Scenario-based construction pathways for a “good urban area” under Nansha’s future urban framework, Guangdong, China

Abstract

Introduction:

Rapid peri-urban expansion poses significant challenges, including spatial inequality, ecological degradation, and mismatches between land development and public-service provision. Addressing these issues requires integrated planning approaches that balance multiple sustainability dimensions. This study aims to evaluate alternative pathways for constructing a “good urban area” in Nansha District, Guangzhou, by systematically comparing scenarios that prioritize equity, growth, environmental protection, or maintain current trends.

Methods:

We developed the Spatial–Participatory–Ecological–Coupled Governance–Smart Equity (SPECS) framework, integrating cellular automata, behavioral relocation rules, ecological constraints, governance allocation, and equity assessment. Data on land use, demographics, economy, zoning, accessibility, and stakeholder inputs were collected from 18 subdistricts. Historical land-use transitions (2005–2010) were used for model training, 2010–2015 for parameter selection, and 2015–2020 for out-of-sample validation. Four 2025–2035 trajectories were simulated: business-as-usual, equity-prioritized, growth-oriented, and environmentally constrained. Model performance was assessed via overall accuracy and Kappa coefficient, while scenario outcomes were compared using composite performance scores, ecological-redline overlap, and equity indices. Robustness was tested through Monte Carlo analysis under parameter variation and policy shocks.

Results:

Historical validation achieved 91.6% overall land-class accuracy and a Kappa coefficient of 0.872. The equity-prioritized trajectory raised procedural equity from 0.53 to 0.71 and distributive equity from 0.49 to 0.65, improved service accessibility, reduced spatial-functional mismatch, and attained the highest composite performance score (0.704). The environmentally constrained trajectory recorded the lowest ecological-redline overlap (3.2%) and strongest ecological retention. Monte Carlo analysis confirmed that both the equity-prioritized and environmentally constrained scenarios maintained robust Pareto membership and rank stability under uncertainty.

Discussion:

The equity-prioritized pathway offers the best overall performance by concurrently enhancing social fairness, service accessibility, and land-use efficiency, while the environmentally constrained option excels in ecological protection. These findings demonstrate that scenario-based planning, facilitated by the SPECS framework, provides a transparent and robust decision-support tool for comparing trade-offs among equity, ecological integrity, spatial suitability, and institutional feasibility in rapidly urbanizing peri-urban regions.

1 Introduction

Rapid urbanization intensifies the need to coordinate spatial growth, ecological protection, equitable access to urban resources, and institutional capacity for implementation. A “good urban area” is defined through five necessary dimensions: spatial compactness and connectivity, functional land-use alignment, ecological integrity, socio-economic inclusiveness, and governance coordination. These dimensions represent the physical, functional, environmental, distributive, and institutional components of urban quality. Spatial compactness and connectivity are measured through construction-land contiguity, road density, transit proximity, suitability, and landscape fragmentation. New urban district design in China similarly links spatial organization with ecological and functional development objectives (Lin et al., 2012). Functional land-use alignment measures correspondence between planned functions and population, infrastructure, and service demand through functional balance, zoning deviation, and spatial-functional mismatch. Ecological integrity is assessed through ecological-redline protection, ecological resistance, vegetation condition, wetland and forest retention, slope, and water proximity. Socio-economic inclusiveness is represented by accessibility and the distribution of services, housing, mobility, green space, and public investment. Governance coordination is assessed through planning approvals, investment capacity, regulatory alignment, stakeholder coordination, administrative readiness, and implementation timing. Urban-growth simulation in Guangzhou further demonstrates the importance of linking land allocation with planning constraints and implementation requirements (Li X. et al., 2023). The five dimensions therefore provide direct theoretical coverage of urban spatial structure, land-use function, ecological capacity, benefit distribution, and institutional implementation, with each dimension linked to measurable indicators used in the scenario assessment.

China’s rapidly transforming edge-urban districts demonstrate the difficulty of reconciling economic expansion with ecological protection, equitable service provision, and coordinated governance. National urbanization policies increasingly emphasize smart growth, ecological civilization, and regional integration, but implementation remains uneven across metropolitan fringe areas (Liu et al., 2019). Development strategies may prioritize land-based revenue, industrial clustering, and infrastructure construction while giving insufficient attention to ecological thresholds and public-service distribution (Zhang et al., 2025). These tensions are especially pronounced where national, municipal, district, and subdistrict responsibilities intersect with decentralized land-development incentives (Liu et al., 2023). Institutional fragmentation may consequently delay ecological protection, transit-oriented development, infrastructure coordination, and the equitable allocation of public services (Zhao et al., 2023).

Nansha District provides an appropriate empirical setting for investigating these interconnected urban-development challenges. Located on Guangzhou’s southern frontier, Nansha is a national-level New Area and a strategic component of the Guangdong–Hong Kong–Macao Greater Bay Area. Its development mandate combines port logistics, advanced manufacturing, regional cooperation, ecological conservation, and large-scale urban expansion (Cheng et al., 2017). The district has evolved from a peripheral development zone into a polycentric edge-urban area containing differentiated administrative, industrial, residential, agricultural, and ecological functions (Cheng and Shaw, 2021). Scenario research indicates that transport connectivity and policy orientation strongly influence the location and scale of future construction land (Zheng and Xian, 2023). Urban growth boundary modeling further identifies substantial development pressure along accessible transport and industrial corridors (Zhang and Liao, 2024). Landscape analysis has documented increasing fragmentation in development-intensive areas and reinforced the importance of protecting the southern ecological core (Huang et al., 2024). Coastal subsidence, flood exposure, and infrastructure loading impose additional restrictions on future construction intensity and location (Wang et al., 2024). Moderate coordination between land-development intensity and green-economic performance also indicates incomplete alignment between urban growth and environmental objectives (Huang et al., 2025).

Land-use simulation has advanced through cellular automata (CA), urban growth boundary (UGB), ecological-security, and scenario-based approaches for representing alternative spatial trajectories. Multi-scenario CA and UGB modelling demonstrates how planning constraints can redirect urban expansion toward spatially suitable locations (Lai et al., 2022). Ecological security pattern–Future Land Use Simulation (ESP–FLUS) modelling incorporates ecological resistance and protected-space constraints into land-transition processes (Liu et al., 2022). Three-dimensional Future Land Use Simulation (FLUS-3D) extends simulation to horizontal expansion and vertical intensification (Xu et al., 2024). Participatory geographic information system (PGIS) research connects stakeholder knowledge with spatial planning information and land-use decisions (Marondedze et al., 2024). Co-creation approaches further incorporate stakeholder priorities into spatial alternatives and implementation choices (Chen, 2026). However, these advances remain fragmented across spatial prediction, ecological protection, participation, equity assessment, behavioral responses, and governance implementation. In Nansha, this fragmentation leaves unclear how ecological resistance, spatial-functional mismatch, stakeholder priorities, household and developer behavior, investment allocation, policy shocks, equity, and institutional feasibility interact across future development trajectories. The resulting research gap is the absence of an integrated scenario-based approach for evaluating these processes within a common spatial and temporal structure across Nansha’s 18 subdistricts.

