbon in each prefecture-level city within the YREB was calculated as the difference between supply and demand. The calculation is defined as follows:

$$S{D}_k=S_k-D_k$$

(4)

where SD denotes the supply–demand balance of water, crop, and carbon; S represents the corresponding supply; and D indicates the respective demand for each resource.

Defined for city c and service k$\in \left\{\mathrm{water}, \mathrm{crop}, \mathrm{carbon}\right\}$, the threshold-aware, symmetric Ecosystem service supply–demand ratio (ESDR) achieves scale normalization of the supply–demand difference and embeds policy-relevant safety margins:

$$x_{c,k}={\displaystyle \frac{D_{c,k}}{S_{c,k}}}$$

(5)

$$ESD{R}_{c,k}={\displaystyle \frac{1-x_{c,k}}{1+x_{c,k}+\epsilon }}$$

(6)

where $D_{c,k}$ denotes the annual service demand within city c and carbon; S represents the annual service supply produced within city c. $x_{c,k}$ denotes the annual service balance within the city c.

In alignment with risk-management practice, service-specific pressure thresholds are estimated. Warning and red-line levels are set at the 75th and 90th percentiles of $ESD{R}_{c,k}$, computed across all prefecture–year observations in the sentinel years 2000, 2005, 2010, 2015, 2020, and 2023. The ESDR was categorized as:

$$Stat{e}_{c,k}=\left\{\begin{array}{l}\mathrm{S}\mathrm{a}\mathrm{f}\mathrm{e}, ESD{R}_{c,k}\ge {\theta }^{\mathrm{warn}}\\ \mathrm{W}\mathrm{a}\mathrm{r}\mathrm{n}\mathrm{i}\mathrm{n}\mathrm{g}, {\theta }^{\mathrm{red}}\le ESD{R}_{c,k}\le {\theta }^{\mathrm{warn}}\\ \mathrm{R}\mathrm{e}\mathrm{d}, ESD{R}_{c,k}\le {\theta }^{\mathrm{red}}\end{array}\right\}$$

(7)

where ${\theta }^{\mathrm{red}}$ denotes the red-line of $ESD{R}_{c,k}$, and ${\theta }^{\mathrm{warn}}$ represents the warning of $ESD{R}_{c,k}$.

Quantitative Estimation Methods of Crop

Grain production for each prefecture-level city in the YREB was estimated based on NDVI data from the years 2000, 2005, 2010, 2015, 2020, and 2023. The city-level grain production was then downscaled to grid cells within each city in proportion to each cell’s share of the city’s total NDVI [37]. This approach enabled spatial disaggregation of city totals to the grid-level grain supply. The grid-level grain supply is calculated as follows:

$$S_{ci}=\sum _{i=1}^n{\displaystyle \frac{NDV{I}_i}{NDV{I}_{sum}}}\times G_{sum}$$

(8)

In the equation, $S_{Ci}$ denotes the grain production (t) of the i-th prefecture-level city, and $G_{sum}$ represents the total grain output of the entire region. Grain demand was estimated from per capita grain consumption and the year-end resident population of each prefecture-level city. The calculation of grain demand is expressed as follows:

$$D_{Ci}=PO{P}_i\times F_{city}\times Rati{o}_{city}+PO{P}_i\times F_{rural}\times Rati{o}_{rural}$$

(9)

where $D_{Ci}$ represents the total grain consumption (t) of the i-th prefecture-level city, and ${POP}_i$ is the total population of that city. $Rati{o}_{city}$ denotes the proportion of the urban population to the total population, $F_{city}$ is the per capita grain consumption in urban areas (kg), $Rati{o}_{rural}$ represents the proportion of the rural population, and $F_{rural}$ is the per capita grain consumption in rural areas (kg).

Quantitative Estimation Methods of Water Resources

Water supply was estimated using the seasonal water yield InVEST 3.14.2 software, based on annual precipitation and evapotranspiration data for the YREB from 2000, 2005, 2010, 2015, 2020, and 2023 [38]. The calculation of water supply is expressed as follows:

$$S_{wi}=P_i-Q{F}_i-AE{T}_i$$

(10)

where $S_{wi}$ denotes the water supply in grid i, $P_i$ is the annual precipitation, $AE{T}_i$ represents the actual evapotranspiration, and $Q{F}_i$ refers to the runoff estimated by the SCS-CN [39] runoff model for grid i.

Water demand was derived from municipal water resource bulletins across the YREB for the years 2000, 2005, 2010, 2015, 2020, and 2023. The calculation of water demand is defined as follows:

$$D_{wi}=I{g}_i\times C_i+I{n}_i\times W{I}_i+D{W}_i\times 365\times PO{P}_i$$

(11)

where $D_{wi}$ denotes the total water demand in grid or city i. $I{g}_i$ represents the irrigation water use per ha (m³/ha), and $C_i$ is the cropland area (m³). $I{n}_i$ denotes the industrial added value (10,000 CNY), and $W{I}_i$ refers to the water use per unit of industrial output (m³/10,000 CNY). $D{W}_i$ is the average daily domestic water use per capita (L/person/day), and $PO{P}_i$ indicates the total population.

