Coordinating Industrial Restructuring and Population Dynamics for Sustainable Land–Sea Coupled Development: An Agent-Based Optimization Framework

Abstract

Coordinating socio-economic development with coastal environmental recovery is a critical challenge in rapidly urbanizing coastal regions. Few studies quantitatively integrate industrial restructuring, population dynamics, and environmental outcomes within a unified optimization framework. This study develops an Agent-Based Optimization Model of Land–Sea Processes Coupled with Socio-Economic Dynamics (ABO-LSED) within the Driver-Pressure-State-Impact-Response (DPSIR) structure to jointly optimize development drivers and regulatory responses for improved environmental outcomes. The model incorporates a pollution intensity (PI) structure as an allocation principle, deriving region-specific integrated indicators (IIs) to guide differentiated adjustments in PI reduction, sectoral growth, and population distribution across ten districts of Qingdao, China. Simulation results show that the optimization approach reduces the time required to achieve seawater quality targets from 26 to 13 years, while maintaining an average annual GDP growth rate of approximately 7%. Structural adjustments include a shift from higher to lower-intensity sectors and a moderated urbanization rate. These findings indicate that environmental recovery and economic growth can be achieved simultaneously when intensity reduction, structural transformation, and spatial redistribution are coordinated within the proposed framework. This study offers a quantitative basis for regionally differentiated policy design and provides a transferable strategy for other coastal regions.

1. Introduction

Coastal regions worldwide are experiencing intensifying pressures from rapid socio-economic development, manifesting as degraded water quality, eutrophication, and ecological disruptions such as harmful algal blooms. These challenges substantially constrain sustainable development in densely industrialized coastal zones [1]. Since the 1972 United Nations Conference on the Human Environment, integrated land-sea management has become a central focus of global environmental governance [2]. Although existing regulatory frameworks—such as the Total Maximum Daily Loads (TMDLs) in the U.S. [3] and the Marine Strategy Framework Directive (MSFD) in the European Union [4]—have yielded notable improvements, including reductions in point-source pollution and partial recovery of water clarity in some regions [5,6], they generally fail to capture the dynamic interactions between socio-economic drivers and environmental responses. Moreover, spatial and sectoral heterogeneity in pollution sources is rarely incorporated, limiting the effectiveness of regional differentiated management strategies [7,8]. These limitations highlight the need for an integrated and mechanistic framework that explicitly links socio-economic dynamics with land–sea environmental change.

The coupling between socio-economic development and coastal environmental quality is governed by complex feedbacks between human activities and ecosystem responses [9]. Conceptual perspectives such as DPSIR framework clarify how socio-economic drivers—including industrial restructuring and population dynamics—translate into environmental pressures and subsequent ecosystem-state changes [10] (Figure A1). Within this causal chain, effective responses require not only emission control but also structural transformation in economic and demographic systems. Consistent with this view, the Environmental Kuznets Curve (EKC) suggests that environmental degradation associated with early-stage economic growth can be reversed through technological upgrading and structural transition [11,12]. Together, these perspectives imply that aligning socio-economic drivers with targeted environmental responses—particularly through industrial restructuring and spatial reconfiguration of population and economic activities—represents a critical pathway for simultaneously sustaining economic development and improving coastal environmental quality [13,14,15].

Building on this causal perspective, coordinated improvement of coastal environmental quality and socio-economic development fundamentally depends on interactions among D, S, and R [16] (Figure 1). A wide range of modeling approaches has been developed to explore these linkages, including input–output analysis, Cobb–Douglas production functions models, system dynamics, and agent-based models (ABMs) [17,18,19,20]. Among them, ABMs are particularly suited to representing heterogeneous agents, spatial interactions, and nonlinear feedbacks within coupled land–sea socio-economic systems, and have therefore been increasingly applied in environmental and economic planning since the early 2000s [21,22,23,24,25,26,27].

Recent studies have further extended ABMs toward multi-objective optimization to support coordinated land–sea management, enabling explicit linkages between water quality improvement, pollution control, and socio-economic development [28,29]. Nevertheless, two critical limitations remain. First, most existing models insufficiently represent the full heterogeneity of industrial sectors and population dynamics that determine pollution generation, preventing a direct and quantitative linkage between industrial restructuring and measurable coastal water quality responses [24,30,31]. Second, optimization frameworks rarely incorporate mechanisms that reconcile efficiency, equity, and cooperative behavior among agents, while integrated bio-economic approaches often treat environmental and economic objectives in isolation [32,33]. Consequently, existing modeling frameworks remain limited in their ability to capture the structural pathways through which socio-economic transformation can simultaneously drive environmental recovery and sustainable development.

Industrial structure optimization is widely regarded as a key pathway for reducing source-level emissions in coastal systems [6,34,35]. However, the absence of a comprehensive benchmark framework for PI across heterogeneous emission sources constrains systematic evaluation of how industrial restructuring translates into measurable coastal environmental improvement [36]. This limitation hinders full integration of socio-economic transformation with land–sea environmental dynamics and underscores the need for modeling approaches that explicitly represent interactions among drivers, environmental states, and policy responses while enabling quantitative operationalization through coordinated indicators.

To address this gap, this study develops an Agent-Based Optimization Model of Land–Sea Processes Coupled with Socio-Economic Dynamics (ABO-LSED). The novelty of ABO-LSED lies in quantitatively coordinating development drivers (D), regulatory responses (R), and environmental state (S) by using the pollution intensity (PI) structure as a bridge between economic output and pollution generation, translating environmental constraints into regionally differentiated adjustments in industrial structure, population distribution, and pollution intensity. By embedding a fairness-oriented constraint that aligns emission responsibility with economic contribution, the framework enables coordinated coastal environmental recovery while sustaining moderate-to-high socio-economic growth.

Model implementation is supported by two complementary datasets that connect emission processes with coastal water quality outcomes. The first is a total maximum allocated load (TMAL) matrix of land-based total nitrogen (TN) for 2381 source units in Qingdao, incorporating industrial discharges, municipal treatment reductions, watershed retention, and land–sea fluxes [37]. The second is a 7 categories, 13-level Environmental Access Benchmark System defining PI across industrial and domestic sources [38]. Together, these datasets enable data-constrained optimization of PI structure, industrial development, and population dynamics, providing an empirically grounded framework applicable to real-world coastal management.

2. Materials and Methods

2.1. Study Area

Qingdao, a major coastal city in China, was selected as the study area due to its intensive industrialization, high population density, and significant pressures on fragile coastal ecosystems [39,40]. Despite efforts since 2000 to mitigate land–sea pollution [41], eutrophication, especially excessive dissolved inorganic nitrogen (DIN) accumulation, has continued to degrade coastal water quality, underscoring the need for coordinated socio-economic and environmental management.

Located along the Yellow Sea on the southern Shandong Peninsula, Qingdao governs approximately 11,300 km² of land and 12,200 km² of marine area, encompassing ten districts. Its coastline extends over 800 km, featuring 49 bays and 120 islands. The city has experienced rapid growth, with GDP rising from CNY 48.03 billion in 1990 to CNY 1.41 trillion in 2021, and its population growing from 6.7 million to 10.26 million over the same period [40]. The concentration of industry and population along the coastline—particularly around Jiaozhou Bay (JZB)—has intensified land-sea pollution linkages, posing significant challenges for sustainable management [39,42]. Qingdao is approaching a critical stage in the EKC, where economic growth and environmental quality are expected to converge [14,41], making it an ideal case for exploring strategies that balance economic development and ecosystem health.

For fine-scale analysis, the 167 administrative jurisdictions across Qingdao’s 10 districts—Shinan, Shibei, Licang, Laoshan, Chengyang, Huangdao, Jiaozhou, Jimo, Laixi, and Pingdu—were subdivided into 2381 source units [43] (Figure 2). This high-resolution spatial framework provides the basis for spatially explicit analysis of pollutant loads and socio-economic activities in the subsequent modeling.

2.2. Data Sources

This study utilizes multiple authoritative datasets from government reports and peer-reviewed literature to ensure analytical robustness and data comprehensiveness. Data on TN pollution loads were compiled for both point sources—including industrial facilities, service sectors, and urban domestic activities—and non-point sources, such as agricultural production and rural households, across Qingdao’s 167 jurisdictions. These data were primarily obtained from [44,45,46]. Socio-economic data, including gross value added (GVA) for agriculture, industry, and services, as well as the population statistics for rural and urban areas, were derived from [44,45] (with [45] covering the period 1980–2021). Data on DIN concentrations in Qingdao’s coastal waters were collected from multiple sources, including [47,48,49] (with [48] for 2011–2017 and [49] for 2002–2016), as well as academic studies [46,50,51,52,53]. Finally, data on TN discharges into the sea through estuaries and outlets was obtained from [48,49,50].