1.1 Operationalizing “good urban area” concepts in the Nansha context

Operationalizing a good urban area requires measurable criteria for spatial form, functional performance, ecological protection, socio-economic inclusion, and institutional capacity. Urban sustainability assessment has increasingly incorporated governance, resilience, participation, and context-sensitive monitoring alongside environmental, social, and economic indicators (Saraswat et al., 2025). Spatial compactness and connectivity are represented through construction-land contiguity, road density, transit proximity, suitability, and landscape fragmentation. Morphological analysis at fine spatial resolution provides measurable evidence of urban density and structural configuration (Yu H. et al., 2025). Multi-level ecological-network analysis further links urban morphology with ecological connectivity across development landscapes (Lu et al., 2026). Functional land-use alignment is assessed through the Land Resource Mismatch Index (LRMI), functional balance, zoning deviation, and service-efficiency divergence. Land-development intensity in China reflects interacting policy, economic, social, and natural drivers, supporting the use of multiple functional and development indicators (Li et al., 2024). Ecological integrity is assessed through ecological-redline overlap, ecological resistance, vegetation condition, wetland and forest retention, slope, and water proximity. Spatiotemporal ecological modelling supports repeated measurement of environmental condition and change across space and time (Liu et al., 2025). Socio-economic inclusiveness is represented through service accessibility, transit distance, housing-cost ratio, affordable-housing benefit, procedural equity, distributive equity, and the Gini coefficient. Equity-oriented urban assessment similarly demonstrates the value of combining accessibility, quality, inequality, and optimization measures when evaluating distributional outcomes (Li et al., 2026). Governance coordination is assessed through planning approvals, investment intensity, coordination frequency, participation diversity, regulatory alignment, administrative readiness, and implementation lag. These five dimensions are operationalized through corresponding indicators, data sources, model functions, and expected planning outcomes (Table 1), providing distinct measures of spatial organization, functional correspondence, ecological capacity, benefit distribution, and implementation capacity.

Chinese land-use research shows that functional zoning, ecological constraints, and spatial mismatch are often evaluated through separate analytical systems, limiting integration with equity and governance processes. Yang et al. (2025) combined major functional zones, ecological redlines, expansion-mode zoning, and cellular automata in Jiangyin, achieving a Kappa coefficient of 0.959 and identifying infill, sprawl, and leapfrog expansion patterns. Its evaluation focused on spatial allocation and planning conformity, leaving procedural equity, benefit distribution, and institutional feasibility outside the model. Yu J. et al. (2025) integrated system dynamics, FLUS, and a vigor–organization–resilience–ecosystem-services framework to simulate land-use change and ecosystem health in the Chengdu metropolitan area, while household behaviour, stakeholder influence, and service accessibility remained only weakly connected to the ecological assessment. Shen et al. (2023) developed the LRMI to diagnose over-allocation and under-allocation among urban land functions, but the index does not dynamically update mismatch in response to changing demand, migration, or investment. Shi et al. (2024) extended mismatch assessment across 285 Chinese cities through measures of planning deviation and economic mismatch from 2002 to 2021, although its city-scale structure does not resolve subdistrict interactions among zoning, accessibility, ecological resistance, and implementation conditions.

Adaptive urban-quality assessment therefore requires direct links among measurable indicators, scenario inputs, behavioural mechanisms, feedback rules, and planning outcomes. Gao et al. (2026) demonstrated through a hierarchical agent-based model that interactions among central, departmental, and village-level actors can shape land acquisition, conversion, and reclamation, yet the model does not jointly evaluate spatial form, ecological compliance, service equity, and implementation feasibility. The remaining gap is a Nansha-specific analytical structure that connects the five good urban area dimensions to a common spatial resolution and annual simulation sequence while retaining their distinct indicators and decision mechanisms. The operational structure links each dimension to its corresponding indicators, data sources, model functions, and expected planning outcomes (Table 1), enabling consistent comparison of alternative development pathways across spatial, ecological, equity, and governance conditions.

1.2 Simulating urban futures through smart-equity and participatory integration

Multi-scenario urban simulation provides a structured basis for comparing how alternative policy, ecological, and development assumptions reshape future land-use patterns. Hu et al. (2024) coupled multilevel urban-flow estimation with Patch-generating Land Use Simulation (PLUS) for the Beijing–Tianjin–Hebei region and reported higher overall accuracy, Kappa, and Figure of Merit (FoM) values than the conventional PLUS configuration, with the coordinated-development scenario producing a more stable expansion trajectory. The model demonstrates the value of incorporating economic, transport, population, and information flows, but its regional intercity structure does not resolve subdistrict-scale household behavior, coastal ecological exposure, or institutional implementation in Nansha. Li L. et al. (2023) integrated ecological-agricultural-urban suitability with FLUS to delineate alternative urban growth boundaries and reduce conflict between development and conservation. However, suitability was primarily imposed through top-down spatial regulation, leaving service distribution, stakeholder influence, and governance feasibility outside the transition process. Aghaloo and Sharifi (2025) combined PLUS and InVEST across Tehran scenarios and found that the sustainable-development pathway better preserved cropland, expanded rangeland, and moderated ecosystem-service losses. The approach clarifies ecological trade-offs, yet it does not determine whether the resulting benefits are equitably distributed or behaviorally and institutionally attainable.

Smart-equity planning requires simulations to assess both the spatial efficiency of development and the distribution of its benefits and decision-making influence. Feitosa et al. (2024) operationalized spatial justice through equality, need, merit, and entitlement criteria, demonstrating that different normative rules produce different territorial allocations. The framework clarifies why equity cannot be represented by a single accessibility or efficiency score, but it remains separate from dynamic land-transition modeling. Maheshwari and Chopra (2025) integrated geospatial, demographic, facility, and Sustainable Development Goal (SDG) indicators at the building level and demonstrated that administrative averages can conceal substantial intra-urban disparities. Their fine-scale assessment improves the identification of underserved populations and locations, although it functions as a diagnostic system instead of a mechanism that changes annual land allocation. These studies indicate that procedural and distributive equity should enter scenario construction and feedback rules directly because post-simulation assessment cannot establish whether participation or benefit distribution altered the development pathway.

Participatory integration becomes analytically meaningful when stakeholder knowledge modifies scenario parameters, spatial priorities, or decision rules. Knowledge-conversion research in participatory planning identifies institutional context as a key condition for translating stakeholder knowledge into planning decisions (Seo, 2022). Digital-twin evidence further shows that citizen influence depends on accessible visualization, interactivity, inclusiveness, privacy, and institutional readiness (Adade and de Vries, 2025). Participatory Scenario Planning (PSP) research identifies uneven stakeholder involvement across scenario-development stages and persistent limitations in empowerment and engagement (Pusingha et al., 2026). Broader evidence on sustainable urban planning similarly documents a shift from consultation toward co-creation while identifying continuing constraints from tokenism, digital divides, and power asymmetries (He et al., 2026). Participatory land-use modelling demonstrates that stakeholder-developed scenario narratives can be translated into land parcel cellular automata (LP-CA) parameters and spatially explicit development configurations (Molinero-Parejo et al., 2023). Computational participatory design further links stakeholder choices with accessibility, environmental performance, resource allocation, and multi-criteria decision analysis (MCDA) within a common spatial decision process (Nourian et al., 2024). Urban-justice research distinguishes procedural justice in decision-making from the distribution of resulting benefits, supporting separate measurement of participation processes and distributive outcomes (Yazar et al., 2025). The unresolved gap is a district-scale framework that converts stakeholder priorities into zoning elasticity, ecological-buffer distances, capital-allocation weights, and corridor coefficients and traces their effects on annual land transitions, ecological compliance, procedural equity, distributive equity, behavioral responses, and implementation feasibility.