Quantitative Estimation Methods of Carbon Sequestration Balance

Carbon sequestration refers to the uptake of atmospheric CO₂ by vegetation through the process of photosynthesis. In this study, the amount of CO₂ assimilated via photosynthesis is used as an indicator of the carbon sequestration capacity of vegetation. According to the stoichiometry of photosynthesis, 1 g of dry biomass corresponds to the uptake of approximately 1.63 g of CO₂ [35]. The carbon sequestration is calculated as follows:

$$S_{ci}=1.63\times NP{P}_i$$

(12)

The annual mean net primary productivity (NPP) was estimated using the Carnegie–Ames–Stanford Approach (CASA) model [40]:

$$NP{P}_{(i,t)}=FRA{R}_{(i,t)}\ast SO{L}_{(i,t)}\ast 0.5\ast {\epsilon }_{(i,t)}$$

(13)

In the equation, $NP{P}_{(i,t)}$ denotes the net primary productivity of pixel i during time period t; $FRA{R}_{(i,t)}$ is the fraction of photosynthetically active radiation (PAR) absorbed by vegetation; $SO{L}_{(i,t)}$ represents the total solar radiation received by pixel i in period t; and ${\epsilon }_{(i,t)}$ refers to the light use efficiency.

Regarding carbon emissions, fossil fuel combustion is the dominant source. Therefore, this study estimated carbon emissions based on eight major fossil fuels: raw coal, coke, crude oil, gasoline, kerosene, diesel, fuel oil, and natural gas. Carbon emissions for each prefecture-level city in the YREB were calculated using the IPCC inventory method. The formula is as follows:

$$E{C}_i=\sum _{j=1}^8{C}_i=\sum _{j=1}^8{E}_j\times F_j\times {\displaystyle \frac{44}{12}}$$

(14)

where $E{C}_i$ denotes the total carbon emissions from energy consumption, and $C_i$ represents the carbon emissions associated with the j-th type of fossil fuel. $E_j$ is the amount of the j-th fuel consumed, and $F_j$ refers to its corresponding carbon emission factor.

2.3.4. SHAP Stacking-Bayesian Optimization Model

The SHAP (Shapley Additive explanations) framework is a widely adopted tool for interpreting machine learning models by assigning contribution scores to each input feature. Based on cooperative game theory, SHAP uses Shapley values to reflect the marginal contribution of each feature across all possible combinations, enabling both global and local interpretability with theoretical consistency [41,42]. However, the standard SHAP approach assumes feature independence and is highly sensitive to data quality and bias.

To overcome these limitations, a SHAP Stacking–Bayesian optimization Model was developed. This model integrates multiple tree-based algorithms, such as Random Forest, Gradient Boosting, XGBoost, and LightGBM, within a stacking framework optimized via Bayesian search. Compared with a single-model SHAP baseline, this ensemble approach improves simulation accuracy, reduces overfitting, enhances robustness, and maintains strong interpretability [43,44]. Using this model, key anthropogenic drivers of ecosystem service SDB, including GDP of the primary, secondary, and tertiary sectors, total GDP, population density, nighttime light intensity, cultivated land area, and urbanization rate, were quantified across prefecture-level cities in the YREB. Detailed evaluations of model performance are provided in the Supplementary Materials.

2.3.5. Analysis of Ecosystem Service Flows

Ecosystem service (ES) flows describe the spatial and temporal transfer of services from service providing areas (SPAs) to service benefiting areas (SBAs), mediated by service connecting areas (SCAs). To optimize the spatial allocation and supply–demand regulation of key services at the regional scale, this study simulated multiservice flow across the Yangtze River Economic Belt, focusing on water resources, crop production, and carbon sequestration. Based on the supply–demand balance of these services, a breakpoint model was employed to estimate the maximum transmission radius of ES flows [45]:

$$P_{ij\prime }={\displaystyle \frac{P_{ij}}{1+\sqrt{V_j/V_i}}}$$

(15)

where $P_{ij\prime }$ is the effective flow radius from SPA i to SBA j, $P_{ij}$ is the geographic distance, and $V_j$ and $V_i$ are the service supply values of each region. Based on these flows, a proximity-based intercity network was constructed in ArcGIS.