2.3. Method

2.3.1. Framework of ABO-LSED

The Agent-Based Optimization Model of Land–Sea Processes Coupled with Socio-Economic Dynamics (ABO-LSED) is structured following the Overview, Design Concepts, and Details (ODD) protocol [54,55], ensuring clarity, reproducibility, and transparency in the model design and assumptions.

(1)

Purpose

The primary objective of the ABO-LSED framework is to integrate socio-economic development (D), regulatory responses (R), and environmental state (S) within the DPSIR framework. The model is designed to explore pathways that enable sustained economic and population growth while meeting coastal environmental quality targets.

By explicitly linking source-level pollutant generation to marine environmental responses, the framework provides a quantitative basis for coordinating development and environmental management across the land–sea system.

(2)

Entities, state variables, and scales

The model comprises 2381 source units as the basic entities, which are treated as decision-making agents in the optimization framework. These units are adopted from a previously established land–sea environmental modeling framework and serve as the fundamental spatial units for both pollution accounting and socio-economic analysis [43]. Each source unit represents a spatially explicit socio-economic–environmental unit characterized by relatively homogeneous land-use patterns and pollutant generation processes.

The delineation of source units is based on an integrated consideration of land-use types, watershed and river network distributions, wastewater catchment areas, and administrative boundaries. This design ensures consistency with both natural hydrological processes and socio-economic statistical units. In particular, the units are aligned with sub-watershed characteristics for non-point source transport and with wastewater collection systems for point sources, thereby providing a unified representation of pollutant generation and transport pathways. Collectively, the 2381 source units cover more than 95% of the jurisdictional area, with each unit characterized by a dominant land-use type [43].

Key state variables associated with each agent include the GVA of different economic sectors, population size, and pollution intensity (PI) levels. These variables jointly determine pollutant generation and socio-economic dynamics within each unit.

The model operates at an annual temporal resolution and a fine spatial resolution at the source-unit level, enabling the representation of spatial heterogeneity in economic activity, population distribution, and pollution generation. Compared with conventional aggregate models based on regional averages, this fine-scale representation reduces aggregation bias and supports differentiated optimization and more targeted land–sea coordination strategies.

(3)

Process overview

Each source unit simultaneously functions as a driver of socio-economic development and a source of pollutant generation. Changes in economic activity and population size directly affect pollutant loads through their interaction with PI levels. These loads are subsequently linked to coastal environmental conditions via a concentration–response relationship [37].

This process establishes a causal chain from socio-economic drivers to environmental outcomes, allowing the model to explicitly capture how adjustments in economic structure, population distribution, and pollution intensity propagate through the system and influence coastal water quality.

(4)

Sub-models

The ABO-LSED framework consists of three interconnected modules that together represent the land–sea socio-environmental system.

The first module describes industrial and population growth dynamics at the source-unit level. It simulates changes in sectoral GVA and population size through adjustable growth rates, thereby capturing the evolution of socio-economic drivers.

The second module translates these socio-economic changes into pollutant loads. Using PI coefficients for industrial and domestic sources [38,56], the model converts economic activity and population into pollutant generation at the source-unit level. In this formulation, PI serves as a key control variable linking development patterns to environmental pressure. These two modules constitute the primary focus of the present study.

The third module links land-based pollutant discharges to coastal water quality through a concentration–response matrix [37]. The concentration–response matrix and the associated land–sea environmental constraints adopted in this study were established in previous work based on a 3D hydrodynamic–water quality model coupled with linear programming. These previous studies quantified source-to-sea transport, concentration responses, and TMAL under specified water-quality targets, and reported satisfactory model performance and reliability [37].

Together, these modules operationalize the DPSIR framework by explicitly linking socio-economic drivers (D) to environmental state changes (S) through intermediate pressure and response processes.

The ABO-LSED framework incorporates a multi-objective optimization module operating at an annual time step to determine optimal development trajectories from the base year (2015) to the target year. Industrial and population growth rates, along with corresponding PI levels for each source unit, are treated as decision variables.

The optimization seeks to maximize socio-economic development potential while satisfying environmental constraints, particularly the TMAL. At the same time, it enforces an optimal PI structure of pollution sources (see Section 2.3.2), ensuring a balanced distribution of environmental pressure.

Through this formulation, the model reconciles economic and population growth with environmental quality objectives, and generates a set of integrated indicators to evaluate system performance under the optimized configuration. The overall modeling framework is illustrated in Figure 3.

2.3.2. Mathematical Formulation of the ABO-LSED Model

The ABO-LSED framework involves several categories of parameters with distinct roles in the formulation and optimization processes. Some parameters are derived from empirical data and previous studies, including pollution-intensity (PI) level system and the TMAL, which define the external environmental boundary conditions of the model.

Other parameters are introduced as policy- or management-relevant assumptions, such as the environmental Gini coefficient threshold used to represent structural equity in pollution allocation. In addition, several parameters are specified as bounded optimization settings, including the allowable growth intervals for GDP and population and the fluctuation range of TMAL-related constraints. These settings together define the feasible solution space of the optimization model (

Table 1. Key parameters used in the ABO-LSED framework.

Parameter	Description	Unit	Basis	Range/Value
PI levels	Discrete pollution intensity levels classified into 7 categories and 13 sub-levels, defining emission intensity per unit GDP	kg/(104 CNY)	Benchmark system from previous study [38]	7 categories, 13 sub-levels
TMAL of TN pollution generated from source unit	Total maximum allocated load of TN pollution generated from source unit	ton/km2/a	Previous study [37]	Fixed value
Environmental Gini coefficient (Gn)	Equity threshold constraining the PI structure	–	Assumed constraint	≤0.4
GDP growth intervals	Allowable growth intervals for sectoral GVA	%	Optimization setting (based on statistical data)	±50% around the 7.9%
Population growth intervals	Allowable growth intervals for urban and rural population	%	Optimization setting (based on statistical data)	±10% around the 2%
TMAL fluctuation	Allowable variation range of pollutant-load constraints under uncertainty	%	Optimization setting (based on literature [37])	±20%

).

Based on the parameter definitions and settings described above, the mathematical formulation of the ABO-LSED model is presented as follows. For ease of reference, the abbreviations used throughout the manuscript are summarized in Abbreviations, and the key variables used in the model are listed in

Table A2. List of Formula Symbols.

Associated Formulation	Symbols	Definition
Land–sea coupling framework	i	Index of wastewater outlet
	j	Index of source unit
	$F$	Emission load from land to sea
	$W$	Pollution generated from source unit
	$E$	Pollution reduction by MSTS
	$T$	Pollution reduction by soil/river-retention
Linking socio-economic variables to pollution loads	m	Industrial sources
	n	Domestic sources
	k	Pollution intensity level
	$V_m$	GVA of economic sectors
	$P_n$	Population size
	$\omega$	PI of pollution sources
	${\theta }_n$	Population–income regression coefficient (103 CNY/person)
	t	Number of years since the base year
	BY	Base year (2015)
	$\overline{V}$	Annual growth rates of the economic sectors
	$\overline{P}$	Annual growth rates of population
Formulation of the PI structure of emission sources	$Gn$	Environmental Gini coefficient, representing the PI structure in this study
	$M_{\left(i,j,k\right)}\%$	Cumulative proportions of sectoral GVA and residents’ income at the PI level of k to their sum across all PI levels
	$L_{\left(i,j,k\right)}\%$	Cumulative proportions of pollution loads from industrial sources and domestic sources at the PI level of k to their sum across all PI levels
Construction of integrated indicators	$\overline{GDP}\left(t\right)$	Projected GDP growth rate across Qindao in the year $t$
	$\overline{Pop}\left(t\right)$	Projected population growth rate across Qindao in the year $t$
	${wg\left(t\right)}_m$	Proportion of sectoral GVA in year $t$ across the agricultural, industry, and service sectors
	${\overline{V}\left(t\right)}_m$	Annual growth rate of sectoral GVA in year $t$ across the agricultural, industry, and service sectors
	${wp\left(t\right)}_n$	Proportion of population in year $t$ across the urban and rural area
	${\overline{P}\left(t\right)}_n$	Annual growth rate of population in year $t$ across the urban and rural area
	*	Optimal results of the optimization model
	$\widehat{D}$	High-pressure districts
	$\check{D}$	Low-pressure districts
	$\widehat{i,j,k}$	High-pressure source units
	$\check{i,j,k}$	Low-pressure source units
	$ind$	Type of indicators, including PI reduction, GVA, population growth
	$RI$	The change value of indicators
	$\overline{RI}$	The change rate of indicators
	$\mathrm{\Delta }{S}_m^{\mathrm{*}}$	Additional industrial land
	$gva$	The GVA indicator
	${\overline{S}}_m$	The industrial GVA per unit area
	$\mathrm{\Delta }{S}_n^{\mathrm{*}}$	Additional residential land
	$pop$	The population indicator
	${\overline{S}}_n$	Mean population density, defined as the number of individuals per unit area
Optimization implementation	$\mathrm{\Delta }k$	Permissible range of variation for graded PI, which is generally set to 1
	DL	Lower limits
	UL	Upper limits

.