1.3 Theoretical and analytical framework

Fragmentation between spatial form, functional alignment, ecological conditions, equity, and governance limits multidimensional assessment of a good urban area, while stakeholder priorities are rarely translated into explicit scenario parameters that can be traced through land-transition and implementation outcomes. The Spatial–Participatory–Ecological–Coupled Governance–Smart Equity (SPECS) model was developed for this study to connect these two analytical gaps within a five-module sequential-feedback architecture (Figure 1). Spatial–Functional Alignment identifies development suitability and spatial-functional mismatch; Participatory Scenario Integration converts stakeholder priorities into zoning elasticity, ecological-buffer, capital-allocation, and corridor parameters; Ecological Constraint Encoding defines protected and high-resistance areas and regulates eligible transitions; Coupled Governance Simulation updates investment allocation, household relocation, developer conversion, and policy-shock responses; and Smart–Equity Validation evaluates procedural equity, distributive equity, ecological compliance, spatial alignment, robustness, and institutional feasibility. The modules operate through annual land-transition cycles in which spatial, ecological, participatory, behavioral, and governance conditions determine allocation decisions, while feedback updates subsequent simulation years.

FIGURE 1

SPECS links urban-quality dimensions, stakeholder-defined scenario assumptions, ecological safeguards, behavioral decisions, governance allocation, and multidimensional outcomes within one traceable analytical system. Its novelty lies in converting participatory and ecological inputs into operational transition rules, coupling those rules with household, developer, and governance responses, and evaluating their spatial, ecological, procedural, distributive, robustness, and feasibility consequences within the same annual simulation pathway. This architecture enables alternative development trajectories to be compared through the mechanisms generating their outcomes and provides a district-scale basis for assessing good urban area performance in Nansha.

1.4 Objectives, research questions and hypotheses

The study aims to: (i) operationalize the good urban area construct through spatial compactness and connectivity, functional land-use alignment, ecological integrity, socio-economic inclusiveness, and governance coordination; (ii) compare land suitability, ecological compliance, accessibility, equity, and institutional feasibility across business-as-usual (BAU), equity-prioritized, growth-oriented, and environmentally constrained scenarios; (iii) determine how ecological constraints, stakeholder-defined parameters, household and developer behaviour, and governance allocation influence projected spatial outcomes; and (iv) evaluate scenario robustness under parameter variation and policy shocks.

The study is guided by the following research questions. RQ1: How do the four development scenarios differ in land suitability, spatial-functional mismatch, ecological encroachment, service accessibility, procedural equity, distributive equity, and institutional feasibility across Nansha’s 18 subdistricts? RQ2: How do ecological constraints, stakeholder-derived parameters, household relocation, developer decisions, and governance allocation explain differences among scenario outcomes? RQ3: How stable are the scenario rankings under changes in suitability, ecological resistance, development costs, population projections, fiscal conditions, flood restrictions, and zoning controls?

The study hypothesizes that H1: Equity-prioritized and environmentally constrained scenarios produce higher multidimensional good urban area performance than BAU and growth-oriented development. H2: scenarios incorporating ecological resistance and zoning constraints generate lower ecological-redline encroachment, high-resistance land conversion, and spatial-functional mismatch than unconstrained growth. H3: service accessibility and procedural equity are positively associated with distributive equity and land suitability. H4: scenarios combining ecological safeguards, equity feedback, and coordinated governance achieve greater stability, Pareto efficiency, and institutional feasibility under parameter perturbations and policy shocks than single-priority scenarios.

2 Methods

2.1 Study area

The study was conducted in Nansha District, Guangzhou, China (22°48′–23°06′N, 113°28′–113°47′E), covering 803 km² (Song et al., 2018) (see Figure 2). Spatial analysis used 100 × 100 m grids under the CGCS2000 coordinate system with Gauss–Kruger Zone 6 projection. Four functional zones were defined and rasterized: administrative core (north), logistics–innovation corridor (east), ecological belt (south), and industrial transition zone (west) based on official planning overlays.

FIGURE 2

Institutional planning was structured across four tiers. National directives followed the Greater Bay Area Plan, designating Nansha as a cooperation zone. Municipal governance applied the Guangzhou Master Plan to allocate infrastructure, housing, and environmental quotas. District policies relied on the Nansha Regulatory Plan (2020–2035) to define zoning intensities. Subdistricts used detailed control plans for boundary and quota data. Governance activity was measured using three indicators: annual planning approvals, infrastructure investment (yuan/km²), and coordination meeting frequency (2018–2023). Stakeholder consultation was conducted under institutional research authorization No. SCPEB-2025–441,902; all 28 participants provided written participation consent, and identifying information was removed before coding and spatial parameterization.

The computational unit was the 100 × 100 m cell, and outputs were aggregated to the 18 Nansha subdistricts for reporting. Four trajectories were simulated annually from 2025 to 2035: business-as-usual (BAU), equity-prioritized, growth-oriented, and environmentally constrained (Figure 3). The operational indicators for the five good urban area dimensions were calculated in Sections 2.2 and 2.5; scenario differences were generated in Section 2.3; household, developer, governance, and shock mechanisms were implemented in Section 2.4; and historical validation and uncertainty estimation were conducted in Section 2.6.

FIGURE 3

2.2 Spatial–ecological modelling framework

Historical model development used three non-overlapping temporal intervals. Land transitions from 2005 to 2010 formed the training interval used to estimate the five-class transition matrix and baseline neighborhood effects. Observations from 2010 to 2015 formed the parameter-selection interval used to select suitability weights, suitability thresholds, ecological-resistance weights, resistance thresholds, inertia coefficients, and conversion-cost multipliers. The resulting specification was frozen before evaluation against the 2015–2020 interval, which served exclusively as out-of-sample validation. Prospective simulations for 2025–2035 used the frozen historical parameter set with only the prespecified scenario parameters varying across trajectories.

Land-use rasters for 2005, 2010, 2015, and 2020 were obtained from the Guangdong Remote Sensing Center and reclassified as construction, agriculture, forest, wetland, or water (see Supplementary Figure S1). The rasters were reprojected to CGCS2000/Gauss–Krüger Zone 6 and resampled to a common 100 × 100 m grid in ArcGIS Pro (Esri, United States). Slope suitability, road accessibility, and elevation suitability were normalized to 0–1 and combined as $$S_i = 0.25L_i+0.50A_i+0.25E_i \tag{1}$$ where is the suitability score for cell , is normalized slope suitability, is normalized road accessibility, and is normalized elevation suitability. Candidate weights were evaluated at 0.05 increments using the 2010–2015 parameter-selection interval while constraining all weights to sum to 1.00. Candidate suitability thresholds from 0.55 to 0.75 were evaluated at 0.01 increments. The retained weights of 0.25, 0.50, and 0.25 and the suitability threshold of 0.65 maximized the mean of Kappa and FoM while maintaining construction-area error below 7%. The retained specification was subsequently applied without retuning to the 2015–2020 validation interval.

Zoning mismatch was assessed using a LRMI adapted from Shen et al. (2023), comparing planned versus actual land allocation proportions per cell. Weighted mismatches emphasized construction and agriculture. Cells exceeding a mismatch threshold were flagged as inconsistent. Service-efficiency divergence was measured by rank correlation between functional zoning and service access, based on road density, transit proximity, and healthcare accessibility. Divergences of two or more quantile ranks indicated zoning inefficiency. Ecological constraint layers from the 2020–2035 Nansha Redline Plan were rasterized at 100 m and used to block construction transitions during simulation.