Multiservice flows were optimized using the Non-dominated Sorting Genetic Algorithm II (NSGA-II), a fast and elitist multi-objective evolutionary algorithm [46]. The model simultaneously minimizes total flow distances for three objectives: $F_1\left(x\right)$—total crop service transport distance; $F_2\left(x\right)$—total water resource flow distance; $F_3\left(x\right)$—total carbon service flow distance. This forms a multi-objective minimization problem:

$$\underset{x}{min}F(x)= [F_1(x),F_2(x),F_3(x)]$$

(16)

Instead of a weighted-sum aggregation, the NSGA-II flow allocation embeds explicit supply–demand risk thresholds and yields the Pareto set, thereby revealing trade-offs among service types. The optimization is subject to key constraints:

Flow balance constraint at each node i for resource k, ensuring that net supply/demand is met:

$$\sum _{(i,j)\in E}{x}_{ij}^k-\sum _{(j,i)\in E}{x}_{ji}^k=b_i^k$$

(17)

Capacity constraint for each edge $(i,j)$, limiting the maximum allowable flow:

$$0\le \sum _{(i,j)\in E}{x}_{ij}^k\le Q_{i,k}^{\mathit{exp}ort}$$

(18)

The total inflow of service k into node i must at least meet its threshold-defined minimum import (relief) requirement:

$$\sum _{(i,j)\in E}{x}_{ji}^k\ge H_{i,k}^{need}$$

(19)

where $H_{i,k}^{need}$ denotes the minimum import requirement of city i for service k.

3. Results

3.1. Spatiotemporal Analysis of Ecosystem Services

3.1.1. Temporal Characteristics of Ecosystem Services

Based on the InVEST 3.14.2 software, five key ecosystem services—including water yield, crop production, carbon sequestration, water purification, and soil retention—were quantified for 130 prefecture-level cities across the YREB from 2000 to 2023. Temporal trends were assessed using the Theil–Sen slope estimator and were categorized into five levels: strong increase, weak increase, stable, weak decrease, and strong decrease. Their spatiotemporal dynamics are illustrated in Figure 3 and summarized in

Table 3. Descriptive statistics of ecosystem services of cities in Yangtze River Economic Belt.

Year	$\mathbf{Water} \mathbf{Yield}/10^{11} {\mathbf{m}}^3$	$\mathbf{Soil} \mathbf{Retention}/10^{11} \mathbf{t}$	$\mathbf{Phosphorus} \mathbf{Load}/10^4 \mathbf{t}$	Nitrogen $\mathbf{Load}/10^5 \mathbf{t}$	$\mathbf{Crop} \mathbf{Yield}/10^8 \mathbf{t}$	$\mathbf{Carbon} \mathbf{Sequestration}/10^9 \mathbf{t}$
2000	7.614	1.536	4.521	1.861	2.097	2.761
2005	6.581	1.361	4.512	1.923	2.021	2.177
2010	8.626	1.589	4.446	1.887	2.127	2.181
2015	7.758	1.538	4.426	1.900	2.333	2.464
2020	8.935	1.588	4.454	1.967	2.391	2.412
2023	7.017	1.284	4.424	1.935	2.415	2.517

.

Water yield exhibited a slight upward trend across the YREB over the study period, maintaining a spatial gradient of “high in the east, low in the west.” Most downstream regions remain stable. In contrast, the upstream highland areas of Sichuan and Yunnan, with relatively low rainfall, showed marginal or locally significant declines in water yield. Crop production increased significantly between 2000 and 2023, driven primarily by productivity gains in the fertile agricultural heartlands of the middle and lower reaches, including the Sichuan Basin. Crop production fell marginally between 2000 and 2005, then increased continuously, reaching a new maximum in 2023. Carbon sequestration declined from 2.76 × 10⁹ t in 2000 to 2.18 × 10⁹ t in 2005, then partially recovered to 2.52 × 10⁹ t by 2023, indicating moderate restoration of carbon sequestration capacity. Notably, the middle and lower reaches exhibited marked increases in ecosystem carbon uptake over the past two decades. Total phosphorus (TP) loads showed a slight downward trend and remained stable between 4.4 × 10⁴ and 4.5 × 10⁴ t. In contrast, total nitrogen (TN) loads increased gradually from 1.86 × 10⁵ t in 2000 to 1.94 × 10⁵ t in 2023, suggesting a persistent input pressure likely associated with urban and agricultural sources. Soil retention showed a slight overall decline during the study period, with relative peaks observed in 2010 and 2020, followed by a pronounced drop in 2023 to the lowest level recorded.

3.1.2. Spatial Characteristics of Ecosystem Services

As shown in Figure 4, the spatial distribution of ecosystem services across 130 prefecture-level cities in the YREB from 2000 to 2023 reveals a clear bipolar pattern. Ecosystem service provision is consistently higher in the southwestern mountainous regions, driven by favorable natural conditions such as abundant precipitation and extensive forest cover, which enhance water yield, soil retention, and carbon sequestration. In contrast, the eastern plains, subject to intense land development, exhibit significantly weaker ecosystem functions due to ecological degradation [47].

Water yield is persistently highest in the upstream areas, particularly in Yingtan (mean: 9 × 10⁷ m³/hm²), while cities like Chuxiong and Panzhihua record the lowest values (8 × 10⁶ m³/hm²). Regionally, the upper basin (3.29 × 10¹¹ m³) yields more water than the middle (2.82 × 10¹¹ m³) and lower reaches (1.63 × 10¹¹ m³).