(1)

Land–sea coupling framework

The land–sea coupling process describes the progression of pollutants from terrestrial sources to coastal waters. The resulting pollutant loads influence marine environmental conditions (e.g., DIN concentrations) through established concentration–response relationships [37] (Figure 3). It can be formulated as:

$$F_{(i,j)}=W_{(i,j)}-E_{(i,j)}-T_{(i,j)}$$

(1)

where the subscript i indicates the index of wastewater outlet, and j indicates the source unit. For each source unit, $F$, $W$, $E$, $T$ indicate the emission load from land to sea, pollution generated from source unit, pollution reduction by Municipal sewage treatment system (MSTS), soil/river-retention, respectively.

It should be noted that this source-to-sea process has been previously established and quantitatively constrained in earlier studies, in which optimal values at different stages were determined under specified environmental targets [37,57]. In the present study, the land–sea coupling is represented in an operational sense as a load-based linkage between socio-economic drivers and coastal environmental outcomes. These results are adopted as exogenous constraints, particularly in the form of optimal source-level pollutant loads.

The focus of this study is to extend the framework on the driver (D) side. Specifically, given the externally defined pollutant load constraints, source-level emissions are further decomposed into economic activity, population, and pollution intensity (PI). The model then explores how these components can be optimally allocated across regions to achieve environmental targets while maintaining socio-economic development.

(2)

Linking socio-economic variables to pollution loads

Consequently, the pollutant loads generated from source unit are estimated based on the 7 categories, 13-level PI classification system [38].

$$W_{(i,j)}=\sum _m\sum _k{W}_{m(i,j,k)}+\sum _n\sum _k{W}_{n(i,j,k)}=\sum _m\sum _k{\omega }_{(m,k)}{V}_{m(i,j,k)}+\sum _n\sum _k{\omega }_{(n,k)}{\theta }_n{P}_{n(i,j,k)}$$

(2)

where the subscript m and n indicate industrial and domestic sources, and k indicates the PI level [56]. $V_m$ and $P_n$ indicate the GVA of economic sectors and the population size, respectively, while $\omega$ indicate the PI of pollution sources. The parameter ${\theta }_n$ indicates the population–income regression coefficient (10³ CNY/person).

Statistical analysis indicates that both sectoral GVA and population in Qingdao exhibit exponential growth over time [45]. Accordingly, ${V\left(t\right)}_{m(i,j,k)}$ and ${P\left(t\right)}_{n(i,j,k)}$ can be expressed as follows:

$${V\left(t\right)}_{m(i,j,k)}={V\left(BY\right)}_{m(i,j,k)}{\left(1+{\overline{V}}_{m(i,j,k)}\right)}^t$$

(3a)

$${P\left(t\right)}_{n(i,j,k)}={\theta }_n{P\left(BY\right)}_{n(i,j,k)}{\left(1+{\overline{P}}_{n(i,j,k)}\right)}^t$$

(3b)

where t indicates the number of years since the base year. BY indicates the base year (2015). $\overline{V}$ and $\overline{P}$ indicate the annual growth rates of the economic sectors and population, respectively.

In general, the PI of agriculture, industry, and service sectors tends to decline with technological progress, while industrial upgrading occurs throughout the process of industrialization. While, during urbanization, the pressure induced by improved living standards can be offset by the expansion of MSTS, resulting in relatively minor changes in the PI of urban and rural domestic sources. Accordingly, Equation (2) can be reformulated as:

$$\begin{gathered} {W\left(t\right)}_{(i,j)}=\sum _m\sum _k\left({\omega \left(t\right)}_{(m,k)}\left({V\left(BY\right)}_{m(i,j,k)}{\left(1+{\overline{V}}_{m(i,j,k)}\right)}^t\right)\right) \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} +\sum _n\sum _k\left({\omega \left(BY\right)}_{\mathrm{n}(i,j,k)}\left({\theta }_n{P\left(BY\right)}_{n(i,j,k)}{\left(1+{\overline{P}}_{m(i,j,k)}\right)}^t\right)\right) \end{gathered}$$

(4)

The above equations formally represent source generated pollutant loads as functions of socio-economic development (drivers) through the PI framework, thereby establishing a direct linkage between socio-economic growth and environmental pressures in the ABO-LSED model.

(3)

Formulation of the PI structure of emission sources

In this study, the PI structure of emission sources is introduced based on the concept of the environmental Gini coefficient, which quantifies the consistency between economic contribution and pollution emissions across different intensity levels. It can be formulated as:

$$Gn=1-\sum _k\left(\left(\sum _i\sum _j{M}_{\left(i,j,k\right)}\%-\sum _i\sum _j{M}_{\left(i,j,k-1\right)}\%\right)\left(\sum _i\sum _j{L}_{\left(i,j,k\right)}\%+\sum _i\sum _j{L}_{\left(i,j,k-1\right)}\%\right)\right)$$

(5)

$$M_{\left(i,j,k\right)}\%=\sum _i\sum _j\left(V_{m\left(i,j,k\right)}+{\theta }_n{P}_{n\left(i,j,k\right)}\right)/\sum _k\sum _i\sum _j\left(V_{m\left(i,j,k\right)}+{\theta }_n{P}_{n\left(i,j,k\right)}\right)$$

(6a)

$$L_{\left(i,j,k\right)}\%=\sum _i\sum _j\left({\omega }_{\left(m,k\right)}{V}_{m\left(i,j,k\right)}+{{\omega }_{\left(n,k\right)}\theta }_n{P}_{n\left(i,j,k\right)}\right)/\sum _k\sum _i\sum _j\left({\omega }_{\left(m,k\right)}{V}_{m\left(i,j,k\right)}+{{\omega }_{\left(n,k\right)}\theta }_n{P}_{n\left(i,j,k\right)}\right)$$

(6b)

where $Gn$ indicates the environmental Gini coefficient, representing the PI structure in this study. $M_{\left(i,j,k\right)}\%$ indicates the cumulative proportions of sectoral GVA and residents’ income at the PI level of k to their sum across all PI levels. Accordingly, $L_{\left(i,j,k\right)}\%$ indicates the cumulative proportions of pollution loads from industrial sources and domestic sources at the PI level of k to their sum across all PI levels.

Within this framework, the PI structure functions as an allocation principle that links economic output to emission responsibility. It ensures that sectors or source units with higher pollution intensity bear proportionally greater reduction requirements, while those with lower intensity are allowed more development space.

At the same time, the PI structure also acts as a structural adjustment mechanism. The optimization process can approach the target PI distribution through two complementary pathways: (i) reducing pollution intensity in high-intensity sectors (e.g., through technological improvement), and (ii) reallocating economic activity toward lower-intensity sectors. These adjustments reflect a coordinated evolution of industrial structure and emission patterns.

This framework implies a trade-off between equity and efficiency. While stricter PI structure constraints may limit short-term growth in high-intensity sectors, they promote long-term environmental capacity and enable more balanced development across sectors. Therefore, the PI structure serves both as an equity constraint and as a mechanism guiding efficient resource reallocation.

In the present study, this principle is operationalized through a set of regionally differentiated indicators for economic growth, population dynamics, and PI adjustment, which guide the system toward the target structure within the specified environmental constraints.

(4)

Construction of integrated indicators

The IIs comprise three primary dimensions. First are regional differentiated PI discharge permit indicators, including the optimal pollutant discharge permit limits for industrial PI and their corresponding annual reduction rates. Second is regional differentiated economic sectoral GVA indicators, including the optimal GVA values for economic sectors and the associated annual growth rates; and third is regional differentiated urban and rural population indicators, including optimal population sizes and annual growth rates for both urban and rural areas.

Collectively, the IIs are designed to facilitate evidence-based policy-making that simultaneously addresses economic growth and environmental goals. By incorporating pollution control, industrial restructuring, and population management into a unified framework, the model provides a systematic approach to promote sustainable development in Qingdao, thereby enhancing both long-term environmental quality and economic resilience.