Ecological resistance combined vegetation condition, terrain, hydrological proximity, and elevation as $$R_i = 0.35V_i+0.25G_i+0.25W_i+0.15H_i \tag{2}$$ where is ecological resistance for cell , is normalized vegetation resistance derived from NDVI, is normalized slope resistance, is normalized water-proximity resistance, and is normalized elevation resistance. Candidate resistance thresholds from 0.55 to 0.75 were evaluated at 0.01 increments using observed ecological non-conversion during 2010–2015. A threshold of 0.65 maximized agreement with observed non-conversion and was frozen before 2015–2020 validation. Cells with were classified as high-resistance land. Ecological restrictions were implemented according to scenario purpose. Ecological redlines and cells with were hard exclusions for new construction under BAU, equity-prioritized, and environmentally constrained trajectories, giving eligible construction probability within these masks. The growth-oriented trajectory functioned as an ecological-enforcement stress test and applied soft penalties: baseline construction probability was multiplied by 0.15 inside ecological redlines and by 0.35 on high-resistance land. Permanent water bodies remained excluded from construction in all trajectories. Existing 2020 construction located inside ecological redlines was retained in the initial land stock and could subsequently undergo restoration or reclassification; ecological-redline overlap therefore measured total construction stock within redlines, whereas high-resistance conversion measured new construction transitions into high-resistance cells. These definitions were applied consistently to annual area accounting.

These resistance rasters were integrated into the ESP–FLUS model, where inverse resistance guided constrained land-use transitions under stochastic cellular automata–Markov simulation. All model layers used a 100 m resolution aligned with subdistrict boundaries. Annual simulation steps from 2025 to 2035 allowed discrete updates from policy, ecological, and stakeholder layers. Validation overlays were aggregated by subdistrict using zonal statistics. No upsampling or smoothing ensured fidelity and accurate cumulative tracking.

Every annual simulation exported a complete 5 × 5 land-class transition matrix and an ecological-area ledger recording beginning area, eligible area, blocked area, new conversion, restoration or reclassification, and ending area for ecological-redline and high-resistance cells. Accounting was performed separately for each scenario and simulation year from 2025 through 2035, permitting direct reconciliation between transition rules and reported land-area outcomes (Supplementary Tables S8–S11). The accompanying analysis scripts retained scenario identifier, simulation year, Monte Carlo seed, eligibility mask, transition probability, realized class change, and final land class for every cell.

2.3 Scenario construction and stakeholder integration

BAU represented continuation of the 2020 planning rules through 2035 and used the frozen historical transition specification with a suitability threshold of 0.65, zoning elasticity of 0, capital redistribution weight of 0, corridor coefficient of 1.00, and a 300 m ecological buffer. The equity-prioritized, growth-oriented, and environmentally constrained trajectories modified only the scenario parameters defined below; historical coefficients, land-class transition structure, and calibration parameters remained fixed across trajectories.

Stakeholders were grouped into four categories: (i) Nansha Planning Bureau officials, (ii) neighbourhood association chairs from high-density zones (>3,000 persons/km²), (iii) commercial-sector actors with land rights or concessions (2018–2022), and (iv) policy advisors with prior involvement in Nansha Strategy 2020–2035. Twenty-eight participants were recruited via institutional correspondence, grouped into seven four-person panels by jurisdiction, and assigned mapping domains and codes to spatially anchor their simulation inputs. All confirmed participation by signed protocol. The two-round Delphi procedure estimated zoning elasticity, ecological buffer distance, capital redistribution, and corridor prioritization. Admissible ranges were 0.20–0.50 for zoning elasticity, 300–1,200 m for ecological buffers, 0.25–0.65 for capital redistribution, and 1.00–1.80 for corridor coefficients. Round 1 collected independent values from the 28 participants. Round 2 returned the group median, interquartile range, and entropy-weight summary for each parameter. Convergence required an interquartile range below 0.25 and a change in the median below 0.10 between rounds. Final parameter combinations were selected by maximum Euclidean separation in the four-dimensional parameter space to retain substantively distinct equity-prioritized, growth-oriented, and environmentally constrained scenarios.

The scenario layers were generated in ArcGIS Urban (Esri, United States) as 2D zoning overlays under CGCS2000/G–Krüger Zone 6. BAU used a suitability threshold of 0.65, a 300 m ecological buffer, zoning elasticity of 0, capital redistribution of 0, and a corridor coefficient of 1.00. The equity-prioritized scenario used a suitability threshold of 0.70, zoning elasticity of 0.22, an 800 m ecological buffer, a capital redistribution weight of 0.61, and a corridor coefficient of 1.30. The growth-oriented scenario used a suitability threshold of 0.60, zoning elasticity of 0.41, a 500 m ecological buffer, a capital redistribution weight of 0.32, and a corridor coefficient of 1.60. The environmentally constrained scenario used a suitability threshold of 0.75, zoning elasticity of 0.17, a 1,200 m ecological buffer, a capital redistribution weight of 0.46, and a corridor coefficient of 1.10. Developable-cell masks were generated from the scenario-specific suitability thresholds. Ecological buffers were applied by raster dilation. Binary zoning masks allowed reclassification of 22%, 41%, and 17% of eligible land in the equity-prioritized, growth-oriented, and environmentally constrained scenarios, respectively. Capital-priority cells were defined as locations exceeding 1.3 times the median investment value and were converted to normalized investment surfaces. The corridor coefficients multiplied the baseline construction-transition probability within designated development corridors. Land-use controls were encoded through five-class raster recoding. Each scenario layer was checked against the 2020 base plan and ID-linked stakeholder records, and parameter differences exceeding ±3% of the Round 2 medians were corrected before the 2025 simulation step.

Scenario parameters were fixed before prospective simulation and represented alternative policy bundles. Zoning elasticity controlled permissible deviation from the 2020 zoning allocation, ecological-buffer distance controlled the protected-space mask, capital redistribution controlled the proportion of discretionary investment redirected across subdistricts, and the corridor coefficient modified construction propensity within designated development corridors. Scenario comparison therefore estimated the conditional consequences of these prespecified parameter combinations. Reported performance was evaluated through criterion-specific outcomes, Pareto membership, uncertainty intervals, and rank stability across alternative weighting specifications.

Variables used to construct a scenario were recorded separately from variables used for post-simulation evaluation. Capital redistribution weight, zoning elasticity, ecological-buffer distance, and corridor coefficient were scenario inputs. Service accessibility, affordable-housing benefit, distributive-equity scores, ecological compliance, zoning compliance, benefit-distribution inequality, shock recovery, and feasibility were recalculated from realized annual outputs. Feedback variables entered the following simulation year only; an outcome calculated for year could modify allocation in year but could not alter the transition that had already generated the year- outcome. This one-year lag is important because it removes same-cycle algebraic feedback.

2.4 Institutional, behavioural, and shock modelling

Three governance strategies guided public investment allocation: centralized top-down, multi-level coordination, and decentralized subdistrict prioritization. Each generated an annual matrix based on projected population and governance-specific weights: 0.45 (centralized), 0.33 (multi-level), 0.25 (decentralized). Investment totals were derived from 2016 to 2023 audited budgets. Annual equity feedback operated with a one-year lag. Subdistricts falling below the prespecified distributive-equity target of 0.55 in year received a 20% increase in the discretionary component of year- investment, whereas subdistricts exceeding 0.70 received a 10% reduction; intermediate values retained the existing allocation. Total annual investment remained fixed at the scenario-specific fiscal envelope, so redistribution altered spatial allocation without increasing aggregate expenditure. The 2025 allocation used observed 2020 baseline deficits, and subsequent allocations used only the immediately preceding simulated year.

Household-location parameters were estimated from geocoded residential transactions and observed relocation records for 2020–2023. Housing-cost ratio, transit accessibility, and service coverage were normalized to 0–1 and entered into a conditional logit model of observed destination choice. Records from 2020 to 2022 formed the calibration sample, and 2023 records formed the temporal validation sample. Annual relocation probabilities were calculated from the fitted coefficients, and relocation occurred only when the destination probability exceeded the current-location probability and the destination cell satisfied zoning and ecological restrictions. Household coefficients were varied by ±10% during uncertainty analysis.