Soil retention is similarly concentrated in the forest-rich upstream (e.g., Nujiang: 3 × 10⁷ t/hm²), but lower in downstream floodplains with high erosion (e.g., Bozhou: 4 × 10⁴ t/hm²) due to sediment redistribution [48,49]. The upper basin’s average (9.69 × 10¹⁰ t) again exceeds that of the middle (3.57 × 10¹⁰ t) and lower (1.55 × 10¹⁰ t) zones.

Total nitrogen (TN) and total phosphorus (TP) loads show a decreasing gradient from upstream to downstream. Nutrient export is concentrated in the upper regions such as the Sichuan Basin due to high agricultural intensity and population density. By contrast, nutrient loads decline downstream owing to reservoir retention and stricter land use policies [50]. Zhoushan exhibits the highest TN and TP intensities (38.27 t/hm², 6.69 t/hm²), while Ganzi reports the lowest (3.55 t/hm², 0.82 t/hm²).

Grain production is the greatest in the upper basin (1.00 × 10⁸ t), followed by the middle (6.1 × 10⁷ t) and the lower (6.2 × 10⁷ t) regions. Ziyang shows the highest yield intensity (3.25 × 10⁴ t/hm²), while Ganzi again ranks the lowest (3.9 × 10² t/hm²).

Carbon sequestration also follows a west-to-east decline: upper (1.36 × 10⁹ t), middle (6.9 × 10⁸ t), and lower reaches (3.5 × 10⁸ t). Xishuangbanna leads in carbon sequestration intensity (2.98 × 10⁵ t/hm²), whereas industrialized cities like Changzhou and Suzhou record the lowest values (6.8 × 10⁴ t/hm² and 6.5 × 10⁴ t/hm²).

3.2. Analysis of Ecosystem Service Supply–Demand Balance

3.2.1. Spatial Assessment of Ecosystem Service Supply–Demand Balance

The YREB’s SDB analysis centered on three spatially transferable services—water yield, crop production, and carbon sequestration. SDB values were quantified for 2000–2023 under the procedures in Section 2.3.4 (Figure 5 and

Table 4. Spatial-temporal dynamics of water, crop, and carbon supply–demand balances in the YREB (2000–2023).

Year	$\mathbf{Water} \mathbf{Resources} \mathbf{Supply}\text{–}\mathbf{Demand} \mathbf{Balance}/10^{11} {\mathbf{m}}^3$	$\mathbf{Crop} \mathbf{Supply}–\mathbf{Demand} \mathbf{Balance}/10^8 \mathbf{t}$	$\mathbf{Carbon} \mathbf{Supply}–\mathbf{Demand} \mathbf{Balance}/10^9 \mathbf{t}$
2000	5.343	1.361	1.514
2005	4.205	1.203	0.066
2010	6.033	1.287	−0.886
2015	5.135	1.446	−1.121
2020	6.452	1.491	−1.409
2023	4.433	1.451	−1.6734

).

Across the YREB, water SDB exhibited a fluctuating downward trajectory, decreasing from 5.343 × 10¹¹ m³ to 4.433 × 10¹¹ m³ during the study period. All subregions—upper, middle, lower—showed net surpluses; the upper reaches, shaped by dense forests and sparse populations, delivered the largest surplus (2.742 × 10¹¹ m³), the middle reaches followed (1.986 × 10¹¹ m³), and the lower reaches registered the smallest value (5.388 × 10¹⁰ m³) under strong urban–industrial demand. At the prefecture scale, Chongqing led the ranking, whereas Shanghai and Jiangsu posted persistent deficits (−7.721 × 10⁸ m³ and –2.236 × 10¹⁰ m³, respectively) [51,52]. Risk grading placed 80 of 130 prefectures in Red, with 25 in Warning and 25 in Safe; the 2023 pattern remained largely unchanged relative to 2000. Theil–Sen diagnostics labeled 122 of 130 as stable, indicating limited trend signals in water ESDR.

In contrast, the carbon SDB deteriorated markedly over the study period. Regional balance fell from 1.514 × 10⁹ t in 2000, reaching −1.673 × 10⁹ t in 2023, providing evidence of a basin-wide shift from sink status toward a net source. The upper reaches remained a persistent carbon sink (6.976 × 10⁸ t), while the middle reaches shifted from a surplus (1.302 × 10⁹ t) to a deficit (−3.811 × 10⁸ t). The lower reaches consistently acted as a carbon source (−1.194 × 10⁹ t) due to higher industrial activity and energy consumption. Pu’er City in Yunnan exhibited the largest surplus (1.034 × 10⁸ t), whereas Shanghai and Jiangsu were major deficit hotspots (−1.525 × 10⁸ t and −6.239 × 10⁸ t, respectively) [53,54]. Spatially, 109 cities were unchanged, 11 improved, and 10 downgraded. Theil–Sen trends indicated 66 cities with a significant decrease and 20 with a slight decrease in the carbon ESDR, suggesting an overall deterioration of carbon supply–demand conditions over 2000–2023.