The annual total economic growth rate is generally represented as a weighted average of the growth rates across the agricultural, industrial, and service sectors. Accordingly, the projected GDP and population growth rate in year $t$ can be expressed as:

$$\overline{GDP}(t)=\sum _m\left({wg\left(t\right)}_m{\overline{V}\left(t\right)}_m\right)$$

(7)

$$\overline{Pop}(t)==\sum _n\left({wp\left(t\right)}_n{\overline{P}\left(t\right)}_n\right)$$

(8)

where $\overline{GDP}\left(t\right)$ and $\overline{Pop}\left(t\right)$ indicate the projected GDP and population growth rate across Qindao in the year $t$. And ${wg\left(t\right)}_m$ indicates the proportion of sectoral GVA in year $t$ across the agricultural, industry, and service sectors and ${\overline{V}\left(t\right)}_m$ indicates the corresponding annual growth rate. Analogously, ${wp\left(t\right)}_n$ indicates the proportion of population in year $t$ across the urban and rural area and ${\overline{P}\left(t\right)}_n$ indicates the corresponding annual growth rate.

Under environmental constraints, achieving economic (GDP) and population growth within the planning horizon requires, according to Equation (4), not only adjustments in pollution intensity (PI), but also optimization of the industrial GVA structure and population distribution.

From a management perspective, administrative districts serve as the primary units for implementing the indicator system. To ensure sustainable development, all districts—irrespective of their pressure level—are required to reduce industrial PI, as current levels remain considerably high. Similarly, achieving high-quality development requires that both industrial GVA and population across all districts grow at a moderate and sustainable rate. However, the magnitude and rate of changes in PI, GVA, and population exhibit substantial spatial heterogeneity.

To characterize this heterogeneity, the deviation between the optimal and base-year values can be employed to estimate the pressure levels of industrial GVA, population, and PI for each source unit. Following the load matrix accuracy (±30%) reported by previous study [37], the pressure levels of source units are categorized into three classes: high-pressure (load exceeding the average by more than 30%), medium-pressure (load within ±30% of the average), and low-pressure (load below the average by more than 30%).

Accordingly, for districts classified as medium-pressure, the optimal values and rates of PI reduction, GVA and population growth are computed as the weighted averages of the respective source units within the district. While for high-pressure districts, it can be formulated as:

$$\overline{RI}{\left(t\right)}_{\left(ind, \widehat{D}\right)}^{\#\#}=RI{\left(t\right)}_{\left(ind,\widehat{D}\right)}^{\#\#}\times 100\%/{RI\left(BY\right)}_{\left(ind,D\right)}$$

(9a)

$$RI{\left(t\right)}_{\left(ind,\widehat{D}\right)}^{\#\#}=\sum _{\widehat{i}}\sum _{\widehat{j}}\sum _{\widehat{k}}RI{\left(t\right)}_{\left(ind,\widehat{i,j,k}\right)}^{\#\#}-\sum _{\check{i}}\sum _{\check{j}}\sum _{\check{k}}R{I\left(t\right)}_{\left(ind,\stackrel{{\displaystyle ˇ}}{i,j,k}\right)}^{\#\#}$$

(9b)

and for low-pressure districts, as:

$$\overline{RI}{\left(t\right)}_{\left(ind, \check{D}\right)}^{\#\#}=RI{\left(t\right)}_{\left(ind,\check{D}\right)}^{\#\#}\times 100\%/{RI\left(BY\right)}_{\left(ind,D\right)}$$

(10a)

$$RI{\left(t\right)}_{\left(ind,\check{D}\right)}^{\#\#}=\sum _{\check{i}}\sum _{\check{j}}\sum _{\check{k}}R{I\left(t\right)}_{\left(ind,\check{i,j,k}\right)}^{\#\#}-\sum _{\widehat{i}}\sum _{\widehat{j}}\sum _{\widehat{k}}R{I\left(t\right)}_{\left(ind,\widehat{i,j,k}\right)}^{\#\#}$$

(10b)

The pressure status of each administrative district is determined by comparing the aggregated high-pressure and low-pressure source units within the district, which can be expressed as follows:

$$R{I\left(t\right)}_{\left(ind,\widehat{i,j,k}\right)}^{\#\#}=R{I\left(t\right)}_{\left(ind,\widehat{i,j,k}\right)}-R{I\left(t\right)}_{\left(ind,\widehat{i,j,k}\right)}^{\#\#}$$

(11a)

$$R{I\left(t\right)}_{\left(ind,\check{i,j,k}\right)}^{\#\#}=R{I\left(t\right)}_{\left(ind,\check{i,j,k}\right)}^{\#\#}-R{I\left(t\right)}_{\left(ind,\check{i,j,k}\right)}$$

(11b)

where the superscript ## indicates the optimal results of the optimization model, and subscript $\widehat{D}$ and $\check{D}$ indicate high-pressure and low-pressure districts, $\widehat{i,j,k}$ and $\check{i,j,k}$ indicates the corresponding high- and low-pressure source units respectively, $ind$ indicates the type of indicators, including PI reduction, GVA, population growth. $RI$ and $\overline{RI}$ indicate the change value and rate of indicators.

For the GVA indicator, If the aggregate decrement of industrial GVA in high-pressure source units is lower than the increment in low-pressure source units, the adjustment of industrial layout can be achieved concurrently with the reduction of industrial pollution, implying that the desired structural optimization does not require additional spatial expansion. Conversely, additional industrial land must be allocated to compensate for the imbalance. It can be quantified as:

$$\mathrm{\Delta }{S}_m^{\#\#}=\left(\sum _{\widehat{i}}\sum _{\widehat{j}}\sum _{\widehat{k}}RI{\left(t\right)}_{gva\left(\widehat{i,j,k}\right)}^{\#\#}-\sum _{\check{i}}\sum _{\check{j}}\sum _{\check{k}}RI{\left(t\right)}_{gva\left(\check{i,j,k}\right)}^{\#\#}\right)/{\overline{S}}_m$$

(12)

where $\mathrm{\Delta }{S}_m^{\#\#}$ indicates the additional industrial land, subscript $gva$ indicates the GVA indicator. ${\overline{S}}_m$ indicates the industrial GVA per unit area.

Similarly, the additional residential land required can be formulated as:

$$\mathrm{\Delta }{S}_n^{\#\#}=\left(\sum _{\widehat{i}}\sum _{\widehat{j}}\sum _{\widehat{k}}RI{\left(t\right)}_{pop\left(\widehat{i,j,k}\right)}^{\#\#}-\sum _{\check{i}}\sum _{\check{j}}\sum _{\check{k}}RI{\left(t\right)}_{pop\left(\check{i,j,k}\right)}^{\#\#}\right)/{\overline{S}}_n$$

(13)

where $\mathrm{\Delta }{S}_n^{\#\#}$ indicates the additional residential land, subscript $pop$ indicates the population indicator. ${\overline{S}}_n$ indicates the mean population density, defined as the number of individuals per unit area.

(5)

Optimization implementation

1.

Objective function

The optimization problem is formulated as a constrained nonlinear programming problem. The objective function is formulated to maximize socio-economic development potential under environmental constraints. This is achieved by combining normalized total GDP and pollutant discharge into a single objective, allowing the model to maximize economic output while fully utilizing the allowable pollution capacity. It can be expressed as:

$$\sum _i\sum _j{W}_{(i,j)}^{\#\#}=max\sum _i\sum _j{W}_{(i,j)}$$

(14a)

$$\sum _i\sum _j{V}_{m(i,j)}^{\#\#}=max\sum _i\sum _j{V}_{(i,j)}$$

(14b)

$$\sum _i\sum _j{P}_{(i,j)}^{\#\#}=max\sum _i\sum _j{P}_{(i,j)}$$

(14c)

According to Equation (2), population is converted into the GDP variable through the population–income regression coefficient.

Population dynamics are represented as a growth-based redistribution process rather than through explicit migration modeling. Population distribution is adjusted via differentiated growth rates across source units, with higher growth assigned to regions with lower environmental pressure and greater development capacity, and more restrictive growth applied to environmentally constrained areas. This formulation captures the gradual reallocation of population toward regions with higher environmental carrying capacity over the planning horizon.

The relationship between industrial restructuring and population change is represented through the response of population distribution to economic opportunities. Regions with higher economic growth are assumed to attract greater population increases, while the transition toward service-oriented industries is associated with urbanization.