Developer parameters were estimated from parcel transactions, construction costs, resale values, and observed land conversion during 2018–2022. Expected resale margin, development cost, zoning eligibility, and terrain feasibility were entered into a binary logistic model of parcel conversion. Records from 2018 to 2020 were used for calibration, and records from 2021 to 2022 were reserved for validation. The payoff threshold of 0.62 was selected from candidate cutoffs between 0.50 and 0.75 in increments of 0.01 by maximizing the Youden index in the validation data. Eligible parcels were sampled probabilistically according to their fitted conversion probabilities after ecological and zoning exclusions had been applied.

Three policy shocks were applied in 2029 for 3 years. First, flood risk blocked development in ≥10% of cells flagged by 2022 Pearl River Delta overlays. Second, fiscal contraction reduced investment by 30% across all subdistricts. Third, zoning lockdown removed industrial-to-mixed-use transitions and penalized developer payoff by 0.2. Impacts were measured as percentage change in spatial equity, ecological compliance, and zoning mismatch, aggregated at subdistrict level and compared to pre-shock baselines. All simulations were synchronized annually, with outputs from governance, behavioral, and shock models fed forward as constraints for the next year. Data were regridded to the 100 m mesh and timestamped. Governance-adjusted equity influenced behavioral investment, which then shaped land transitions, conditional on shock exclusions. Fixed precedence rules governed coupling, with overrides allowed only for equity breaches or triggered shocks.

Each annual iteration followed a fixed update order. Scenario-specific ecological masks and shock restrictions first defined eligible or penalized cells; governance weights then updated public investment, household probabilities updated residential demand, and developer probabilities updated parcel-conversion demand. FLUS resolved competition among eligible land classes, after which procedural equity, distributive equity, ecological compliance, and zoning mismatch were recalculated. The recalculated year- equity state modified year- investment allocation according to the lagged feedback rule. Flood, fiscal-contraction, and zoning-lockdown rules superseded the standard allocation sequence during 2029–2031.

2.5 Equity validation and optimization

Equity outcomes were calculated separately for the BAU, equity-prioritized, growth-oriented, and environmentally constrained simulations. Procedural equity was calculated annually from simulated participation activity rather than participant count alone. The 28-member stakeholder pool remained fixed, but seven jurisdictional review panels generated up to 28 valid participant-response opportunities during each annual scenario-review cycle. For scenario in year , , where is the normalized frequency of valid annual stakeholder contributions and is Shannon diversity normalized across the four stakeholder categories. Annual contribution realization was based on category-specific participation frequencies derived from the 2018–2023 coordination records and updated by governance coordination frequency and implementation delay. was calculated after each annual land-transition step and did not enter the same-year construction-transition equation. Distributive equity was calculated as, , where is normalized service accessibility, is normalized green-space access, and is normalized affordable-housing benefit for scenario and year . Each component was scaled to 0–1 before aggregation. All components were normalized to 0–1 before aggregation. Both indices were mapped in ArcGIS Pro (Esri, United States) using subdistrict zonal statistics. For each simulation year, , , and the cumulative benefit index were co-registered to the 100 × 100 m grid. The cell-level equity gap was calculated as , where denoted the subdistrict containing cell . Cells in the highest 20% of values were classified as persistent equity-deficit locations. Getis-Ord Gi* analysis identified statistically concentrated high-gap clusters across the annual outputs.

A separate post-simulation validation set was used to assess whether scenario rankings persisted when variables closely related to scenario construction were removed. The independent set comprised zoning compliance, benefit-distribution Gini coefficient, forest-and-wetland retention, construction-land contiguity, institutional-feasibility score, and post-shock recovery time. None of these six criteria entered the construction-transition probability, capital-redistribution rule, or household destination equation. Scenario ordering was recalculated with equal weights across the six independent criteria and compared with the primary ranking using Spearman rank correlation and Pareto membership.

Scenario trade-offs were summarized using mean equity, ecological compliance, and fiscal efficiency. Round 2 Delphi median importance scores of 4.6, 4.0, and 2.9 were normalized to weights of 0.40, 0.35, and 0.25, respectively. The composite score was calculated as , where , , and denoted scenario-level equity, ecological compliance, and fiscal efficiency. Pareto analysis retained every non-dominated scenario. Dominance required equal or higher performance on all three objectives and higher performance on at least one objective. Equal weighting and ±0.05 perturbations of the primary weights were used to assess rank stability, with the weights rescaled to sum to 1.00. Spearman correlations of at least 0.90 between the primary and alternative rankings were interpreted as stable.

2.6 Model calibration, ablation, uncertainty, and feasibility

Historical calibration used land-use transitions from 2005 to 2010 as the training interval, transitions from 2010 to 2015 as the parameter-selection interval, and transitions from 2015 to 2020 as the out-of-sample validation interval. Neighbourhood coefficients were searched from 0.10 to 1.00 in increments of 0.10, inertia coefficients from 0.50 to 0.95 in increments of 0.05, and conversion-cost multipliers from 0.80 to 1.20 in increments of 0.05. The retained FLUS parameter set maximized the mean of Kappa, macro-F1, and FoM while maintaining construction-area prediction error below 7%. All suitability, resistance, neighborhood, inertia, and conversion-cost parameters were frozen after the 2010–2015 parameter-selection stage. The 2015–2020 interval contributed only final validation outcomes and was excluded from weight selection, threshold selection, coefficient fitting, and model revision.

The calibrated parameter set was held constant across BAU, equity-prioritized, growth-oriented, and environmentally constrained simulations, except for the scenario-specific suitability threshold, ecological buffer, zoning elasticity, capital redistribution weight, and corridor coefficient defined in Sections 2.2 and 2.3. Uncertainty was estimated from 500 Monte Carlo runs per scenario. Land-conversion costs were perturbed by ±10%, ecological-resistance weights by ±0.15, and population projections by ±5%. Corresponding BAU and alternative-scenario runs used identical random-number seeds and parameter draws. Cell-level uncertainty was calculated from the frequency and entropy of simulated land-class outcomes. Scenario uncertainty was summarized using the 2.5th, 50th, and 97.5th percentiles of paired scenario-minus-BAU differences within the same subdistrict, year, and seed.

Component attribution was evaluated using four simplified specifications against the full SPECS configuration. The spatial–ecological core retained suitability, zoning, ecological resistance, and FLUS transitions while holding stakeholder, behavioral, governance, and equity-feedback components inactive. The participatory specification added stakeholder-derived zoning, buffer, capital, and corridor parameters to the spatial–ecological core. The coupled-behavior specification additionally activated household relocation, developer conversion, and governance allocation while retaining fixed annual investment shares. Full SPECS added the lagged equity-feedback mechanism. Each specification used the same historical parameters, scenario inputs, 500 Monte Carlo draws, random-number seeds, population perturbations, cost perturbations, and ecological-resistance perturbations. Incremental component effects were calculated from paired differences in suitability, mismatch, ecological encroachment, accessibility, distributive equity, zoning compliance, feasibility, and the independent six-criterion validation score.

Scenario robustness was calculated as the proportion of paired Monte Carlo runs in which the prespecified directional improvement over BAU was retained. The outcome-dispersion index was calculated from normalized cell-level variation across the 500 runs, with classified as volatile. Institutional feasibility was calculated as , where represented regulatory alignment, represented budget capacity, represented implementation-timeline readiness, and represented participatory reach. Each component was normalized to 0–1. Scenarios with were classified as implementable. Rollout simulations applied policy delays of 0, 1, 2, and 3 years to quantify the effect of implementation lag on projected equity outcomes.