By comparison, crop SDB remained in surplus, rose from 1.361 × 10⁸ to 1.450 × 10⁸ t over the study period. The upper reaches dominated (6.422 × 10⁷ t), exceeding the middle (3.728 × 10⁷ t) and lower reaches (3.582 × 10⁷ t). Chongqing recorded the highest surplus (7.388 × 10⁶ t), while Shanghai and Zhejiang appeared as principal deficit areas (−1.05 × 10⁶ t and −1.019 × 10⁶ t, respectively). Under rapid population and economic expansion, pronounced ESDR declines emerged in Jiangxi, Shanghai, and the Suzhou–Wuxi–Changzhou urban agglomeration; by contrast, Anhui, one of China’s major grain bases, showed sustained gains.

3.2.2. Drivers of Ecosystem Service Supply–Demand Balance

A Stacking–Bayesian optimization model quantified the effects of eight anthropogenic drivers on the YREB’s SDB of water, crop, and carbon services during 2000–2023, namely urbanization rate (UR), nighttime light intensity (NLI), total and sectoral GDP (PIGDP, SIGDP, TIGDP), cultivated land area (CLA), and population density (PD) (Figure 6 and Figure 7).

For water resources, NLI was the most influential factor throughout the study period, peaking in 2000, declining until 2015, and rising thereafter. This trend aligns with the environmental Kuznets curve and previous research [51], where early-stage decoupling of economic growth and water use was followed by renewed pressure due to economic expansion [55]. Secondary influences such as SIGDP, TIGDP, and urbanization rate showed increasing contributions over time, consistent with rising domestic and service-sector water demands [56].

Similarly, NLI was the dominant driver of the carbon supply–demand balance, reflecting the tight coupling between regional economic activity and emissions. Its temporal pattern also followed a decoupling–recoupling trajectory, with emissions rebounding in later development stages [57].

In contrast, the crop supply–demand balance was primarily determined by cultivated land area. Its SHAP value peaked in 2000 (0.671), declined, and then rebounded after 2015, highlighting the growing constraints from urban expansion and population growth.

Spatially, highly developed eastern cities (e.g., Shanghai, Jiangsu) exhibited consistent deficits in both water and carbon balances due to high NLI and PD, with strongly negative SHAP values. In contrast, ecologically rich upstream regions (e.g., Ganzi, Aba) showed surpluses and positive SHAP values, suggesting a functional role as ecological suppliers [58]. For crops, productive plains (e.g., the Sichuan Basin and the Jianghan Plain) showed large positive SHAP values, while coastal megacities like Shanghai exhibited negative values, underscoring reliance on external grain transfers [59].

3.3. Optimization of Ecosystem Service Supply–Demand Balance

3.3.1. Optimization of Supply–Demand Balance for Individual Ecosystem Services

Water resource supply–demand flow dynamics revealed an increase in total flow volume from 1.1 × 10¹⁰ m³ in 2000 to a peak of 1.57 × 10¹⁰ m³ in 2010, followed by a decline to 1.1 × 10¹⁰ m³ in 2020 due to stricter water governance. However, by 2023, flow volumes rebounded to 1.6 × 10¹⁰ m³ in response to extreme climate conditions such as the severe drought event in the Yangtze River Basin in 2022 [60,61,62]. At the prefecture scale, Shanghai consistently led as the largest net-importer (7.9 × 10⁹ m³ in 2010, accounting for more than 50% of total flows), followed by Nanjing and Wuhan. Xuancheng, Ningbo, and Shaoxing remained major net-exporters, forming a spatial configuration of “peri-urban water supply to metropolitan demand” (Figure 8).

Crop supply–demand flow dynamics showed an accelerated increase in total intercity flows, from 2.7 × 10⁵ tons in 2000 to 4.32 × 10⁶ tons in 2023, particularly between 2000 and 2005. This reflected intensifying urban crop demand due to urbanization and demographic growth, alongside limited farmland availability [63]. Shanghai emerged as the largest crop importer after 2005, with inflows reaching 1.65 × 10⁶ tons by 2023. Guiyang and Hangzhou also recorded high inflows. Suzhou transitioned from a major crop exporter in 2005 (4.4 × 10⁵ t to Shanghai) to a net importer in 2020–2023 (2.6 × 10⁵ t), highlighting evolving urban food supply roles.

Carbon supply–demand flow dynamics demonstrated a sustained upward trend, from 3.3 × 10⁸ t in 2000 to 7.3 × 10⁸ t in 2023. The steepest growth occurred between 2005 and 2010, with subsequent stabilization around 6.8 × 10⁸ t by 2020. In 2000, major inflows were concentrated in lower reaches (e.g., Wuhan: 3.9 × 10⁷ t), whereas by 2023, Chongqing became the largest carbon sequestration importer (8.27 × 10⁷ t), reflecting a westward shift of carbon burdens driven by industrial relocation [49]. Conversely, cities such as Wenshan, Ganzhou, and Lishui consistently exported surplus carbon sequestration services, playing vital roles as ecological compensation zones.