Population growth is constrained within predefined bounds derived from historical trends, reflecting limits imposed by infrastructure, resource availability, and environmental capacity. Given the medium- to long-term perspective of this study, short-term migration dynamics and temporal lags are not explicitly modeled; instead, population adjustment is treated as an aggregated response to economic restructuring.

2.

Constraints

The model is subject to multiple constraints, including (i) PI structure constraints represented by the environmental Gini coefficient (Gn ≤ 0.4), (ii) upper and lower bounds of PI levels based on benchmark standards [38], (iii) allowable growth intervals for GDP and population derived from historical trends, (iv) total pollutant load limits (TMAL) and (v) sectoral structure constraints to promote a transition toward lower-intensity industries. All constraints are incorporated as explicit inequality constraints in the optimization problem.

(i)

PI structure of pollution sources

Following international practices [58], A threshold of 0.4 is adopted following established practice, representing an acceptable level of distributional equity of PI among pollution sources.

$$Gn\le 0.4$$

(15)

(ii)

Coefficient interval of industrial PI

Due to the spatial heterogeneity, industrial PI may exhibit considerable variability among source units. Accordingly, the feasible coefficient interval can be expressed as:

$${\omega }_{m\left(i,j,k-\mathrm{\Delta }k\right)}\le {\omega }_{m\left(i,j,k\right)}^{\#\#}\le {\omega }_{m\left(i,j,k+\mathrm{\Delta }k\right)}$$

(16)

where $\mathrm{\Delta }k$ indicates the permissible range of variation for graded PI, and $\mathrm{\Delta }k$ is generally set to 1.

(iii)

Growth rate interval of economy and population

In both the five-year socio-economic development plan and the annual government plan, the expected targets for GDP and population growth are typically specified as intervals [59]. According to Equation (7), the constraint on the annual GDP growth rate interval can be expressed as follows:

$${\overline{GDP}}_{DL}\le {\overline{GDP}}^{\#\#}\le {\overline{GDP}}_{UL}$$

(17)

where the subscripts DL and UL indicate the lower and upper limits, respectively. Based on statistical data, the average annual GDP growth rate of Qingdao since 2010 has been approximately 7.9%, with an average fluctuation of about 50%. Accordingly, the lower and upper bounds for GDP growth were set within a ±50% fluctuation range. Similarly, according to Equation (8), the annual population growth rate constraint can be expressed as:

$${\overline{Pop}}_{DL}\le {\overline{Pop}}^{\#\#}\le {\overline{Pop}}_{UL}$$

(18)

Based on statistical data, the annual population growth rate is approximately 2%, with the lower and upper limits set within a ±10% fluctuation range.

(iv)

TMAL interval of pollution discharged by industrial and domestic sources

Each source unit typically comprises multiple pollution sources across varying PI levels. Accordingly, the constraints on the TMAL intervals for industrial and domestic sources can be expressed as follows:

$$\sum _k{W}_{m\_DL(i,j,k)}^{\#\#}\le \sum _k{W}_{m(i,j,k)}^{\#\#}\le \sum _k{W}_{m\_UL(i,j,k)}^{\#\#}$$

(19)

$$\sum _k{W}_{n\_DL(i,j,k)}^{\#\#}\le \sum _k{W}_{n(i,j,k)}^{\#\#}\le \sum _k{W}_{n\_UL(i,j,k)}^{\#\#}$$

(20)

The average calculation accuracy of TMAL for pollution sources—including agriculture, industry, services, urban life, and rural life—is approximately 20% in Qingdao [37]. Therefore, the upper and lower bounds of both TMAL intervals for industrial and domestic sources were set within a ±20% fluctuation.

2.3.3. Assessment of Applicability

Given the intrinsic coupling among environmental quality improvement (S), pollution control (R), and socio-economic development (D), the applicability of the proposed IIs was evaluated from four complementary perspectives [58].

First, accuracy was quantified by comparing the mean pollutant loads derived from Equation (4) with the corresponding outputs of a coupled 3D hydrodynamic–water quality and linear programming model [37]. Second, equity in IIs across districts—an essential requirement for coordinated environmental governance—was assessed using the t-test and the polymerization degree (σ) to quantify differences between the index values of high-pressure and low-pressure source units [56]. Third, proportionality of industrial and domestic pollution contributions within each source unit was evaluated through the distribution proportionality coefficient (β) [60], reflecting the internal coordination of pollution control responsibilities. Finally, efficiency was examined by comparing the time required to achieve full compliance with the DIN concentration target in Qingdao’s coastal waters [61] under three policy scenarios: the current reduction rate (3%), a one-size-fits-all (OSFA) strategy (4.5%), and the optimal IIs proposed in this study.

2.3.4. Model Evaluation and Robustness Assessment

Validation Scope and Strategy

The source-to-sea pollutant transport pathway, including source-level emissions, engineering reduction, and land–sea transport processes, is adopted from previously established and validated studies and is treated as an external environmental constraint in this work. Accordingly, the present study does not re-calibrate or re-validate this transport module.

Instead, the validation focus of this study is placed on the newly developed socio-economic optimization layer. Specifically, the model is assessed in terms of whether it can produce adjustment patterns in GDP, population, and PI that are broadly consistent with historical development trends. This historical-consistency-based validation strategy helps ensure that the optimized pathways are anchored in realistic socio-economic dynamics, while the environmental response module remains fixed as an externally defined boundary condition.

Sensitivity Analysis

To assess the robustness of the ABO-LSED framework, a scenario-based sensitivity analysis was performed by systematically varying key constraint ranges rather than empirically derived parameter values [62]. This approach reflects the constraint-based nature of the optimization framework. Three groups of constraints were considered: (i) the allowable adjustment range of PI within each intensity class, (ii) the TMAL, which represents environmental carrying capacity, and (iii) the bounds on economic and population growth.

The PI range was relaxed and tightened by ±10% to represent uncertainty in technological improvement and emission-control flexibility. The TMAL constraint was varied by ±20% to account for potential uncertainty in environmental capacity and load estimation. The GDP and population growth bounds were proportionally adjusted by ±20% to reflect alternative socio-economic development conditions.

For each scenario, changes in key outputs—including the average PI reduction rate, GDP growth rate, population growth rate, and spatial allocation pattern—were compared with the baseline solution. This procedure was used to evaluate the structural robustness of the optimization results under plausible variations in constraint settings.

3. Results

3.1. Regional Differentiated PI Reduction Indicators

To achieve the targeted improvement in coastal water quality by 2030, substantial cross-sectoral reductions in PI are required. The weighted average annual PI reduction rates in agriculture, industry, and service across Qingdao were 6.3%, 8.5%, and 7.1%, respectively (Figure 4a). Accordingly, the permissible levels of PI in agriculture, industry, and service in Qingdao were projected to decreased annually from V, IV, and IV in 2015 to IV, III, and III in 2030, corresponding to overall reductions of 85%, 90%, and 87%, respectively (Figure 5a). These results indicate that meeting the coastal water-quality target requires coordinated source-level mitigation across sectors rather than reliance on isolated end-of-pipe controls. Moreover, the required reduction rates exhibit pronounced spatial heterogeneity among districts, suggesting that future emission-reduction strategies should follow regional differentiated pathways aligned with local industrial composition and environmental capacity.

Across sectors, high-pressure districts accounted for 60%, 70%, and 70% of agricultural, industrial, and service PI, respectively, whereas medium-pressure districts accounted for 0% in all sectors and low-pressure districts accounted for the remaining 40%, 30%, and 30%. This uneven distribution reveals substantial spatial disparities in emission pressure and consequently in required mitigation intensity. Specifically, the required annual reduction rates in high-pressure districts increased by 26%, 70%, and 42% for agriculture, industry, and services, respectively, whereas those in low-pressure districts decreased by 6%, 26%, and 28%, highlighting the necessity of differentiated regional control strategies.

3.2. Regional Differentiated Industrial GVA Indicators

Under the optimal configuration, sustained economic growth in Qingdao remains achievable while meeting emission-reduction requirements. Assuming an average annual economic growth rate of 7.0%, the weighted average annual GVA growth rates in agriculture, industry, and services were 4.6%, 7.0%, and 8.6%, respectively (Figure 4b). Accordingly, sectoral GVA is projected to increase from CNY 36, 409, and 485 billion in 2015 to CNY 137, 720, and 1709 billion by 2030, respectively (Figure 5b). These results demonstrate the feasibility of decoupling economic expansion from pollution emissions under the optimal pathway. Correspondingly, the sectoral structure of agriculture, industry, and services shifts from 4:44:52 in 2015 to 5:28:67 in 2030, indicating a transition toward a service-dominated and lower-intensity economic system. This structural transformation is consistent with patterns observed in developed economies, where the service sector typically contributes more than 60% of GDP [63].

