Detecting and Predicting the Multiscale Geographical and Endogenous Relationship in Regional Economic–Ecological Imbalances

Abstract

Addressing the economic–ecological imbalance within urban agglomeration integration and sustainable development is crucial, particularly in the context of achieving the Sustainable Development Goals of sustainable cities and communities. This study examines this imbalance using a unique ecosystem services (ESs) balance index that evaluates “supply” and “demand” tradeoffs. It emphasizes localization, mobility, and cooperation as regionalization strategies, utilizing multisource datasets. To address gaps from endogeneity and heterogeneity, the study regresses these strategies on ESs balance values, incorporating landscape patterns as endogenous variables across 214 YRDCA counties or districts in 2020, using a multilevel geographically weighted instrumental variable regression model. Employing the patch-generating land use simulation method, three scenarios were explored: non-intervened development (ND), mobility priority (MD), and localization priority (LP). These scenarios were assessed for their 2025 mitigation effects and health benefits to optimize balanced development strategies. Key findings include (1) a severe ecological–economic imbalance in supply and demand patterns; (2) localization boosts economic development, mobility enhances ecological development, and cooperation promotes both; and (3) LP and MP strategies, compared to ND, show promising potential to reduce the imbalance and generate health benefits, although the extent of the impact may depend on the implementation scale and regional context. By promoting inclusive urbanization and participatory and integrated planning, and enhancing urban resilience through targeted risk-reduction strategies, this study provides insights into fostering balanced development among cities.

1. Introduction

Regional imbalances have become a substantial concern because of their impact on sustainable development, political cohesion, and social equality [1]. While researchers have predominantly focused on regional inequality, particularly in capital, markets, and the workforce [2,3,4], the importance of the regional imbalance between economic and ecological subsystems is evident, especially in the context of urban integration and the global consensus on the SDG 11 Sustainable Cities and Communities [5,6], which underscored the urgency of fostering a harmonious relationship between economic growth and environmental sustainability.

Although existing studies have laid the groundwork for understanding the relationship between ecological and economic balances, a considerable knowledge gap remains in comprehensively examining ecological–economic imbalances as a whole and in investigating economic disparities from an ecological standpoint [2]. Some research addresses ecological and economic imbalances in isolation [7,8], whereas other studies are primarily concerned with economic inequality [9,10]. However, studies that approach these imbalances in an integrated framework are scarce, limiting our ability to design holistic policies for sustainable regional development.

Moreover, there is still a lack of in-depth analysis characterizing the trade-offs between ecological sustainability and economic development [11]. While the importance of identifying and leveraging synergy between these two systems has been widely acknowledged [4], few studies have attempted to quantitatively assess the trade-offs and spatial heterogeneity of these imbalances [12]. Therefore, examining the balance between ecology and economy through an integrated lens is not only timely but necessary. This study seeks to address this gap by using a novel ecosystem services (ESs) balance index to evaluate how regional development strategies can align with broader sustainability objectives.

The tradeoff between supply and demand for ecosystem services reveals regional imbalances from ecological and economic perspectives that have promoted regional sustainable development. Ecosystem services (ESs) supply contributes to both tangible physical products and intangible services, enhancing human welfare through service flows [13]. Simultaneously, the demand for ESs influences societal systems. Surpluses or deficits in ESs supply or demand have been proven to be detrimental [14,15,16]. In an urban environment, the ESs supply reflects the level of ecological functions that urban systems provide to humans. Conversely, the ESs demand illustrates the economic benefits that the system would bring to the urban population [17]. Therefore, achieving a balance between ESs supply and demand is not only crucial for the ecosystem but also serves as a prompt for achieving regional equilibrium between ecology and the economy. We use the ecosystem services (ESs) balance index to evaluate and address the equilibrium between ecological sustainability and economic development at the regional scale. However, some research tends to focus more on analyzing ESs supply than on demand [18], neglecting the importance of the latter. Alternatively, some studies focus on interpreting the index of the supply and demand, failing to identify regional indicators influencing their balance [19]. To address the need for comprehensive and balanced coordination of the economy and ecology, one prevailing solution is to calibrate the ESs balance through a supply and demand matrix based on land use and land cover (LULC) [19]. This approach considers ESs supply and demand and the balance index to be representative of regional ecology and economy as well as their inherent trade-offs.

To better understand the relationship between ecosystem services (ESs) supply–demand balance and regional development, this study adopts regionalization strategies, a conceptual framework that encompasses localization, mobility, and cooperation as its core dimensions [20]. These three strategies collectively capture the spatial, functional, and institutional mechanisms through which regions pursue integrated ecological and economic goals [21,22,23,24]. Localization reflects the spatial concentration of economic activities, such as industries and labor, and typically influences ESs demand due to intensified land use and resource consumption in specific areas [25,26]. Mobility, on the other hand, represents the physical flow of people, goods, and services, and can help redistribute ESs supply, thereby reducing ecological pressure in densely developed or ecologically sensitive zones [22,27]. Cooperation focuses on the institutional and governance dimension, referring to interjurisdictional partnerships, policy coordination, and shared planning efforts that aim to harmonize regional functions and facilitate balanced supply and demand outcomes [28]. Together, localization, mobility, and cooperation provide a comprehensive lens through which regionalization strategies can be operationalized and empirically tested. Their interactions are particularly relevant in complex urbanized regions where economic growth and ecological preservation must be balanced. However, despite their theoretical relevance, prior research tends to emphasize localization while overlooking the quantitative impact of mobility and cooperation on ESs balance [20,25,27]. This study addresses that gap by analyzing all three strategies in an integrated framework, offering a more holistic understanding of how different dimensions of regionalization influence the ecological–economic equilibrium.

Numerous studies have employed diverse metrics to define the concepts of localization, mobility, and cooperation. Indicators of localization include employment wages, land prices, development zones, urbanization rates, and population size [29,30]. Mobility indicators have been measured using transportation prices [31,32], travel time estimation [33], service levels [34], transportation infrastructure systems [35], and urban railways [36]. Cooperation indicators involve exchanges of information and culture, the streamlining and pooling of infrastructure services, economic integration, political association, and integration [28]. Spontaneous economic ties or contacts generated by market sectors such as industrial cooperation and international investment have led to intercity collaboration in various areas [37]. A substantial portion of the data sources rely on county statistical yearbooks, which can deviate markedly from the current situation. Moreover, these studies do not consider urban real-time activity levels, the true extent of economic development, or accurate transportation conditions due to inherent data limitations. This study introduces a more comprehensive approach using multisource big data, such as (1) Weibo check-in location, population density, built-up area, and nighttime light intensity to reflect localizations; (2) road junction density, highway and railway length, and high-speed train frequency to reflect mobility; and (3) indicators of point of interest (POIs) of Apple stores, KFC stores, water and electricity data, logistics companies, and bike renting companies to symbolize cooperation.