3.3.2. Optimization of the Supply–Demand Balance Across Multiple Ecosystem Services

Water resource supply–demand flows showed a fluctuating trend from 2000 to 2023. The total flow volume increased from 8.5 × 10⁹ m³ in 2000 to a peak of 1.45 × 10¹⁰ m³ in 2010, declined to 8.5 × 10⁹ m³ in 2020, and rebounded to 1.43 × 10¹⁰ m³ in 2023 (

Table 5. The Total flow volume of supply–demand Equilibrium across multiple ecosystem services in the YREB (2000–2023).

Year	$\mathbf{Water} \mathbf{Balance} \mathbf{Flow}/10^{11} {\mathbf{m}}^3$	$\mathbf{Crop} \mathbf{Balance} \mathbf{Flow} /10^8 \mathbf{t}$	$\mathbf{Carbon} \mathbf{Balance} \mathbf{Flow}/10^8 \mathbf{t}$
2000	0.851	0.085	0.957
2005	1.319	0.131	1.526
2010	1.449	0.145	2.441
2015	1.031	0.103	2.367
2020	0.845	0.084	2.785
2023	1.426	0.142	2.891

). This pattern reflected growing urban water demand in the early 2000s and the intensification of intra-basin redistribution, facilitated by large-scale inter-basin transfer projects such as the Yangtze-to-Taihu and Yangtze-to-Hanjiang schemes [64]. After 2010, frequent droughts (e.g., in 2011 and 2022) and water quota regulations reduced flow volumes [65]. The 2023 recovery indicated enhanced allocation capacity. Spatially, cities like Ningbo, Shaoxing, and Xuancheng consistently acted as supply hubs, while Shanghai, Nanjing, and Wuhan remained the largest recipients. In 2010, Shanghai accounted for 36.4% of total inflows, while by 2015, Wuhan’s share rose to 13.3%. From 2020 to 2023, Chuzhou and Shaoxing became key source cities for the Yangtze River Delta.

Crop supply–demand flows expanded substantially, rising from 3.3 × 10⁵ tons in 2000 to 3.93 × 10⁶ tons in 2023, a nearly twelvefold increase. The sharpest rise occurred between 2000 and 2005, driven by urbanization and dietary shifts [66]. The network evolved from local self-sufficiency to regional complementarity. Nantong, Huzhou, and Luzhou emerged as major exporters and Shanghai, Guiyang, and Ningbo as key importers. Average transfer distances lengthened, highlighting growing regional interdependence for food security.

Carbon supply–demand flows more than tripled from 9.576 × 10⁷ tons in 2000 to 2.89 × 10⁸ tons in 2023. The sharpest increase occurred between 2000 and 2010, in line with rapid industrialization and surging energy use [67]. While carbon emissions soared from 5.9 × 10⁸ tons in 2000 to 1.63 × 10⁹ tons in 2010, carbon sequestration grew only 37%, driving a shift toward net carbon deficits [68]. Spatially, carbon flows shifted from central forest zones to western mountainous regions. In 2000, Anhui and Hubei served as key suppliers. After 2005, cities like Lishui, Fuzhou (Jiangxi), and Shiyan became carbon exporters, offsetting emissions in industrial provinces like Hunan and Sichuan. By 2020–2023, Wenshan and Ganzhou further strengthened their sequestration capacity, supporting major deficit areas such as Chongqing and Chengdu. The expanding flow network and rising “compensation distance” reflected the increasing spatial coupling of carbon balance across the basin [49].

4. Discussion

For the sustainable use and conservation of the YREB, it is essential to understand how ecosystem services (ESs) circulate within and between cities. Using a SHAP based modeling framework together with an improved minimum-cost algorithm, we disentangled the anthropogenic drivers of ES supply–demand balance and mapped the ES supply–demand balance flows. Over 2000–2023, the flows of water yield, crop provisioning, and carbon sequestration intensified and became more structured, revealing two concurrent patterns: (i) strengthened regional coordination through increasingly complementary exchanges among subregions, and (ii) an amplified siphon effect, whereby large metropolitan centers concentrated demand and drew services from surrounding cities.