A circularity sensitivity analysis repeated the scenario ranking with equity feedback disabled, capital redistribution fixed at the BAU level, and all equity measures calculated only after land allocation. A second specification excluded accessibility and affordable-housing benefit from the ranking criterion set. A scenario effect was classified as structurally robust when its direction was retained in at least 90% of paired Monte Carlo runs and remained in the Pareto set under both sensitivity specifications.

2.7 Statistical analysis

All analyses were conducted using R (R Computing, Austria). The 100 m cell was the computational unit, and outputs were aggregated to the 18 subdistricts. BAU served as the reference trajectory for the equity-prioritized, growth-oriented, and environmentally constrained scenarios. Scenario differences were calculated within the same subdistrict, simulation year, and Monte Carlo seed. The terms “intervention” and “control” were not applied because the four configurations were simulated trajectories generated from the same calibrated historical baseline. Projected scenario outcomes were summarized using the median scenario-minus-BAU difference, interquartile range, 95% simulation uncertainty interval, percentage change from BAU, and empirical probability of improvement. The empirical probability was calculated as the proportion of paired runs in which an alternative scenario exceeded BAU for beneficial outcomes or remained below BAU for adverse outcomes. Independent-samples t-tests, one-way ANOVA, repeated-measures ANOVA, Tukey HSD tests, and conventional null-hypothesis p-values were not applied to the projected outputs because the 18 subdistricts represented the complete study area and were not randomly sampled or assigned units. Associations among suitability, ecological resistance, public investment, accessibility, land conversion, and equity were summarized using Spearman coefficients with percentile intervals derived from the paired Monte Carlo ensemble.

Historical validation used the held-out 2015–2020 interval. Classification performance was evaluated using AUC, class-specific F1 scores, Kappa, overall accuracy, Figure of Merit, and allocation disagreement. Continuous prediction error was evaluated using RMSE for construction area and MAPE for population. Scenario ranking was based on Pareto dominance, composite percentiles, 95% simulation uncertainty intervals, empirical improvement probabilities, and rank stability under alternative parameter and weighting specifications. Associations specified under H3 were treated as model-output associations and were additionally recalculated in the no-equity-feedback ablation to assess dependence on the embedded allocation rules. Robust association required retention of the same coefficient direction in the full and ablated specifications and exclusion of zero from the corresponding 95% Monte Carlo percentile interval. Scenario rankings from the primary criteria were also compared with rankings from the independent six-criterion validation set using Spearman rank correlation. Projected scenario inference was based on paired simulation uncertainty intervals, empirical improvement probabilities, Pareto membership, and rank stability. Significance thresholds were restricted to procedures with an explicit sampling reference distribution, including Getis–Ord Gi* cluster detection, for which two-sided was applied after false-discovery-rate correction.

3 Results

3.1 Baseline spatial and functional conditions

Baseline spatial conditions establish the common 2020 starting configuration used to evaluate scenario differences under RQ1 and the multidimensional performance comparison specified in H1 and H2.

Development pressure was concentrated in the administrative core and logistics–innovation corridor before scenario simulation. Across the 18 subdistricts, median population density was 1,445 residents/km², median construction-land share was 31.8%, and median road density was 3.21 km/km². Nansha Central recorded the highest construction-land share at 42.6%, while the ecological-belt subdistricts retained the largest forest and wetland share at 61.4%. Baseline service accessibility remained uneven, ranging from 0.43 in peripheral ecological areas to 0.71 in the administrative core. The median suitability score was 0.56, and 38.7% of cells exceeded the BAU suitability threshold of 0.65. Ecological-redline overlap affected 9.6% of existing construction cells, while median spatial-functional mismatch was 0.184. Procedural equity and distributive equity were 0.53 and 0.49, respectively, with a benefit-distribution Gini coefficient of 0.212 (Table 2). These conditions defined the ecological, accessibility, and equity deficits against which the four 2025–2035 trajectories were evaluated.

3.2 Scenario-based land-use and ecological outcomes

Land suitability, spatial-functional mismatch, and ecological encroachment answer RQ1, while changes produced by scenario-specific suitability thresholds, ecological buffers, and zoning elasticity test H1 and H2.

Construction remained highest under the growth-oriented trajectory, which added a median 3.18 km² per subdistrict by 2035, compared with 2.46 km² under equity prioritization and 1.72 km² under environmental constraint. Suitable-cell allocation distinguished planned expansion from unconstrained quantity: 88.6% of equity-prioritized construction and 91.8% of environmentally constrained construction occurred in cells meeting their scenario thresholds, compared with 72.4% under growth orientation. The equity-prioritized trajectory increased composite service accessibility from 0.58 to 0.69 while reducing median mismatch from 0.184 to 0.112. Its median paired improvement over BAU was 0.104 for suitability and −0.068 for mismatch, with empirical improvement probabilities of 0.946 and 0.932, respectively (Supplementary Table S1). Development was therefore redirected toward accessible and functionally aligned cells without reproducing the higher conversion volume of the growth-oriented trajectory.

Ecological restrictions produced the clearest separation among the four trajectories. Median total construction overlap with ecological redlines reached 11.7% under growth orientation, remained at 9.4% under BAU, declined to 5.3% under equity prioritization, and fell to 3.2% under environmental constraint. No new construction entered ecological redlines or high-resistance cells under BAU, equity prioritization, or environmental constraint because these areas remained hard exclusions; existing 2020 overlap accounted for the residual redline values. Under the growth-oriented soft-constraint specification, 2.6% of new construction entered ecological-redline cells and 9.8% entered high-resistance cells. Annual transition and ecological-area accounting reconciled these conversions with beginning and ending land stocks (Supplementary Tables S8–S11). Wetland and forest retention reached 94.6% and 91.7%, respectively, under environmental constraint, while agricultural-land loss remained lowest at 8.7%. Equity prioritization and environmental constraint reduced redline overlap relative to BAU by 4.1 and 6.2 percentage points, respectively (Supplementary Table S2). Ecological exclusions also produced higher allocation agreement and lower spatial entropy (Figure 4).

FIGURE 4

3.3 Equity, behavioural, governance, and shock outcomes

Service allocation, household relocation, developer conversion, and governance feedback answer RQ2, while their relationships with procedural equity, distributive equity, and suitability test H3.

Equity prioritization produced the largest increase in participation activity and benefit distribution. Procedural equity increased from 0.53 to 0.71 as median valid contributions rose from 17 to 23 of 28 annual response opportunities and normalized stakeholder-category diversity increased from 0.45 to 0.60. Distributive equity increased from 0.49 to 0.65, while the benefit-distribution Gini coefficient declined to 0.139. Median paired improvements over BAU were 0.142 for procedural equity and 0.126 for distributive equity, with empirical improvement probabilities of 0.968 and 0.956. Residence in high-suitability cells increased by 16.8 percentage points, annual relocation declined from 7.2% to 3.6%, and developer conversion retained a median payoff of 0.84 with 91.3% zoning compliance (Supplementary Table S3). Distributive equity remained associated with accessibility in the full specification, ρ = 0.76, 95% interval [0.61, 0.86], and the no-equity-feedback ablation, ρ = 0.68 [0.52, 0.80]. Its association with suitability similarly remained positive at ρ = 0.71 [0.54, 0.83] and ρ = 0.63 [0.47, 0.76], respectively.