Spatially, the distribution of GVA growth requirements varies markedly across districts. High-pressure districts accounted for approximately 20%, 50%, and 40% of agricultural, industrial, and service GVA, respectively, whereas medium-pressure districts accounted for about 30%, 0%, and 0%, and low-pressure districts accounted for roughly 50%, 50%, and 60%. This pattern reveals pronounced spatial heterogeneity in required GVA growth trajectories. Specifically, annual growth rates in low-pressure districts increased by 26%, 21%, and 33% across the three sectors, whereas those in high-pressure districts decreased by 54%, 11%, and 2%, respectively, reflecting a spatial reallocation of economic expansion toward environmentally sustainable regions.

3.3. Regional Differentiated Urban and Rural Population Indicators

Under the optimal scenario, population growth in Qingdao remains compatible with environmental improvement targets while exhibiting clear spatial differentiation. Assuming an average annual population growth rate of 2.0%, the corresponding demand growth rates for the urban and rural population were 2.1% and 0.5%, respectively (Figure 4c). Accordingly, the urban population and rural population were projected to increase annually from 6.5 and 2.6 million in 2015 to 8.8 million and 3.0 million in 2030, respectively (Figure 5c). This demographic trend indicates intensified urban coastal pressure, highlighting the need for spatially integrated land–sea population management.

Spatial distribution patterns further reveal pronounced regional differentiation. High-pressure districts contain approximately 70% of the urban population but only 30% of the rural population, whereas medium-pressure districts account for about 30% of both, and low-pressure districts host nearly 40% of the rural population with negligible urban concentration. Over the planning horizon, annual population growth in low-pressure districts increases by 0% in urban areas and 25% in rural areas, while growth in high-pressure districts declines by 60% and 40%, respectively. This redistribution pattern demonstrates the effectiveness of the optimization framework in steering population expansion toward environmentally sustainable regions, thereby alleviating coastal ecological pressure while maintaining overall demographic growth.

To provide an integrated view of the optimized pathway,

Table 2. Comparison between baseline (base year or historical trajectory) and optimized scenarios.

Indicator	Baseline (Historical or Base Year)	Optimized Scenario	Change
Permissible PI levels (Agri–Ind–Serv)	V, IV, and IV (base year)	IV, III, and III	Reduced
Avg. GDP growth (%)	~7.9 (historical trajectory)	~7	Maintained
Avg. Population growth (%)	~2 (historical trajectory)	~2	Maintained
Urbanization rate (%)	71 (base year)	74	Increased
Industrial structure (Agri:Ind:Serv)	4:44:52 (base year)	5:28:67	Service-oriented shift
Time to meet DIN target (years)	~26	~13	Reduced by ~50%

“Avg” indicates average value; “Agri”, “Ind” and “Serv” indicate agriculture, industry, and service sector, respectively.

compares the baseline conditions with the optimized scenario in terms of PI levels, economic and population growth, industrial structure, and the time required to meet the DIN target. The comparison shows that the optimized pathway substantially accelerates environmental improvement while maintaining socio-economic development, thereby providing an overall context for evaluating the applicability of the integrated indicators.

3.4. Applicability of the Integrated Indicators

The applicability of the IIs was evaluated in terms of accuracy, equity, proportionality, and environmental effectiveness. First, the relative deviation (RD) between the average pollutant loads predicted by Equation (4) and those simulated using a coupled 3D hydrodynamic-water quality and linear programming model [37] was 15%, demonstrating strong agreement with an independent modeling framework and confirming the accuracy of the IIs. Second, the mean polymerization degree ($\sigma$) between high-pressure and low-pressure source units was 4 and statistically significant (p < 0.05), indicating clear differentiation in pollution reduction responsibilities among agents according to their relative contributions to marine pollution and thereby reflecting the equity of the IIs. Third, the proportionality coefficient of distribution (β) for industrial and domestic pollution loads within each source unit declined from 0.8 in 2015 to 0.4 in 2030, suggesting a progressively more balanced allocation of pollution sources under the optimal configuration. Finally, the time required to achieve full compliance with the DIN concentration target in Qingdao’s coastal waters was projected to be 26, 20, and 13 years under the current policy reduction rate (3.0%), the OSFA strategy (4.5%), and the IIs proposed in this study, respectively. This substantial acceleration of environmental recovery highlights the effectiveness of the IIs as a decision-support tool for coordinated land–sea management. Collectively, these results confirm that the ABO-LSED-derived IIs provide a robust, equitable, and efficient framework for simultaneously advancing coastal environmental restoration and sustainable socio-economic development.

3.5. Sensitivity Analysis of Constraint Settings

To further evaluate the robustness of the ABO-LSED framework, sensitivity analysis was conducted by varying three key constraint settings, namely the PI adjustment range, TMAL, and growth bounds (

Table 3. Sensitivity analysis of constraint settings in the ABO-LSED framework (relative changes compared to baseline).

Scenario	Constraint Adjustment	$∆$ PI Reduction Rate (%)	$∆$ GDP Growth Rate (%)	$∆$ Population Growth Rate (%)	Spatial Allocation Pattern
S0	Baseline	0	0	0	Reference pattern
S1	PI range relaxed (+10%)	+1.4%	+1.3%	~0	Similar
S2	PI range tightened (−10%)	−2.1%	−1.4%	−1.2%	Similar
S3	TMAL relaxed (+20%)	−7.0%	+2.8%	+2.5%	Slightly more balanced
S4	TMAL tightened (−20%)	+9.3%	−2.0%	−1.7%	Stronger shift to low-pressure sectors
S5	Growth bounds relaxed (+20%)	+2.1%	+3.3%	+2.7%	Similar
S6	Growth bounds tightened (−20%)	−1.4%	−4.7%	−3.8%	Slightly constrained

). The results show that changes in these constraints primarily affect the magnitude of the optimized outcomes rather than the direction of the solution, indicating that the model is structurally stable under moderate perturbations. This pattern reflects the internal logic of the optimization framework: because the objective function maximizes socio-economic development under environmental and structural constraints, the optimal solution tends to evolve along the feasible boundary defined by these constraints.

Adjustments to the PI constraint range mainly influence the balance between economic growth and emission-intensity reduction. Relaxing the PI range (+10%) increases flexibility in intensity adjustment, particularly for high-intensity sectors, and therefore allows slightly higher GDP growth (+1.3%) together with a moderate increase in the PI reduction rate (+1.4%), while population growth remains nearly unchanged. By contrast, tightening the PI range (−10%) restricts the scope for emission-intensity adjustment and shifts the optimization toward tighter control of economic activity, resulting in lower PI reduction (−2.1%), GDP growth (−1.4%), and population growth (−1.2%). Because PI functions as an intermediate variable linking economic activity and pollutant generation, changes in its allowable range have only a limited effect on the overall spatial allocation pattern.

By comparison, TMAL exerts the strongest influence on model outcomes because it directly defines the environmental carrying capacity of the system. Relaxing TMAL (+20%) expands the feasible solution space, allowing higher GDP growth (+2.8%) and population growth (+2.5%) while reducing the need for intensive PI reduction (−7.0%). Conversely, tightening TMAL (−20%) reduces the allowable pollutant load and therefore requires substantially stronger PI reduction (+9.3%), accompanied by lower GDP growth (−2.0%) and population growth (−1.7%) and a more pronounced shift toward low-intensity regions. These results indicate that TMAL acts as the most binding environmental constraint in the optimization framework.

Growth bounds mainly regulate the scale of socio-economic expansion. Relaxing these bounds (+20%) allows the model to accommodate higher GDP growth (+3.3%) and population growth (+2.7%), together with a slight increase in PI reduction (+2.1%) under fixed environmental constraints. In contrast, tightening growth bounds (−20%) directly limits the scale of development, reducing GDP growth (−4.7%) and population growth (−3.8%), while correspondingly lowering the pressure for PI reduction (−1.4%). Compared with TMAL, changes in growth bounds mainly affect the scale of adjustment rather than the underlying allocation logic.

Overall, the sensitivity of the model can be ranked as TMAL > growth bounds > PI range. Across all scenarios, variations in key indicators remain within a relatively limited range, and the spatial allocation pattern is largely preserved. This suggests that the optimization results are structurally robust under reasonable variations in constraint settings. Such robustness is likely supported by the combined effects of bounded constraint design and the PI-structure constraint, which help prevent extreme or unrealistic solutions during optimization. At the same time, the results also indicate that constraint settings must remain within a reasonable range: excessively large deviations may lead to infeasible solutions or convergence failure, whereas overly relaxed constraints may yield mathematically feasible but practically unrealistic outcomes. The constraint ranges adopted in this study therefore represent a balanced configuration that supports both model stability and practical relevance.