Moreover, this study recognizes the changes in spatial heterogeneity due to the impact of different regionalization strategies on the supply–demand imbalance, thus emphasizing the need to explore these relationships at the local scale [2,38]. For example, localization was found to have a geographical impact on ESs supply [39], while cooperation affected supply–demand imbalances unevenly across different spatial areas [40]. Furthermore, increasing evidence indicates that landscape patterns exert an endogenous influence on supply–demand imbalances [41,42,43]. Initially, regionalization was closely linked to urban land use, resulting in observable shifts in landscape patterns [44,45,46]. Subsequent alterations in landscape patterns have implications for the economic–ecological equilibria of city agglomerations [47,48,49]. Therefore, the pivotal role of landscape patterns in elucidating the impact of regionalization on supply–demand imbalances is evident. However, prior research has overlooked the issue of endogeneity. This endogeneity, along with the spatial heterogeneity, necessitates further exploration of a geographically weighted regression model. This study incorporates a weighted instrumental variable regression (MGWIVR) model that addresses endogeneity, considering the endogenous effect of evolving landscape patterns on regional economic–ecological imbalance.

Predicting the long-term impacts of localization, mobility, and cooperation on the supply–demand imbalance is crucial [50,51]. While prior studies have predominantly focused on three land conversion scenarios, economic, ecological, and natural development [52,53], simulations under regionalization strategy scenarios have been lacking [54,55]. Therefore, this study employed a patch-generating land use simulation (PLUS) approach based on scenarios that prioritized localization, mobility, and non-intervention development. This research aims to fill this knowledge gap by conducting simulations under regionalization strategy scenarios.

To bridge the research gaps, this study assesses the status of the supply–demand imbalance using the ES balance index, explores the impact of regionalization strategies, and predicts the future association between these strategies and the supply–demand imbalance. The significance of this study lies in providing both theoretical and practical knowledge for the sustainable and integrated development of YRDCA. First, uncover the intercity inequality phenomenon considering both economic and ecological development in YRDCA and reveal its spatial autocorrelation at the intercounty level. Second, note that localization strategies positively affect the demand for ESs, whereas mobility strategies positively influence the supply of ESs; cooperation strategies work on both supply and demand. This study has proved the directional relationship between regionalization strategies and inequality and the endogenous effect of landscape pattern. Last, this study put forward the MGWIVR-PLUS method to apply the three ND, MD, LP scenarios to Zhejiang, Shanghai, Jiangsu, Anhui, and four subareas in YRDCA and verified that the regional imbalance condition in 2020 is mitigated in 2025 in general, which proves the viability and aid in the creation of pertinent policies. These contributions include utilizing multisource big data, addressing heterogeneity and endogeneity, and providing valuable insights into medium- and long-term regional planning to promote sustainable development goals.

2. Study Area, Data, and Methods

2.1. Study Area

Yangtze River Delta City Cluster (YRDCA) is China’s most developed urban agglomeration, making a substantial contribution to the country’s gross domestic product. It encompasses 26 cities, including 198 counties across the Zhejiang, Jiangsu, and Anhui provinces, along with the municipality of Shanghai (16 districts) (Figure 1). The YRDCA plays a leading role in China’s sustainable and integrated development, as evidenced by key policies like the 2019 “Outline” and the 2021 “Ecological and Environmental Co-protection Plan”, which set goals for economic, ecological, and comprehensive integration. Despite these initiatives, regional imbalances persist—such as Shanghai’s limited spillover effects, Anhui’s weak integration, and inadequate interprovincial transit links. The “Plan” also highlights issues like high development intensity, ecological degradation, and mismatches between environmental quality and economic growth. Addressing these challenges by 2025 and 2035 is essential for balanced, long-term regional integration. Recognizing the urgency of providing enduring political solutions for integrated development, lessons derived from the YRDCA experience have become increasingly pertinent. Streamlining policies for sustained economic, ecological, and overall integration are crucial for the long-term prosperity of urban regions worldwide.

2.2. Data Sources and Processing

The land use and land cover change (LULCC) data utilized in this study were retrieved from Landsat TM/ETM imagery, interpreted by the Data Center for Resources and Environmental Sciences at the Chinese Academy of Sciences (source: http://www.resdc.cn, accessed on 1 January 2024). The original land use classification comprised seven primary categories (cropland, forest, shrub, grassland, water body, impervious land, and barren land) at a spatial resolution of 30 × 30 m. Detailed land use data for 2015 and 2020 from the YRDCA are provided in Supplementary Figure S1.

The Weibo check-in location data are accessible through the dynamic location service interface of the Sina Weibo platform (https://api.weibo.com/2/place/nearby/photos.json, accessed on 1 January 2024). In 2020, the total number of Weibo check-in location data exceeded 6.8 million, predominantly concentrated in the southeast of YRDCA.

POIs data collected from Amap encompass various urban functions categorized into 23 genres and 267 subtypes. The total number of POIs in the YRDCA for the year 2020 amounted to 13 million. POIs data was selected using keywords including “Apple stores, KFC stores, water and electricity companies, logistics companies, and bike renting companies”. Subsequently, incorrect items with different meanings, such as “fruit shop” items in the “Apple stores”, were removed.

Remote sensing and spatially gridded datasets were used in this study. Population density and built-up area were raster data provided by the ESA World Cover (https://esa-worldcover.org/en, accessed on January 2024), and nighttime light data were extracted from (http://www.resdc.cn/DOI/DOI.aspx?DOIid=32, accessed on 1 January 2024). Population density data were extracted from 1 × 1 km² gridded population data for 2020. The built-up area data utilized land use data with 300 m × 300 m² remote sensing images. Nighttime light data with a 500 × 500 m² spatial resolution were obtained from the NPP-VIIRS for 2020.