Governance and shock responses differed by scenario configuration. Equity-based allocation redirected 14.6% of annual investment toward subdistricts below the distributive-equity target, compared with no redistribution under BAU. During the 2029–2031 flood shock, environmentally constrained development retained 89.8% of its pre-shock ecological-compliance score, while growth orientation retained 72.6%. Fiscal contraction reduced service-access gains by 5.8% under equity prioritization and 12.7% under growth orientation. Zoning lockdown reduced developer conversion by 9.4% and 18.3%, respectively. Equity prioritization maintained a median distributive-equity gain of 0.091 during the combined shock period, whereas growth orientation fell 0.017 below BAU (Table 3). Governance feedback therefore preserved access and equity most effectively where investment redistribution and ecological restrictions were already encoded, while growth-dependent conversion remained sensitive to capital reduction and transition prohibitions (Supplementary Figure S2).

3.4 Scenario optimization, robustness, and trade-offs

Pareto dominance, composite performance, weighting sensitivity, and shock stability answer RQ3 and test whether multi-objective configurations retain the performance and robustness expected under H4.

The equity-prioritized trajectory achieved the highest composite score at 0.704, followed by environmental constraint at 0.687, BAU at 0.546, and growth orientation at 0.531. Equity prioritization combined an equity component of 0.286, ecological component of 0.238, and fiscal-efficiency component of 0.180. Environmental constraint produced the strongest ecological component at 0.272 but a lower fiscal contribution of 0.155. Both scenarios were non-dominated, while BAU and growth orientation were excluded from the Pareto set because they were simultaneously weaker on equity and ecological compliance. The median composite difference from BAU was 0.158 for equity prioritization and 0.141 for environmental constraint, with empirical improvement probabilities of 0.964 and 0.951 (Supplementary Table S4). The alternative rankings therefore reflected distinct equity–ecology trade-offs instead of a single universally dominant trajectory.

Scenario ordering remained stable under the prespecified perturbations. Equity prioritization retained first rank in 84.2% of the 500 ensemble runs and remained within the Pareto set in 96.4%. Environmental constraint retained Pareto membership in 97.8% but ranked first in 14.6%, mainly when ecological weights increased. Equal weighting produced a rank correlation of 0.96 with the primary specification, while ±0.05 weight perturbations produced correlations of 0.92–0.98. The equity-prioritized outcome-dispersion index was 0.912, compared with 0.692 for BAU and 0.664 for growth orientation. Its median rank shift was 0.41 positions, and its expected improvement survived synchronized suitability, resistance, cost, population, flood, fiscal, and zoning perturbations in 91.6% of paired runs (Supplementary Table S5). Stability therefore depended on combining equity feedback, ecological safeguards, and coordinated investment within the same configuration (Figure 5).

FIGURE 5

Component ablation showed progressive performance gains rather than an improvement attributable solely to full-model complexity. For the equity-prioritized trajectory, the composite score increased from 0.612 with the spatial–ecological core to 0.645 after participatory parameterization, 0.676 after behavioral and governance coupling, and 0.704 under full SPECS. Independent evaluation using zoning compliance, benefit-distribution inequality, forest–wetland retention, construction-land contiguity, institutional feasibility, and shock recovery produced scores of 0.681 for equity prioritization, 0.674 for environmental constraint, 0.548 for BAU, and 0.526 for growth orientation, with primary and independent rankings correlated at ρ = 0.94. Removing equity feedback and fixing capital redistribution at the BAU level reduced the equity-prioritized composite advantage over BAU from 0.158 to 0.103, but Pareto membership remained at 91.8%. Excluding accessibility and affordable-housing benefit from ranking retained the equity-prioritized trajectory in the Pareto set in 94.2% of paired runs (Supplementary Table S5).

3.5 Historical validation and institutional feasibility

Held-out historical agreement assesses the calibrated transition model, while feasibility and implementation-delay outputs quantify whether the preferred scenarios remain administratively executable.

The 2015–2020 out-of-sample interval produced an overall land-class accuracy of 91.6%, Kappa of 0.872, macro-F1 of 0.861, and Figure of Merit of 0.743. Construction-class discrimination reached an AUC of 0.894, with F1 of 0.887 and allocation disagreement of 0.061. Construction-area RMSE was 2.81 km², while population MAPE was 5.67%. Wetland and forest classes retained F1 values of 0.842 and 0.874, indicating that the model reproduced ecological persistence as well as construction conversion. The developer conversion model produced a validation AUC of 0.812, sensitivity of 0.781, and specificity of 0.754 at the 0.62 threshold. Household destination prediction reached a top-choice accuracy of 0.736 and mean rank probability of 0.684 (Supplementary Table S6). These metrics supported application of the calibrated parameter set to the four prospective trajectories without treating the projected outputs as observed future conditions. These validation estimates were generated from the parameter set frozen after 2010–2015 selection, with no coefficient, threshold, weight, or transition-rule adjustment using the 2015–2020 observations.

Institutional feasibility was highest for equity prioritization at 0.846 and environmental constraint at 0.812, exceeding the 0.75 implementation threshold. Equity prioritization combined regulatory alignment of 0.88, budget capacity of 0.81, implementation readiness of 0.82, and participatory reach of 0.86. Growth orientation scored 0.681 because lower regulatory alignment and ecological compatibility offset its higher budget capacity. A two-year implementation delay reduced projected equity gains by 11.3% under equity prioritization and 13.8% under environmental constraint, while a three-year delay produced reductions of 16.7% and 19.2%. BAU remained below the feasibility threshold at 0.694 (Supplementary Table S7). Feasibility therefore reflected alignment among regulation, funding, implementation timing, and stakeholder reach, while delayed rollout reduced the gains generated by the scenario parameters (Supplementary Figure S3).

4 Discussion

The equity-prioritized trajectory achieved the highest composite performance, strongest accessibility and equity gains, and the most stable first-rank position, while the environmentally constrained trajectory produced the lowest ecological-redline overlap and strongest forest, wetland, and agricultural-land retention. New construction within ecological redlines and high-resistance land occurred only under the growth-oriented soft-constraint specification. These findings answered RQ1 by demonstrating clear scenario differences in suitability, mismatch, encroachment, accessibility, equity, and feasibility. RQ2 was answered by the links between governance allocation, household relocation, developer conversion, ecological restrictions, and subdistrict outcomes. RQ3 was answered by the retention of scenario rankings under synchronized parameter and policy perturbations. H1 was supported because equity prioritization and environmental constraint exceeded BAU and growth orientation on multidimensional performance. H2 was supported by lower ecological-redline overlap and spatial-functional mismatch under the constrained trajectories, while hard ecological exclusions prevented new high-resistance conversion. H3 was supported by positive associations of distributive equity with accessibility and suitability that retained their direction after equity feedback was removed. H4 was supported because both preferred trajectories retained Pareto membership and stable rankings across the ensemble runs (Chen, 2025; de Feijter, 2023; Gao et al., 2021; Gao et al., 2024; Mahmoud et al., 2021; Wang, 2020; Wei et al., 2024; Wu et al., 2024; Zhao et al., 2022; Zheng et al., 2024; Lai et al., 2022; Lin et al., 2012; Liu et al., 2022; Liu et al., 2025; Marondedze et al., 2024).

The stronger suitability alignment and lower ecological encroachment indicate that the constrained scenarios redirected development toward locations with higher physical capacity and lower environmental conflict. Hu et al. (2024) similarly found that cooperative spatial allocation reduced edge sprawl by redistributing growth toward more suitable nodes, while Zheng and Xian (2023) showed that policy-connectivity controls shifted expansion toward ecologically viable areas in Nansha. The present study extends this logic by embedding ecological resistance, suitability thresholds, and zoning mismatch within the same annual transition process. Development probabilities were reduced where slope, habitat sensitivity, or functional misalignment increased, while ecological buffers removed critical cells from the conversion set. This combination explains why improved spatial alignment occurred without equivalent increases in ecological pressure and clarifies how policy feedback and biophysical constraints jointly shaped the projected land-use pattern.