4.1. Regional-Scale Evaluation Framework for Ecosystem Service Supply–Demand Balance

An integrated workflow combining the InVEST 3.14.2 software, grid-based demand estimation and Theil–Sen trend analysis was developed to quantify water yield, crop production and carbon sequestration across 130 prefecture–level cities in the Yangtze River Economic Belt (YREB) from 2000 to 2023. By simultaneously considering the supply provided by ecosystems and the anthropogenic demand for water, crops and carbon sequestration, the framework captures how human activities shape the ecosystem service provision and intensify service demand. The resulting SDB maps revealed a persistent “bipolar” pattern: water SDB exhibited a fluctuating downward trajectory, decreasing from 5.343 × 10¹¹ m³ to 4.433 × 10¹¹ m³, whereas the carbon SDB shifted from surplus (+1.514 × 10⁹ t) to deficit (−1.673 × 10⁹ t) during the study period. The crop SDB rose from 1.361 × 10⁸ to 1.450 × 10⁸ t across the study period. The temporal trends in the supply–demand balance (SDB) of water resources, crop production, and carbon were consistent with previous studies [68]. On that basis, SHAP results revealed the anthropogenic drivers of the YREB’s SDB, the nighttime lights intensity (NLI), a proxy for economic activity, dominated water and carbon SDBs [69], whereas cultivated land area (CLA) was the primary control on crop SDB. Nighttime light intensity (NLI) is a practical proxy for regional economic prosperity and the intensity of urban activity. Rising economic development provides the foundation for coordinating urban growth with ecological protection [70]. Mechanistically, NLI influences the supply–demand balance of water and carbon as follows: for water SDB, higher NLI signals greater domestic or industrial water demand and densification of built-up land; the expansion of impervious surfaces reduces infiltration and baseflow while increasing runoff and pollutant loads, weakening natural water provision and purification capacity [71]. For carbon SDB, higher energy use, transport, and industrial output elevate carbon emissions while urban expansion diminishes vegetative carbon sequestration through habitat loss and fragmentation [72]. By identifying critical mismatches and dominant human pressures such as economic activity and land use intensity, the framework offers valuable insights for targeted ecosystem management and sustainable regional development. These findings provide a scientific basis for balancing ecological protection with socioeconomic growth.

4.2. Optimizing the Spatial Allocation of Ecosystem Service Supply and Demand

The spatial flows of water yield, crop production, and carbon sequestration across the YREB reveal distinct yet interconnected patterns shaped by regional ecological heterogeneity and anthropogenic pressures. Water flow patterns fluctuate in response to hydrological change and urban expansion; crop-related flows expand nearly tenfold, reflecting intensified agricultural output and urban–rural food dependencies. Concurrently, carbon flows increase threefold, yet the regional carbon balance shifts from surplus to deficit, underscoring the ecological cost of continued economic growth and insufficient sequestration capacity. These divergent trajectories highlight the emergence of a nascent yet inefficient ES circulation network across the YREB, characterized by spatial mismatches and fragmented governance. At the city level, the results demonstrate a progressive polarization of ecosystem service (ES) flows, wherein urban agglomerations such as Shanghai, Wuhan, and Nanjing act as persistent demand centers across at least one or two ES domains, heavily reliant on nearby prefectures, such as Xuancheng, Ningbo, and Ganzhou, which provide systematic donor functions (Figure 9). This cross-service concordance explains why local SDB deficits in metropolitan areas cannot be closed by within-boundary measures alone and motivates cross-regional governance with compensation. The results are consistent with the findings reported by Qu (2024) [54].

4.3. Spatial Strategies for Coordinated Regional Sustainability

Distinct spatial scales are associated with characteristic spatial management regimes, and multiscale approaches can diagnose supply–demand disequilibria with intertemporal correlations while furnishing cross scale evidence to better inform diverse socio-ecological governance contexts. At the basin scale, both the water SDB and the crop SDB indicate surpluses; however, pronounced heterogeneity emerges at the prefecture level, where developed eastern cities such as Shanghai, Nanjing, and Suzhou exhibit persistent deficits across multiple ecosystem services and therefore draw increasingly on ecological resources from neighboring jurisdictions and surplus areas. In effect, the ecological resources of these contributing regions are being used without remuneration by wealthier eastern localities. For compensation receiving regions, more granular eco compensation policies are needed to support development. Recommended measures include:

Strengthen and refine the central government fiscal transfer system to implement differentiated compensation, and appropriately recalibrate transfer mechanisms to expand budgetary support for key ecological function areas, including ecological redlines and nature reserves, in order to secure basic expenditure needs and enhance these regions’ capacity to deliver ecosystem services.

Establish provincial dedicated eco compensation funds to advance horizontal eco compensation, finance these funds through provincial budget appropriations, and deploy them exclusively for ecological protection and restoration within compensated regions so as to ensure the continuity and stability of compensation efforts [65,73]. The funds should be supervised by a specialized management authority that, drawing on ecosystem service balance assessments, identifies eligible recipients and sets compensation standards, thereby ensuring rational and equitable allocation [74].

Mobilize eco-compensation finance through multiple channels, including resource allocation payments and green bonds, and develop location specific, diversified investment management models to enhance the overall effectiveness of compensation [75]; at the same time, encourage local governments to cultivate green industries such as ecotourism and organic agriculture so that local surpluses in ecosystem services can be converted into green economic returns, thereby attracting private capital to participate in eco compensation. Improve market-based trading mechanisms for ecosystem services, including carbon emission allowances, pollution discharge permits, and water rights, and establish standardized, transparent, and convenient trading platforms; encourage enterprises, nonprofit organizations, and local governments to give priority, in accordance with market rules, to purchasing development right quotas from recipient regions.

For recipient regions, the most effective pathway to restore local supply–demand balance is to raise resource-use efficiency and curb avoidable consumption through sector-specific measures across carbon, water, and food systems.