Road junction data were obtained from OpenStreetMap (source: https://download.geofabrik.de/, accessed on), and highway and railway length datasets are from https://www.gscloud.cn, accessed on 1 January 2024. Train frequency data were collected from the official mobile ticketing application (source: www.12306.cn accessed on 1 January 2024) for the YRDCA area in 2020.

2.2.1. Selected Indicators for Evaluating Regionalization Strategies: Localization, Mobility, and Cooperation

Localization indicators underscore the integration of local resources and self-development, crucial factors closely linked to construction land occupation and the demand for ESs [19]. To gauge localization, we employed indicators, including Weibo check-in density (user ages from 18 to 35), density of water and electricity infrastructure, nighttime light (NTL) intensity, population density, and built-up areas [56]. The data were collected in 2020. The density of Weibo is derived from user location records within the Sina Weibo application, representing self-organization and activity within the local social network [57]. Population density data, sourced from raster data with a resolution of 1 × 1 km, are based on national population statistics and reflect their direct impact on local activity intensity in the YRDCA. Built-up areas refer to physical environments that support local activities, whereas NTL serves as an approximation of local economic activity.

In the context of mobility indicators, which play a crucial role in enabling ecological restoration and reconstruction in highly developed areas and benefiting the supply of ESs, transportation indicators were selected based on previous studies [58,59]. These indicators include road junctions, railway lengths, highway lengths, and high-speed train frequencies [60]. Road junction density, extracted from Amap and involving two steps of data processing according to Long and Liu (2017) [61], reflects intercity connectivity. Railway and highway lengths serve as indicators of transportation accessibility, whereas the train frequency signifies the level of connectivity between counties.

Cooperation indicators measure regional cooperation in terms of market ties, focusing on logistics companies, Apple stores, bike rentals, and KFC stores. A higher density of POIs in a spatial distribution indicates a higher level of cooperation. The cooperation index is expected to influence the ES supply and demand in both directions, thus promoting a balanced outcome [62]. All potential independent variables were computed for each county, and the data sources are summarized in

Table 1. Data sources of independent variables classification.

Independent Variables Classify	MGWIVR Model (2020)	Year	Data Source
Localization	Weibo check-in density	2015, 2020	https://api.weibo.com/2/place/nearby/photos.json (1 January 2024)
	Nighttime light	2015, 2020	http://www.resdc.cn (1 January 2024)
	Population density	2015, 2020	http://www.resdc.cn/DOI/DOI.aspx?DOIid=32 (1 January 2024)
	Built-up area	2015, 2020	WorldCover (https://esa-worldcover.org/en, 1 January 2024)
Mobility	Density of junction of road	2015, 2020	OpenStreetMap (https://download.geofabrik.de/, 1 January 2024)
	Train frequency	2015, 2020	www.12306.cn (1 January 2024)
	Railway/highway length	2020	OpenStreetMap (https://download.geofabrik.de/, 1 January 2024)
Cooperation	POIs of Apple stores, KFC stores, water and electricity companies, logistics companies, and bike renting companies	2015, 2020	Amap electronic navigation map

.

2.2.2. Endogenous Variable: Landscape Patterns

Landscape pattern indices, reflecting the configuration and structural composition of land use, serve as potential endogenous variables in the relationship between regionalization strategies and supply–demand inequality.

Table 2. Explanation of landscape pattern metrics.

Pattern Metrics	Definition	Explanation
Patch density (PD)	$\mathrm{P}\mathrm{D}=N_I/{lA}_I$	$N_i$ is the number of patches; ${LA}_i$ is the total area of landscape $i$.
Landscape shape index (LSI)	$\mathrm{L}\mathrm{S}\mathrm{I}={\displaystyle \frac{0.25{\sum }_{k=1}^m{e}^{ik}}{\sqrt{LA}}}$	$e^{ik}$ is the total length of edge in the landscape between ith and kth patches. ${LA}_i$ is the total area of landscape $i$.
Patch cohesion index (COHESION)	$\mathrm{C}\mathrm{O}\mathrm{H}\mathrm{E}\mathrm{S}\mathrm{I}\mathrm{O}\mathrm{N}=\left[1-{\displaystyle \frac{{\sum }_{i=1}^m{\sum }_{j=1}^m{p}^{ij}}{\sum _{i=1}^m\sum _{j=1}^n{p}^{ij\sqrt{a^{ij}}}}}\right]\times {\left[1-{\displaystyle \frac{1}{\sqrt{Z}}}\right]}^{-1}\times 100$	$p^{ij}$ is perimeter of patch $i$ which is the number of cell surfaces. $a^{ij}$ is area of patch $ij$; Z indicates number of cells in landscape, from 0 to 100.
Perimeter–area fractal dimension (PAFRAC)	$\mathrm{P}\mathrm{A}\mathrm{F}\mathrm{R}\mathrm{A}\mathrm{C}={\displaystyle \frac{2}{\frac{\left[n_i\sum _{j=1}^n\left(\ln p_{ij}-\ln a_{ij}\right)\right]-\left[\left(\sum _{i=1}^n{p}_i\right)\left(\sum _{i=1}^n{a}_i\right)\right]}{\left(n_i\sum _{i=1}^n\ln p_{ij}^2\right)-\left(\sum _{i=1}^n\ln p_{ij}\right)}}}$	$a_{ij}$ is the area of patch $ij$, $p_{ij}$ is the perimeter of patch $ij$, and $n_i$ indicates patch number.

is equally adept at explaining the ES balance. The calculations for these indices in the 214 counties for 2020 were executed using Fragstats 4.2 [63]. All the descriptive statistics values of variables are listed in

Table 3. Descriptive statistics of county level Z-score values of variables.

Variable	Min	Max	Quartile
			25th	50th	75th
Weibo location Density	−0.339	11.651	−0.308	−0.248	−0.041
NTL	−1.084	8.533	−0.631	−0.227	0.314
Water and electricity infrastructure	−1.469	5.363	−0.711	−0.189	0.484
Population	−1.520	9.074	−0.623	−0.149	0.477
Build up area	−1.076	4.953	−0.665	−0.298	0.266
Density of junction of road	−1.302	8.194	−0.648	−0.159	0.469
Frequency of the high-speed train	−0.496	5.851	−0.496	−0.463	−0.015
Railway length	−0.837	5.173	−0.661	−0.312	0.287
Highway length	−1.318	7.276	−0.704	−0.179	0.444
poi_apple	−0.818	9.389	−0.596	−0.300	0.291
poi_kfc	−0.729	7.847	−0.555	−0.382	0.224
poi_logistics	−0.845	7.563	−0.561	−0.343	0.188
poi_bike	−0.690	5.127	−0.669	−0.396	0.315
PD	−2.569	2.620	−0.640	0.039	0.716
LSI	−2.379	3.015	−0.681	0.024	0.585
PAFRAC	−6.171	1.942	−0.296	0.092	0.532
COHESION20	−5.905	0.976	−0.372	0.304	0.692

. This table presents the descriptive statistics of Z-score standardized values for various county-level variables used in the study. After applying standardization, the data for each variable have been transformed to a common scale, which allows for comparability across different units and magnitudes.