4. Discussion

4.1. Coordinating Development Drivers and Regulatory Responses Under Integrated Environmental and Growth Constraints

Improving coastal environmental quality under a binding TMAL constraint fundamentally requires coordinating development drivers (D), regulatory responses (R), and environmental outcomes (S), rather than merely tightening emission standards. The optimization results demonstrate that achieving the targeted reduction in pollutant loads by 2030—while maintaining an average annual economic growth rate of approximately 7%—necessitates simultaneous adjustments in PI, sectoral composition, and population distribution. Under the combined constraints of TMAL, PI-structure equity, and bounded economic and demographic growth, structural transformation naturally emerges as a necessary outcome of the optimization process. This mechanism aligns with recent evidence showing that comprehensive environmental programs can significantly promote industrial structure upgrading and coastal water quality improvement in Chinese coastal regions [64].

The PI-structure constraint plays a central role in this coordination mechanism. By allocating proportionally greater reduction responsibilities to high-intensity sources while allowing lower-intensity sectors to maintain or moderately expand within the overall load cap, the model embeds an equity-based distribution principle into the optimization. Although this arrangement may incur higher short-term costs, it releases more environmental capacity for higher-margin, lower-intensity industries over time, leading to an optimal allocation of environmental resources across sectors, and ensuring a decline in total pollutant discharge without imposing uniform contraction across all regions [65]. This dynamic is reflected in the projected shift of the agriculture–industry–service structure toward a more service-oriented economy. The results suggest that maintaining economic expansion under environmental constraints depends on reallocating incremental growth to sectors with lower marginal PI [66]. Thus, structural upgrading is not externally imposed but arises from the need to simultaneously satisfy environmental capacity and growth limits. The transformation of population distribution further supports this structural adjustment. The projected increase in urbanization from 71% to 74% by 2030 signals continued demographic concentration in urban areas. However, under optimal constraints, this concentration is moderated in environmentally sensitive zones.

Consequently, economic growth, PI reduction, and demographic allocation evolve as mutually reinforcing processes, rather than competing objectives, indicating that reconciling economic expansion with coastal environmental recovery depends on integrating regulatory responses into development trajectories, and environmental improvement (S) becomes achievable when development drivers (D) and regulatory controls (R) are optimized together, allowing structural upgrades and PI reductions to emerge as coordinated outcomes [67].

4.2. Spatial Heterogeneity and Policy-Operational Implications

Spatial heterogeneity plays a crucial role in the efficiency of environmental governance within the context of a fixed total allowable load. The optimization results reveal significant differences in pollution intensity, sectoral growth potential, and demographic expansion across districts. High-pressure districts, for example, contribute most of the industrial pollution, and their required annual PI reduction rates significantly exceed those of low-pressure districts—consistent with earlier findings that heavy industry remains the dominant source of nitrogen emissions in coastal regions [41,46]. At the same time, projected economic and population growth adjustments vary greatly between these pressure categories, as districts differ markedly in their contributions to marine pollutant loads. This discrepancy leads to moderated demographic expansion in high-pressure areas, while lower-pressure regions can accommodate more growth. These disparities demonstrate that environmental pressure is unevenly distributed across space, and that uniform reduction standards would impose disproportionate constraints on some regions while insufficiently addressing others [68].

Under the differentiated pathway, high-pressure districts primarily accommodate adjustment through more aggressive PI reductions and moderated expansion of high-intensity sectors, whereas lower-pressure districts take on a larger share of incremental economic and demographic growth within the overall environmental cap. These differentiated adjustments also have significant implications for spatial planning. For instance, sectoral growth differences translate into varied land-use requirements, with high-pressure districts requiring relatively less additional industrial land, while lower-pressure areas have more room for expansion. Similarly, population adjustments imply differentiated residential land expansion needs, estimated at 30 km² in core urban districts and up to 200 km² at the metropolitan scale based on projected density assumptions—a trend consistent with the ongoing functional urban restructuring that characterizes China’s broader urbanization trajectory [69]. These spatial considerations show that environmental capacity constraints not only reshape economic structure but also urban form and land allocation patterns [70].

The IIs operationalize this differentiation by specifying district-level PI reduction intervals, sectoral growth boundaries, and population adjustment targets. Compared with a uniform reduction scenario, the differentiated pathway significantly reduces the projected time required to achieve compliance with DIN concentration targets. This demonstrates that aligning regulatory intensity with localized environmental pressure can enhance governance efficiency while sustaining aggregate growth.

These results suggest that regional policy design should treat environmental capacity as a spatially heterogeneous constraint. Differentiated, quantitatively derived targets offer a more adaptive and economically efficient strategy for coordinating development and environmental recovery in land–sea coupled systems.

4.3. Historical Consistency and Feasibility of Regional Differentiated Indicators

The practical feasibility of the optimal pathway can be evaluated by a retrospective consistency check by comparing the proposed targets with historical development trajectories.

Sustained reductions in PI have already been observed in China, with average annual declines of approximately 8.1% for industrial sources and 3.7% for agricultural sources [34]. In Qingdao, the long-term average annual PI reduction rate since 1980 has approached 9.0%, exceeding the 7.0% reduction target projected under the IIs. This comparison indicates that the PI reduction targets derived from the model are consistent with historically observed magnitudes, supporting the reliability of the optimization results in reflecting realistic environmental improvement trajectories [66].

A similar conclusion emerges from the economic and demographic dimensions (

Table A1. Comparisons between the annual growth rates since 2010 (historical data) and the demand annual growth rates (optimal solution) of GVA of agriculture, industry, and service; and urban and rural population in 10 districts of Qingdao.

Districts	AGR-A (%)		AGR-A (%)		AGR-A (%)		AGR-P (%)		AGR-P (%)
	Agriculture		Industry		Service		Rural		Urban
	HD	OS	HD	OS	HD	OS	HD	OS	HD	OS
Shinan	0.00	0.00	8.32	2.54	11.18	11.18	0.00	0.00	0.55	0.00
Shibei	0.00	0.00	21.63	13.33	14.95	14.95	0.00	0.00	0.56	0.00
Licang	0.00	0.00	3.42	8.56	16.47	16.47	0.00	0.00	1.49	−0.14
Laosahn	7.79	1.21	7.76	2.18	13.89	7.31	−1.12	3.59	4.45	0.00
Chengyang	−5.75	−0.88	3.99	−1.65	11.13	16.00	−1.81	−1.77	3.22	−1.32
Huangdao	10.00	10.79	11.05	9.56	17.45	18.24	−3.29	4.89	6.01	−1.53
Jiaozhou	5.76	2.12	8.38	−1.21	13.93	10.29	−2.03	0.58	7.85	−0.88
Jimo	7.67	8.57	12.44	−2.85	13.70	14.59	−2.34	2.25	11.24	−2.15
Laixi	6.33	5.82	6.56	0.26	9.09	8.58	−0.50	−0.81	5.29	−1.51
Pingdu	5.46	1.20	8.07	−0.86	10.22	−0.29	−0.36	−2.84	6.36	−2.87

“AGR-A” indicates annual growth rate of GVA; “AGR-P” indicates annual growth rate of Population; “HD” indicates historical data; “OS” indicates optimal solution.

). Between 2010 and 2021, average annual growth rates in Qingdao reached 3.0% in agriculture, 8.7% in industry, and 9.8% in services [45], deviating only moderately from the optimal sectoral growth intervals derived in this study, with an average deviation of 28%, The model-implied sectoral growth patterns are therefore broadly consistent with historical trends, particularly in reflecting the relative expansion of the service sector and the moderated growth of agriculture and industry. Urban and rural population growth rates (4.7% and –1.6%, respectively) were likewise broadly consistent with the differentiated adjustment ranges proposed by the model.

It should also be noted that the baseline year (2015) was adopted as a consistent reference point for integrating socio-economic and environmental data across all source units. Since then, Qingdao has undergone substantial changes, including industrial restructuring, rapid urbanization, and continued declines in pollution intensity. These shifts may affect the absolute magnitude of the adjustments required to achieve the 2030 targets. In particular, the expansion of the service sector and the decline in PI suggest that the current system may already be closer to a lower-intensity development trajectory, whereas increasing population concentration in coastal areas may intensify localized environmental pressure. Nevertheless, these changes do not fundamentally alter the structural implications of the model, which is intended to identify coordinated development pathways under environmental constraints rather than to provide point predictions tied to a single base year.