On the water side, utilities should prioritize demand-side management by deploying universal smart metering with user feedback, implementing dynamic or inclining-block tariffs to dampen peak use, and delivering targeted customer programs [76]. These measures consistently reduce household and commercial consumption while enabling rapid, low-cost savings. They should be coupled with systematic control of non-revenue water through active leak detection, pressure management, and district metering, which together curb physical losses, reduce abstractions, and enhance local supply reliability [77]. In parallel, scaling fit-for-purpose reuse, for example, tertiary-treated supplies for industrial cooling and landscape irrigation, can expand effective availability without additional potable withdrawals. Where residual scarcity persists, flexible allocation instruments, such as transferable permits or temporary allocation trades, can reallocate water toward higher-value and critical uses during stress periods while safeguarding environmental flows and equity considerations.

In food systems, safeguarding food security where crop shortfalls are worsening requires strict protection and quality upgrading of Permanent Basic Farmland (PBF) together with the construction of high-standard farmland, which empirical studies show curbs non-agricultural conversion and stabilizes output [78]. Within protected areas, precision irrigation such as drip or subsurface drip, agronomic efficiency measures including 4R nutrient stewardship and balanced potassium fertilization, and digitally enabled services using drones and in-field sensors can raise water- and input-use efficiency without expanding cropland [79]. Urban and peri-urban production, for example, vertical farming and other forms of controlled-environment agriculture, provides near-market buffers that shorten supply chains and enhance resilience during disruptions. In parallel, measures to reduce food loss and waste along the supply chain through cold-chain upgrades, improved on-farm and post-harvest storage, better packaging, and consumer education campaigns can eliminate a material share of avoidable demand. Together, these interventions reduce pressure on external inflows, stabilize local supply, and improve the overall supply-demand balance.

On carbon side, industrial decarbonization should proceed in a sequence of reduce, electrify, and abate. First, capture low-regret efficiency gains through process integration, waste-heat recovery, and optimization of motors and steam systems, which deliver immediate reductions in emissions intensity. Next, accelerate electrification and fuel switching supported by clean power, including pilot deployments of green hydrogen in steelmaking, and complement these steps with clinker substitution and novel low-carbon binders in the cement sector. For residual, hard-to-abate process emissions, deploy carbon capture, utilization, and storage while advancing material efficiency and circularity to curb demand for primary energy-intensive materials. Credible procurement standards and product-level markets, for example, for green steel and low-carbon cement, can create demand pull and mobilize private investment.

Taken together, these measures implement an efficiency-first approach to demand management that lowers both the volume and the volatility of imported services, strengthens local resilience, and promotes coordinated multifunctionality across the water, food, and carbon nexus.

4.4. Limitations and Future Research

The methodological framework offers a replicable approach for analyzing inter-city carbon emission linkages and ecosystem service flows in both domestic and international urban cluster. By capturing the spatial-temporal dynamics of ecosystem service supply-demand imbalances and their coupling with anthropogenic activities, the model is particularly suited for application in rapidly urbanizing and resource-constrained regions. Nonetheless, several limitations merit discussion.

First, the spatial resolution adopted 1 km for land use, precipitation, and evapotranspiration datasets was chosen based on data availability and the geographical scope of the study area. While sufficient for regional-scale representation, this may constrain the model’s precision. Future research could benefit from higher-resolution datasets and multi-scale integration to improve accuracy, particularly for fine-grained urban analyses.

Second, this study focuses on optimizing the flows of three critical ecosystem services with strong transfer characteristics. However, other ecosystem services such as biodiversity conservation, climate regulation, and pollutant filtration remain outside the current optimization scope. Expanding the model to include these dimensions would enhance its comprehensiveness and policy relevance in future applications.

5. Conclusions

This study charts the changing supply and demand dynamics of water yield, crop production and carbon sequestration across the Yangtze River Economic Belt (YREB) from 2000 to 2023 by coupling InVEST 3.14.2 software with a SHAP-informed Stacking Bayesian optimization model and a threshold-aware spatial flow optimization based on an enhanced NSGA-II Minicost algorithm. The YREB’s supply–demand balance (SDB) showed significant spatial heterogeneity. Water SDB exhibited a fluctuating downward trajectory, decreasing from 5.343 × 10¹¹ m³ to 4.433 × 10¹¹ m³, whereas carbon SDB shifted from a surplus (+1.514 × 10⁹ t) to a deficit (−1.673 × 10⁹ t) during the study period. Crop SDB rose from 1.361 × 10⁸ to 1.450 × 10⁸ t across the study period. Nighttime light intensity (NLI), a proxy for economic activity, dominates water and carbon SDB, whereas cultivated land area (CLA) is the primary control on crop SDB. The spatial flows of water yield, crop production, and carbon sequestration across the YREB reveal distinct yet interconnected patterns. Water flow patterns fluctuate in response to hydrological change and urban expansion; crop flows expand nearly tenfold. Concurrently, carbon flows increase threefold, yet the regional carbon balance shifts from a surplus to a deficit. At the city level, the results demonstrate a progressive polarization of ecosystem service flows, wherein urban agglomerations act as persistent demand centers across at least one or two ES domains, heavily reliant on nearby prefectures.