2.3. Methods

2.3.1. Trade-Off Between Economy and Ecology: ESs Supply, Demand, and Balance Index

The ESs index calculation was originally proposed by Burkhard et al. (2012) [19]; then, the way to rescore each ESs value to fit for Chinese land use condition was proposed by Chen et al. (2020) [11]. We employ the ESs balance index as the dependent variable. We utilized the ecosystem services matrix method to calculate the ES balance index for the 214 counties or districts within the YRDCA based on LULCC data from 2015 and 2020 at a 30 m resolution. Within each county, we assigned scores ranging from 0 to 5 credits to each of the 23 ecosystem services, resulting in an ecosystem services score matrix for subsequent calculations (Supplementary Figure S2).

Subsequently, we constructed three matrices representing the supply, demand, and balance of ecosystem services in 214 counties. The balance matrix was derived by subtracting the demand matrix from the supply matrix following the methodology outlined by Wu et al. (2019) and Zhang et al. (2017) [62,64]. Three indexes can be calculated based on the matrixes respectively: the ecosystem services supply index (ESSI), demand index (ESDI), and balance index (ESBI). The formulae are as follows:

$${ESSI}_t=\sum _{j=1}^m\sum _{i=1}^n({LUA}_{i,t}\times {SM}_{ij,t})/\sum _{i=1}^n{LUA}_{i,t}$$

(1)

$${ESDI}_t=\sum _{j=1}^m\sum _{i=1}^n({LUA}_{i,t}\times {DM}_{ij,t})/\sum _{i=1}^n{LUA}_{i,t}$$

(2)

$${ESBI}_t=\sum _{j=1}^m\sum _{i=1}^n({LUA}_{i,t}\times {BM}_{ij,t})/\sum _{i=1}^n{LUA}_{i,t}$$

(3)

In these formulas, ${SM}_{ij,t}$, ${DM}_{ij,t}$, and ${BM}_{ij,t}$ refer to the supply, demand, and balance matrices of the ith land use type and jth ecosystem service at time $t$. ${LUA}_{i,t}$ is the area of the ith land type at time $t$. The variables m and n represent the ecosystem services categories and number of land use types, respectively ($m$,$n$ = 1, 2, 3…).

In this study, higher ESSI values within specific areas indicate a stronger emphasis on ecological development than on economic development. In contrast, higher ESDI values signify a greater emphasis on economic development than on ecological concerns. Because the ESBI value is derived by subtracting ESDI from ESSI, a value closer to zero implies a more balanced condition reflecting an economy–ecology imbalance in terms of ESs. The ESs balance index was built after the above steps, which is shown in the below graph (Figure 2).

2.3.2. Identify Imbalance of ESs Surplus or Deficit by BiLISA

Bivariate local indicators of spatial association (BiLISA) illustrate the spatial autocorrelation between one attribute value of a spatial grid and another attribute value of an adjacent grid [65]. In this study, the spatial autocorrelation of ESs supply and demand was calibrated in adjacent YRDCA counties. The following equation is used to identify the bivariate spatial agglomeration of ESs supply and demand:

$${LISA}_i={\displaystyle \frac{(x_i-\overline{x})}{n{\sum }_i{(x_i-\overline{x})}^2}}\sum _j{w}_{ij}(x_i-\overline{x})$$

(4)

$w_{ij}$ is the spatial weight matrix between grids $i$ and $j$,$x_i$ is grid $i$’s attribute value while $\overline{x}$ is the average of all the attributes, and $n$ is the number of grids in the area. Therefore, ${LISA}_i>0$ represents a positive autocorrelation between the two variables, whereas ${LISA}_i<0$ indicates a negative relationship.

Five ESs supply–demand spatial autocorrelation patterns can be observed: “not significant”, “high–high”, “high–low”, “low–high”, and “low–low”. By combining these patterns with the mapping of ESs balance index (ESBI) values, we can identify the critical areas of unbalancing: either ESs surplus (ESBI > 0, with “H–L” clusters) or ESs deficit (ESBI < 0, with “L–H” clusters) (

Table 4. Identify critical areas of ESs surplus or deficit.

First Class	Second Class
	H–H	H–L	L–H	L–L
ESBI > 0	Mild surplus	Surplus	Potential deficit	-
ESBI < 0	-	Potential surplus	Deficit	Mild deficit

) [66]. Clusters are formed when the supply index in one county is correlated with the neighboring effect of the demand index.

2.3.3. Explore the MultiScale Geographical and Endogenous Relationship by MGWIVR

To examine the multiscale geographical and endogenous relationships between regionalization strategies, landscape patterns, and ESBI, this study employs the multiscale geographically MGWIVR model [67], which is an extension of the multiscale GWR (MGWR) and adds a global two-stage least squares (2SLS) method. The model can simultaneously determine the instrumental variable and its multiscale geographical effects at the same time.

The first step is the global estimation phase, using the 2SLS model. It examines global endogeneity and strong or weak instrumental effects using STATA (version of 17)’s ivreg2 and weakiv commands. The 2SLS approach corrects for endogeneity by predicting the values of endogenous variables using valid instruments in the first stage and then uses these predicted values in the second-stage regression.

In the second step, which is also MGWIVR Stage 1, the model transitions from a global to a spatially explicit analysis. Here, geographically weighted regressions are performed using each endogenous variable as the dependent variable, regressed on a set of control variables that includes the instrumental variables identified earlier. This step allows the model to capture how the determinants of the endogenous variables vary across space.