Taken together, these comparisons indicate that the optimization does not impose unrealistic structural constraints but instead redistributes growth within historically observed bounds, thereby enhancing policy implementability.

Evaluating the feasibility of the approach also requires comparison with international practices. In the United States and European Union, discharge permits operate under robust legal and monitoring systems [6,36]. Conventional permit systems generally apply uniform reduction standards, with limited ability to account for spatial heterogeneity or sector-specific characteristics. In contrast, the ABO-LSED framework enhances traditional permit systems by integrating three key features: (i) spatial differentiation, identifying high-, medium-, and low-pressure zones for targeted interventions; (ii) sector-specific allocation, assigning PI reduction tasks according to industry contributions and economic significance; and (iii) coupled optimization, simultaneously balancing economic growth and environmental quality under explicit constraints.

4.4. Policy Implications of the Indicator System

A key policy value of the ABO-LSED framework is that it translates environmental constraints into operational regional indicators for economic development, population growth, and PI, thereby moving beyond conventional assessment- or scenario-oriented approaches. Unlike many DPSIR-based, land–sea coordination, system-dynamics, and scenario-based studies, which are more effective in diagnosis and trajectory exploration, the present framework explicitly links environmental targets to differentiated regional adjustment pathways and thus provides a more direct basis for policy design [71,72,73,74,75]. Compared with behavioral agent-based models that focus primarily on micro-level responses, this framework is more directly oriented toward optimization-based allocation under environmental constraints [76,77].

The optimized IIs therefore provide quantitative guidance for differentiated governance across high-, medium-, and low-pressure regions. High-pressure regions require stricter PI reduction targets, tighter discharge permitting, and faster technological upgrading of high-intensity sectors, whereas medium-pressure regions should emphasize controlled growth together with gradual industrial restructuring. Low-pressure regions retain greater capacity to accommodate incremental economic and population growth under the same environmental constraints and may therefore be supported through spatial planning, infrastructure provision, development quotas, and regional development incentives. In practice, these pathways can be implemented through linked policy instruments rather than isolated measures: regulatory and technological actions, such as emission permitting, stricter discharge standards, wastewater treatment upgrading, cleaner production technologies, stormwater management, and best management practices (BMPs), can typically be initiated in the short to medium term, while industrial restructuring and population adjustment are more appropriately understood as medium- to long-term governance processes shaped through zoning, industrial upgrading, entry restrictions for high-intensity sectors, urban planning, and infrastructure allocation [78,79,80]. This phased interpretation is consistent with practical experience in load-based governance. For example, the Chesapeake Bay Total Maximum Daily Load (TMDL) program adopted staged milestones from 2010 to 2025, with 60% of required reductions targeted by 2017 and full implementation by 2025, while measures such as wastewater upgrades and BMP adoption generated measurable progress within shorter time frames [81,82]. The optimized pathways proposed here should therefore be interpreted not as one-step prescriptions, but as differentiated medium- to long-term policy guidance.

Effective implementation also requires institutional arrangements capable of coordinating environmental responsibilities and development opportunities across districts. Cross-regional quota allocation, ecological compensation, and performance-based evaluation systems could help redistribute responsibilities and incentives under shared environmental constraints. In this context, the environmental Gini coefficient (Gn ≤ 0.4) provides an operational benchmark for equitable allocation by assessing whether emission responsibilities are broadly aligned with regional economic contributions. Embedding this principle into discharge permitting or quota allocation systems may help avoid excessive concentration of pollution burdens in specific regions or sectors. The sectoral implications of the indicator system are likewise important. In particular, the relatively large reduction requirement in agriculture should be interpreted as the cumulative outcome of long-term improvement rather than a one-time technological adjustment, reflecting both the relatively high initial PI of agricultural activities and the diffuse nature of non-point source emissions. Accordingly, the implied transition depends on sustained improvements in nutrient-use efficiency, livestock waste recycling and treatment, non-point source control, and gradual adjustment of high-intensity production practices over multiple planning periods.

4.5. Uncertainty and Broader Applicability

Despite its demonstrated feasibility, the framework remains subject to several sources of uncertainty. A primary limitation arises from the way land–sea linkage is represented. In this study, source-to-sea environmental effects are incorporated through an externally established concentration–response matrix and linearized environmental constraints, rather than being re-simulated within the optimization model itself. Although the environmental response module was developed and validated in previous studies for Qingdao using a 3D hydrodynamic–water quality model and linear programming, it still simplifies seasonal variability, monitoring uncertainty, extreme events, and complex biogeochemical interactions. As a result, the framework is more suitable for representing long-term environmental constraints than for precisely simulating short-term ecological dynamics [37].

Additional uncertainty stems from the socio-economic optimization layer. Some model inputs, including pollution intensity coefficients and growth bounds, rely on aggregated data and reasonable assumptions, which may affect the quantitative estimates. Moreover, linearized environmental constraints and bounded growth intervals improve computational tractability, but may not fully capture nonlinear socio-economic transitions, abrupt structural changes, or ecological tipping behavior over extended planning horizons. In this sense, the framework should be interpreted as a constrained optimization model for identifying feasible and differentiated allocation pathways under environmental constraints, rather than as a dynamic process model or a behavioral simulation model. Within this scope, TMAL and water-quality targets function as effective environmental boundary conditions. Consistent with this modeling objective, robustness analysis in the present study focuses on the socio-economic optimization layer, and the main differentiated allocation patterns remain stable under variations in key constraint settings, including PI ranges, GDP and population growth bounds, and TMAL.

Uncertainty also exists at the implementation stage. The effectiveness of the IIs depends on consistent regulatory enforcement, monitoring capacity, and inter-district coordination, so realized outcomes may deviate from the optimal projections if institutional conditions vary [83]. Nevertheless, the underlying coordination logic of the framework is not location-specific. Many rapidly urbanizing coastal regions face similar tensions among industrial upgrading, population concentration, and limited marine environmental capacity. By explicitly linking PI structure, sectoral growth bounds, and demographic allocation within an environmental load cap, the framework provides a structured basis for reconciling economic development with marine ecological recovery. Application to other coastal regions would require recalibration of PI coefficients, environmental capacity estimates, and growth parameters to reflect local conditions, but the core coordination mechanism remains transferable.

Future research could further strengthen the framework by incorporating probabilistic or scenario-based uncertainty analysis, closer coupling with process-based environmental models, and more explicit treatment of behavioral and temporal feedbacks under evolving socio-economic and environmental conditions. Continuous monitoring and periodic recalibration would also improve the long-term applicability of model-based decision support in land–sea coupled systems.

5. Conclusions

This study addressed the challenge of coordinating socio-economic development and coastal environmental improvement by developing an Agent-Based Optimization Model of Land–Sea Processes Coupled with Socio-Economic Dynamics (ABO-LSED) within the DPSIR framework to quantitatively coordinate development drivers (D) and regulatory responses (R) for coastal environmental improvement (S). By embedding the PI structure as an allocation principle into the joint optimization of industrial growth, sectoral composition, and population distribution, the model translates environmental quality targets into region-specific, operational indicators. The resulting IIs quantify allowable PI levels and differentiated reduction rates, sectoral GVA growth intervals, and population adjustment ranges at the district level.

The results demonstrate that coastal environmental improvement and sustained economic growth can be achieved simultaneously when coordinated structural adjustment is endogenously driven by the combined consideration of environmental targets, differentiated PI responsibilities, and bounded economic and demographic growth. Rather than relying on uniform emission standards, the optimal pathway reallocates reduction responsibilities toward high-intensity areas while permitting appropriate expansion of lower-intensity sectors within the overall load cap. This coordinated adjustment of PI, industrial structure, and population distribution enables pollutant reduction while maintaining a growth trajectory consistent with historical development trends and significantly shortens the projected time required to achieve water-quality compliance.

Methodologically, this study advances land–sea coupling research by moving beyond descriptive linkage analysis toward coordinated optimization. The derived IIs provide an implementable basis for regional differentiated governance, avoiding uniform reduction prescriptions and instead aligning adjustment intensity with local development structure and pollution contributions.

Overall, the proposed framework provides a quantitative and operational approach for integrating pollution control, industrial restructuring, and demographic dynamics in land–sea coupled systems. With appropriate recalibration of pollution inventories, PI benchmarks, and environmental capacity parameters, the coordination logic underlying ABO-LSED can be extended to other rapidly urbanizing coastal regions facing similar tensions between economic expansion and marine environmental recovery.