The final step, MGWIVR Stage 2, focuses directly on the ecosystem service balance index (ESBI) as the dependent variable. In this stage, ESBI is regressed on the same set of control variables used previously, excluding the instrumental variables. By using the insights gained from the previous stages, this regression captures the spatially varying effects of different socio-environmental drivers on ESBI. To achieve this, we employed the MGWR software 3.8, running it twice for the two stages of the MGWIVR model. The two stages of the MGWIVR are as follows.

$$\mathrm{p}=\sum _q{\omega }_{bwq}\left(u_i,v_i\right)w_{q,i}+\sum _k{\beta }_{bwq}\left(u_i,v_i\right)x_{k,i}+{ϵ}_i$$

(5)

in which $w_{q,i}$ is the $q^{th}$ excluded instrument at location $i$, while ${\omega }_{bwq}\left(u_i,v_i\right)$ are the locally varying coefficients of the excluded instruments, representing the bandwidth utilized for calculating the $q^{th}$ conditional relationship.

$$\begin{gathered} \mathrm{y}={\eta }_{bw}\left(u_i,v_i\right)\widehat{p_i}+\sum _k{\beta }_{bwq}\left(u_i,v_i\right)x_{k,i}+ϵi \\ \widehat{p_i}=\sum _q{\widehat{\omega }}_{bwq}\left(u_i,v_i\right)w_{q,i}+\sum _k{\widehat{\beta }}_{bwq}\left(u_i,v_i\right)x_{k,i} \end{gathered}$$

(6)

This study follows the two stages of MGWIVR: First, the potential endogenous variable-landscape patterns are regressed on all the included and excluded regionalization factors; second, ESBI is regressed on the predicted value of the landscape pattern index ($\widehat{p}$) and all the regionalization factors from the first stage, and then, the critical areas sensitive to regionalization factors are identified.

2.3.4. Local Scenario-Based Simulation for ESBI Change in 2025 by MGWIVR-PLUS

In the integrated MGWIVR-PLUS model, global land use scenarios traditionally employed in the PLUS framework are replaced with spatially differentiated local scenarios. Specifically, subareas identified by the MGWIVR model as having negative regression coefficients—indicating a need to enhance ecosystem service balance index (ESBI)—are assigned a mobility priority (MP) scenario. Conversely, subareas with positive coefficients—suggesting a need to reduce ESBI—are allocated a localization priority (LP) scenario. This localized approach enables the simulation to better reflect regional priorities and ecological sensitivities. This was achieved by using python 3.8 to modify the PLUS model. The above MGWIVR-PLUS method is further explained in the flow chart below (Figure 3).

The operation of the integrated model consists of two parts: a land expansion analysis strategy (LEAS) and a CA model using multi-type random seeds (CARS). LEAS defines the probability of each land use type change using a random forest (RF) [68]. The overall probability overall $\mathrm{O}{P}_{i,k}^{d=1,t}$ of land use type k is as follows:

$$\mathrm{O}{P}_{i,k}^{d=1,t}=\left\{\begin{array}{c}{P}_{i,k}^{d=1}\times \left(r\times u_k\right)\times D_k^t, If {\mathsf{\Omega }}_{i,k}^t=0 and r<P_{i,k}^{d=1}\\ P_{i,k}^{d=1}\times {\mathsf{\Omega }}_{i,k}^t\times D_k^t, all others \end{array}\right.$$

(7)

where $P_{i,k}^{d=1}$ is the probability of type $k$ of land use developed at grid $i$; $D_k^t$ indicates the self -adaptive inertia coefficient of land type $k$. ${\mathsf{\Omega }}_{i,k}^t$ stands for the neighboring effect of grid $i$.

To form a probability map in LEAS, we used the localization, mobility, and cooperation indicators in the MGWIVR as the driving forces. We transferred the county-level vectorial data into a raster form (500 m × 500 m) (see Supplementary Figure S3) and calculated the probability of land use change based on the contribution of these variables and the input of land use maps for 2015 and 2020.

Subsequently, for the CARS to perform a Markov CA process, we present the transition table for three specific scenarios (see Supplementary Figure S4) according to the local efficiencies from the MGWIVR model. In detail, we maintain localization and mobility as the two scenarios in PLUS, while excluding the cooperation strategy. This decision was based on the observation that localization and mobility had a significant influence on the local balance of ESs in the MGWIVR model. The ND scenario was set as the third scenario for comparison. Because the impact of these strategies on ESs was integrated into the land use change process, the PLUS scenarios serve as a representation of the strategies in the future. The rules of the transitional table were that the MP scenario eliminated transfers from grassland and forest land to cropland and impervious land, whereas the LP scenario encouraged such transitions to impervious land and cropland.

After presetting the ND, MP, and LP scenarios, we implemented them locally in the YRDCA based on the MGWIVR’s geographical results to improve the supply–demand imbalance with a specific target. Finally, we obtained the simulated land use change results for 2025 and calibrated the ESBI for 2025 based on the LULCC map. The critical areas for ESBI changes under different pre-set scenarios serve as the basis for evaluating the viability of regionalization strategies. The total steps of the research methods are shown in the flow chart (Figure 4).

3. Results

3.1. Spatial Characteristics of Economy–Ecology Imbalance Measured by ESBI

The spatial distribution of the supply–demand imbalance, measured by three ESs indices (ESSI, ESDI, and ESBI) in 2020, is illustrated in Figure 5a–c. Additionally, Figure 5d,e depict the spatial clusters representing the autocorrelation effect between ES supply and demand and their significance, respectively. The deficit and surplus areas of ESBI were identified by overlapping ESBI with the spatial clusters shown in Figure 5f.

The results reveal a severe imbalance in the YRDCA, characterized by a disparity in supply and demand growth within the region. In detail, higher values of ESSI are concentrated in southern Anhui and southwestern Zhejiang, ranging from 46.4 to 67.7 (Figure 5a). The ESDI prevails in the metropolitan areas of Shanghai and the Suzhou–Wuxi–Changzhou (SZ–WX–CZ) urban agglomeration, with values ranging from 51.5 to 66.4 (Figure 5b). In the case of ESBI, a value approaching 0 indicates a better balance, but the result diverges in the aforementioned regions, varying from −52.2 to 62.8, illustrating severe imbalance in YRDCA (Figure 5c). Broadly, the northeast region leans towards economic development, whereas the southwest region exhibits an ecological bias. This suggests that the regional clusters in southern Anhui and southwestern Zhejiang prioritize ecological protection, whereas the metropolitan areas of Shanghai and Jiangsu prioritize economic development.

Spatial clusters of autocorrelation effects between ES supply and demand in the YRDCA were revealed using BiLiSA, as shown in Figure 5d. The prevailing clusters include high–low and low–high clusters, indicating a correlation between high supply values and neighboring low demand values, and vice versa. According to Figure 5e, autocorrelations in cities such as Suzhou, Wuxi, Nantong, Huzhou, and Jinhua are particularly significant (p-value < 0.001), promoting spatial heterogeneity and accentuating clustering imbalances between supply and demand in two prominent urban agglomerations: a low–high supply–demand cluster represented by Shanghai City and a high–low supply–demand cluster implied by cities in the Anhui and Zhejiang provinces. This illustrates how high ecological value and low economic demand, or low ecological value and high economic demand, occur simultaneously in most areas, which can exacerbate economic hardship or ecological depression in some areas.

By overlapping the aforementioned results, the ESs deficit and surplus areas of the ESBI were determined (Figure 5f). The deficit areas include the cities of Shanghai, Suzhou–Wuxi–Changzhou, Nantong, and Zhenjiang in Jiangsu Province, which are also the most developed city agglomerations with an economic boom in the YRDCA. The surplus area contains the cities of Anqing and Chizhou in Anhui Province and Taizhou and Jinhua in Zhejiang Province, indicating that ecologically rich areas are also clustered. This polarized clustering pattern further shows the unevenness in economic and ecological growth across different regions.

3.2. The Endogenous and Heterogeneous Impact of Regionalization and Landscape Pattern Indicators on Supply–Demand by MGWIVR Estimates

3.2.1. Global 2SLS Model

Table 5. Second stage of global 2SLS estimates.

Outcome Variable	ESBI of 2020
Constant	−0.000 (−0.000)
Localization
Weibo location density	−0.207 (−3.795) ***
NTL intensity	−0.076 (−0.489)
Population	−0.131 (−1.215)
Build up area	−0.244 (−2.595) ***
Mobility
Road junctions	0.393 (2.530) **
High-speed train frequency	−0.020 (−0.366)
Highway length	0.161 (1.859) *
Railway length	0.006 (0.009)
Cooperation
Poi_apple store	0.195 (1.762) *
Poi_bike	−0.019 (−0.300)
Poi_kfc	−0.473 (−4.731) ***
Poi_logistics	−0.149 (−1.475)
Landscape index
PD	−0.573 (−8.207) ***
cohesion	−0.245 (−3.821) ***
Landscape shape index (LSI)	0.299 (4.320) ***
patch fraction	0.249 (4.478) ***
Number of observations	214 county areas
Endogeneity tests
Robust score chi2	9.22 (p = 0.0024)
RobustF
Patch density (PD)	15.65
Weakiv
AR	22.25 (p = 0.000)
Wald	13.93 (p = 0.000)

*, **, and *** show significance levels of 10%, 5%, and 1%, respectively.

presents the outcomes of the 2SLS regression analysis, in which all independent variables (except the instrumental variable of the POI of infrastructure) and the endogenous variable were regressed on ESBI. The global 2SLS model serves as the foundation for MGWIVR and is used to detect instrumental and endogenous variables globally. PD was the most significant among all potential endogenous variables related to landscape pattern indicators. Other significant variables in Table 5 encompass the Weibo check-in density and built-up area growth within the localization category, road junctions and highway length within the mobility category, and the POIs of Apple retail stores and KFC stores within the cooperation category. The other landscape indices also showed significant associations with ESBI (p-value < 10%). The above findings provide empirical evidence that regionalization strategies, including localization, mobility, and cooperation, are significantly correlated with supply–demand imbalances.

The diagnosis of endogeneity and weak instruments suggests that PD can be deemed endogenous globally (robust score chi2 p < 1%) and that the POI of infrastructure is not a weak instrument (Weakiv test, p < 1%). The robust score test produced a p-value less than 1%, indicating strong evidence against the null hypothesis of exogeneity. So, PD should be treated as endogenous, and methods like instrumental variable regression are necessary to correct for this. The instrument POI of infrastructure was tested using a Weak IV test. A p-value less than 1% here suggests that this instrument is strong, meaning it has a sufficiently strong correlation with PD and can reliably be used in two-stage least squares (2SLS) regression. This confirms that the effect of infrastructure POIs on ESBI can be elucidated and replaced by PD. Since the POI of infrastructure represents the regionalization strategy of localization, it also demonstrates that PD can explain the impact of localization on the supply–demand imbalance. Nevertheless, the global 2SLS model, which assumes that the relationships between independent and dependent variables remain constant across different geographical locations, requires the integration of geographical tests to enhance its effectiveness [69] (Table 5).

The study reveals that all localization indicators demonstrate a negative association with ESBI (i.e., favorable to the economy), with standardized coefficients of −0.21, −0.08, −0.13, and −0.24. Conversely, three of the mobility variables displayed positive effects (i.e., conducive to ecology), with standardized coefficients of 0.39, 0.16, and 0.01, while train frequency exhibited a coefficient of −0.02. The findings align with those of prior scholarly works, indicating that the implementation of regionalization strategies can have varying and perhaps contrasting effects on the economic–ecology imbalance. Nevertheless, it is crucial to consider the local implications of these regionalization indicators. The global model fails to capture such nuances, thus necessitating a more in-depth examination of the positive and negative aspects within the local framework.

3.2.2. Geographical Effect of PD in MGWIVR Models

The local effect, which was absent in the global 2SLS model, was further scrutinized and enhanced in the second stage of the MGWIVR approach. This refinement allows for the estimation of localized spatial variation in regionalization strategies, contributing to a more nuanced understanding of the relationship between these strategies and supply–demand imbalances.

Figure 6 delineates the impact of the endogenous variable PD on the supply–demand imbalance at the local level. Initial observations consistently revealed an adverse effect on ESBI, suggesting that as PD increased, ESBI decreased. Because PD functions as a proxy for the level of economic development and diversity of land functions, it provides valuable insights into how localization influences the demand for ESs and the economic sector. Spatially, the observations indicate that cities in Anhui Province, which are undergoing rapid development, are more responsive to changes in PD than cities in economically prosperous regions, such as Shanghai, Zhejiang, and Jiangsu. These heterogeneous patterns suggest that the developmental circumstances of cities influence the relationship between PD or localization and economic–ecological imbalance.

3.2.3. The Regulation of Local Effect of Regionalization Strategies and Tendency of Supply–Demand Imbalance

Spatial coefficient patterns indicative of regionalization factors impacting the supply–demand imbalance are similarly observed. In Figure 7a, the impact of localization factors, including Weibo check-in density, nighttime light intensity, and population density, was notably diminished in the northwestern region of YRDCA, encompassing cities such as Hefei, Chuzhou, Anqing, Yancheng, Nanjing, and Yangzhou. To comprehensively evaluate disparities among provinces, scatter plots were employed to represent the Weibo check-in density as a localization factor within the three provinces and Shanghai metropolis (Figure 7b). The ESBI values demonstrated a distinct stratified pattern among the provinces, with varying amounts across the four regions. In Zhejiang Province, the coefficient of Weibo check-in density exhibits the highest mean value of −0.140, suggesting that the localization factor stimulates ESs demand. However, the provinces of Jiangsu and Anhui, along with Shanghai, which are experiencing an ESs supply deficit, may exacerbate this imbalance by implementing localization techniques.

Figure 7c illustrates the heightened impact of mobility factors, including road junctions, highway length, and high-speed rail frequency, in the southwestern region of Zhejiang Province, which encompasses cities such as Jinhua, Taizhou, and Quzhou. The scatter plot in Figure 7d employs highway length to delineate various mobility characteristics, specifically focusing on four distinct regions. The layering pattern among the regions is evident through the presence of regional coefficients. In Zhejiang Province, the highway length yielded the highest mean coefficient value of 0.43, whereas Anhui Province had the lowest mean coefficient value of 0.1. One plausible explanation is that mobility may initially enhance the supply of ecosystem services (ESs), resulting in positive coefficients for all four places. However, the scenario in Zhejiang, with an excess supply of ESs compared to the demand, can be exacerbated by the increasing use of mobility plans. Conversely, in Anhui and Jiangsu provinces, as well as in the metropolis of Shanghai, where there is an ESs supply shortage, increased mobility can contribute to achieving a more equitable distribution of ESs.

The spatial distribution patterns of localized impacts resulting from regionalization indicators project future trends in mitigating or exacerbating economic–ecological imbalances. Figure 5f shows that Anhui, Jiangsu, and Shanghai are particularly susceptible to localization causes and currently face an ESs deficit, indicating sustained growth in ESs demand. In contrast, Zhejiang, influenced by mobility and facing a surplus, has the potential to amplify ESs supply (Figure 7). This observation highlights the enduring nature of the imbalance problem. It is crucial to mitigate localized strategies to halt these deteriorating trends and coordinate the balance between the regional economy and ecology.

3.3. Local Scenario-Based Strategies for Coordinating Imbalance Supported by PLUS

Using MGWIVR analysis, the provinces of Anhui and Jiangsu, along with the city of Shanghai, were designated for the MP scenario because of their high ESs demand. In contrast, Zhejiang Province, with an excess supply of ESs, is earmarked for the LP scenario. The ND scenario encompassed the entire YRDCA region. Independent simulations were conducted for each region using the PLUS model to project ESBI values for 2025 and evaluate the mitigating effects of these scenarios. The three scenarios were pre-established within the PLUS framework using the land use transition table detailed in Supplementary Figure S5. For initial parameters, I set the neighborhood size as 3, thread as 5, patch generation as 0.9, and expansion coefficient as 0.1.

The initial observations revealed diagnostic values for the kappa and figure of merit (FoM) near 1 (kappa = 0.94, FoM = 0.88). This reliability was achieved by employing land use maps from 2010 and 2015 as input, assessing the similarity between the simulated 2020 land use map and the actual 2020 map, and ensuring the accurate prediction of future ESBI values. The kernel density of land use change and land use type change from 2020 to 2025, calculated by the “variety calculation tool” from the “focal statistics” toolbox in Arcmap10.8 with a unit size setting of 7 × 7, is shown in Supplementary Figure S6.

Figure 8 illustrates the rate of change in the ESBI metrics from 2020 to 2025. The ND scenario is expected to exacerbate the deficit in ESs in the provinces of Anhui, Jiangsu, and Shanghai, significantly widening the existing imbalance, with the deficit ranging from −91.9% to −490%. Nevertheless, the MP scenario effectively addressed the deficit issue by increasing the ESBI from 17.6% to 133.4% in Anhui, Jiangsu, and Shanghai, signifying substantial mitigation. The LP scenario presents a potential solution to the surplus problem in Zhejiang, reducing the ESBI by 12.0% to 65.7%. Overall, the MP and LP scenarios contribute to mitigating the disparity between economic and ecological development, as ESBI approaches zero. These findings empirically support the implementation of localized scenario-based solutions aimed at effectively coordinating and addressing the economy–ecology imbalance.

Figure 9 illustrates the variations in ESBI values across 2015, 2020, and 2025, providing a measure of imbalances. A sample of 10 cities from the YRDCA region was selected to graphically depict and compare the data, considering the health benefits calculated based on green space and population growth, using the methodology proposed by Richard and Popham (2008) [70]. The health benefit level is proxy to total population and total green space area. The comparison included the simulated health benefits in 2025 under the MP and LP scenarios and contrasted them with the ND scenario.

In 2025, compared with the ND scenario, all cities (except those in Zhejiang) exhibited improved health benefits under the LP and MP scenarios. For instance, Shanghai, when comparing 2020 and 2025, demonstrates a move towards a greater balance, with the ESBI value shifting from −43.6 to −43.2. Although its health benefit value decreases from −2.5 to −4.5 under the MP scenario, it remained higher than that under the ND scenario (−4.6). Hence, the ecologically and economically balanced state achieved by the local implementation of the MP or LP scenario may not immediately enhance health benefits in that region. However, it alleviates pressure and increases health benefits compared with the ND scenario. It can also be hypothesized that continued population growth and the loss of green spaces may diminish health benefits that would otherwise accrue. Nevertheless, at the local level, incorporating MP or LP scenarios can effectively balance economic–ecological differences and address health-related challenges.

4. Discussion

As observed in the result, the northwestern and southeastern areas of Yangtze River Delta, which are divided by the northwest adjacent boundary of Zhejiang, Jiangsu, and Anhui, have huge differences in the ecosystem services balance as well as the impacts of regionalization strategies. In detail, the strategies of localization, mobility, and cooperation along with the landscape patterns work differently. Localization and the landscape pattern weigh more in the northwestern areas, while mobility influences the southeastern areas more, which has led to ESs imbalance, with “deficit” problems in the north and “surplus” issue in the south observed in 2020. Then, the simulation uncovers that the future mobility-priority (MP) and localization-priority (LP) scenarios compensate for the ESs imbalance purposefully in northwestern and southeastern areas, respectively, in 2025. This key finding of regional difference becomes an interesting phenomenon and could be explained in various ways.

One is urban sprawl, which is a counterbalance with ecological resources (shrub, grass, and forest) [71]. Urban sprawl has eliminated the land use type of cropland, forest, and all greenery land [72] and, thus, fosters the urban clusters and competes out of the possible ESs supply providers [73] in the Su-Xi-Chang, Ning-Hang, and Shanghai metropolitan areas with ecological “deficit” problems. In contrast, natural elements such as geographical topography factors along with economic weakness, which slow down the urban sprawl [74], has limited the urban agglomeration formed in southwestern Zhejiang and the southern part of Anhui. Therefore, these areas need a unique way to enhance localization.

The other reason is uneven policy implementation in Yangtze River Delta, which decided the fundings and efficiencies in different cities, including urban planning control [75], ecological security control [76], urban–rural cooperation [77], urban developing control [41], etc. The political factor not only controls for urban sprawl but also decides the urban vitality. For example, policies implementation in Zhejiang, eastern Jiangsu, and Shanghai, limited by geographical space but with the advantage of coastal trading, support mobility and cooperation more, with stricter regulations for ecological protection at same time. Moreover, Zhejiang is leading in practicing the “rural revitalization strategy” [16], adding to its urban–rural intercommunication and mobility, and forming a “community-based” approach. To compare, Anhui Province, as the hinterland of Yangtze River Delta, is as the pilot of the “new-type urbanization era” with its effort to support the citizenization of rural workers and update the household registration system, leading a “people-oriented urbanization” [22], is thus under the great influence of the localization process. Other aspects like economy development level, local cultural difference, climate change, ecological redlines, and carbon “source-sink” can also affect the regional difference between the southern and northern areas in ESs balance and management level.

Endogenous relationships support the “community-based” approach. There were few studies seeking to explore the endogenous and geographical relationships among regionalization strategy, landscape patterns, and ecological–economy imbalance. The result of this study is consistent with existing research, showing that the regionalization process can promote the self-organization of landscape change, therefore promoting a bottom-up, community-based approach [78,79]. After regionalization activities bringing about vast changes in land use, the changes in landscape configuration and structure then severely alter ecosystem processes and services [80]. There are studies also supporting the human factors of policies and planning changing the spatial imbalances in ecosystem services [81]. However, the types of regionalization processes and the endogeneity of the landscape patterns are not evaluated in these studies. This study has clarified the instrumental function of regionalization factors and endogeneity of landscape patterns, with further demonstration that landscape patterns can help explain the effect of regionalization on ESs, thus making sure that the first order of policies being imposed should be the local community-based strategies.

Future projections of the ESBI and the identification of critical areas for policy intervention underscore the importance of targeted strategies. The rapid pace of economic growth significantly shapes the spatial disparities observed in the LP and MP scenarios. Prior studies have shown that cities experiencing accelerated development and land use transition exert a greater influence on economy–ecology imbalances [16,82], which aligns with our findings on the uneven spatial relationship between regionalization strategies and these imbalances. Our analysis confirms that the effects of localization and mobility on ESs imbalance are closely linked to regional economic trajectories. Therefore, it is essential to account for differences in urban development levels when formulating strategies aimed at fostering equitable regional growth. In this context, the allocation and distribution of regional funding play a pivotal role in determining the pace and direction of sustainable development. Rather than applying a one-size-fits-all, globalized approach, this study offers a basis for adjusting funding strategies at the regional level to more effectively support the development of sustainable communities based on local needs and conditions.

5. Conclusions

The study begins by assessing the economy–ecology imbalance using the ESs matrix and then explores geographical clustering in 2020. Subsequently, we examined the relationships among regionalization strategies, landscape patterns, and economic–ecological imbalance by employing the MGWIVR model to analyze endogeneity and geographical heterogeneity. Additionally, we predicted the effectiveness of regionalization strategies, specifically, localization and mobility, by generating land transition scenarios (MP, LP, and ND) in the PLUS model to forecast the imbalanced condition and health benefits in 2025. The overarching aim of this study was to modify the imbalance between economic and ecological development through the strategic implementation of localization, mobility, and cooperation in future-oriented strategies, and to fulfil the SDG11 of developing sustainable cities and communities.

Primary observations indicate a severe economy–ecology imbalance, with ESBI values ranging from −52.2 to 62.8. Localization, mobility, and cooperation strategies distinctly influence the supply or demand of ESs. A positive association exists between localization and ESs demand (coefficients ranging from −0.244 to −0.077), whereas mobility affects ESs supply (coefficients ranging from 0.01 to 0.39). Cooperation influences localization and mobility. Additionally, various local development scenarios are expected to mitigate the imbalance by 2025, with MP strategies increasing ESBI from 17.6% to 133.4% and LP strategies reducing ESBI from 12.0% to 65.7%. Notably, there is a stronger relationship between the LP and MP scenarios and the health benefits they provide.

This study contributes significantly to understanding the economy–ecology imbalance by addressing three key perspectives. First, it compares essential regionalization strategies—localization, mobility, and cooperation—using a multisource dataset and regression analysis against an economic–ecology imbalance. Second, it elucidates the endogenous and multiscaled geographical factors contributing to regional disparities between the economy and ecology, addressing a knowledge gap. Third, it presents local scenario-based methods (MP and LP) evaluated against ND using land transition rules designed to mitigate future imbalance issues and inform government decision making for holistic equilibrium and sustainable regional growth. These findings not only enhance academic understanding but also provide actionable policy implications: They inform region-specific interventions that can support more equitable and sustainable growth, guide land use planning under differentiated local conditions, and serve as a decision-making reference for achieving coordinated development and long-term goals set forth in national and regional sustainable cities and communities plans.

These limitations and future directions are acknowledged. The ESBI method may benefit from comprehensively assessing the balance between supply and demand at finer scales, such as communities or grids. Improvements in the integration and usability of the MGWIVR and PLUS models, possibly by incorporating a regional variation module into the PLUS model, are anticipated for more refined analyses and recommendations for future policy planning.