The positive association between suitability and equity indicates that spatially appropriate development also improved the distribution of services and planning benefits. Chen et al. (2025) reported comparable suitability–equity synergies under climate-sensitive land-use scenarios, while Li L. et al. (2023) found that coordinated digital infrastructure, green systems, and service accessibility reduced spatial inequality alongside efficiency gains. In the present model, this relationship emerged because household relocation, developer selection, and public-investment allocation were constrained by suitability, accessibility, and zoning conditions. High-cost, poorly connected, or ecologically sensitive cells consequently received lower transition probabilities, while accessible and compliant locations attracted both development and service investment. The resulting spatial–equity coupling links physical capacity with procedural participation and distributive benefit allocation, explaining why improved suitability was accompanied by stronger service parity and lower spatial exclusion.

Scenario-specific ecological controls operated within the same spatial decision sequence: hard exclusions prevented new conversion in protected and high-resistance cells under BAU, equity prioritization, and environmental constraint, while soft penalties allowed limited conversion under growth orientation. Wang et al. (2023) similarly associated smart-equity governance with stronger resilience and wider service coverage, while Liu et al. (2022) showed that ecological redline constraints reduced expansion pressure and improved functional compactness in ESP–FLUS simulations. The present results suggest that ecological protection did not require isolating conservation from accessibility planning. Suitability thresholds, zoning compliance, and service proximity jointly narrowed the developable set, reducing spatial exclusion and concentrating growth where infrastructure could support additional demand. This mechanism explains why ecological compliance, procedural equity, and service access improved within the same scenarios.

Compact, zoning-compliant development remained feasible because infill locations combined lower transport burdens, stronger service proximity, and acceptable developer returns. Wang et al. (2021) reported similar advantages for compact infill strategies, while Xu (2025) linked land-finance capacity in Nansha to transit-oriented infrastructure investment. Zhang and Liao (2024) further showed that compact expansion can reduce fragmentation and improve service access, and Shen et al. (2023) identified zoning mismatch as a source of governance inefficiency and weaker equity. In the present study, fiscal allocation and parcel attractiveness were linked before conversion occurred, allowing infrastructure investment to improve the viability of compliant sites while reducing pressure for peripheral expansion. Suitability-aligned zoning therefore supported compactness through selective redevelopment and service coordination, with spatial form, institutional capacity, and distributive outcomes remaining interdependent.

4.1 Theoretical and practical implications

The SPECS model extends land-use simulation by placing procedural equity, distributive equity, governance allocation, household relocation, and developer conversion within the annual transition sequence. This structure distinguishes stakeholder participation from benefit distribution and links both to subsequent land and investment decisions. For Nansha, the equity-prioritized trajectory provides the strongest balance under the prespecified criteria where service inequality and spatial mismatch are the principal planning concerns, while the environmentally constrained trajectory is more appropriate for subdistricts exposed to wetlands, redlines, flood risk, or high ecological resistance. Implementation can begin with annual identification of subdistricts below the distributive-equity target, followed by targeted infrastructure allocation, maintenance of the scenario-specific ecological buffers, and monitoring of conversion within suitable cells. These actions align with Sustainable Development Goal 10 through reduced spatial inequality, Sustainable Development Goal 11 through accessible and resilient urban development, Sustainable Development Goal 13 through climate-sensitive land allocation, and Sustainable Development Goal 15 through wetland, forest, and ecological-redline protection.

Ablation and independent-criterion analyses further showed that the principal scenario ordering persisted after equity feedback was removed and overlapping equity criteria were excluded, indicating that the reported advantages were not generated solely by the full model structure.

4.2 Limitations and future research

The 2015–2020 held-out interval assessed historical land-class prediction but could not verify the projected 2025–2035 trajectories. The future outputs remain conditional on the demographic, fiscal, behavioural, ecological, and governance assumptions encoded in the model. Scenario rankings are also conditional on the specified policy bundles, evaluation criteria, and weighting structure, although their ordering remained stable across the ablation and sensitivity specifications.

Household relocation and developer conversion were calibrated from observed transaction and conversion records, although informal occupancy, tenure variation, mortgage conditions, and strategic land banking may alter future behaviour. The Delphi parameters may also reflect the institutional composition of the 28 participants, while the 100 m grid cannot represent parcel-level differences in ownership, building condition, or redevelopment cost. Future research should compare annual projections with newly observed land-use and service data, update agent coefficients as market conditions change, test the framework in another Greater Bay Area district, and conduct participatory assessment in subdistricts with persistent equity deficits.

4.3 Conclusion

Under the prespecified evaluation criteria, the equity-prioritized trajectory achieved the strongest overall balance across accessibility, procedural equity, distributive equity, ecological compliance, fiscal efficiency, and rank stability, while the environmentally constrained trajectory produced the lowest ecological encroachment and highest land-retention performance. BAU and growth-oriented development remained weaker where expansion pressure increased spatial-functional mismatch, ecological conflict, and unequal benefit distribution. The scenario comparison therefore indicates that a good urban area in Nansha depends on coordinating suitability thresholds, ecological buffers, household relocation, developer incentives, and equity-based investment allocation within the same planning sequence. Policy implementation should prioritize annual identification of equity-deficit subdistricts, targeted service and infrastructure investment, protection of ecological redlines and high-resistance land, and development approval linked to suitability, accessibility, and zoning compliance.

References

• 1

  AdadeD.de VriesW. T. (2025). A systematic review of digital twins’ potential for citizen participation and influence in land use agenda-setting. Discov. Sustain.6, 354. Article 354. 10.1007/s43621-025-01204-x

  • CrossRef
  • Google Scholar

• 2

  AghalooK.SharifiA. (2025). Balancing priorities for a sustainable future in cities: land use change and urban ecosystem service dynamics. J. Environ. Manag.382, 125460. Article 125460. 10.1016/j.jenvman.2025.125460

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 3

  ChenF. (2025). Spatial design professional-led multi-stakeholders and community participation in co-creating a walkable and inclusive public space. J. Asian Archit. Build. Eng.25, 1–17.

  • Google Scholar

• 4

  ChenF. (2026). Spatial design professional-led multi-stakeholders and community participation in co-creating a walkable and inclusive public space. J. Asian Archit. Build. Eng.25 (3), 1961–1977. 10.1080/13467581.2025.2498723

  • CrossRef
  • Google Scholar

• 5

  ChenG.ZhangD.ZhaoJ.ZhangL. (2025). Land use and climate change-based multi-scenario simulation of ecosystem service trade-offs/synergies: a case study of the central Yunnan urban agglomeration, China. PLOS ONE20 (6), e0324015. 10.1371/journal.pone.0324015

  • CrossRef
  • Google Scholar

• 6

  ChengH.ShawD. (2021). Polycentric development and the formation of edge urban areas in China's mega city regions: case study of nansha, guangzhou. Int. J. Urban Regional Res., 45(6), 1009–1027. 10.1111/1468-2427.13045

  • CrossRef
  • Google Scholar

• 7

  ChengH.LiuY.HeS.ShawD. (2017). From development zones to edge urban areas in China: a case study of nansha, Guangzhou City. Cities, 71, 110–122. 10.1016/j.cities.2017.07.015

  • CrossRef
  • Google Scholar

• 8

  de FeijterF. J. (2023). Trust in circular design: active stakeholder participation in Chinese and Dutch housing retrofit projects. Build. Res. and Inf.51 (1), 105–118. 10.1080/09613218.2022.2121905

  • CrossRef
  • Google Scholar

• 9

  FeitosaF. O.WolfJ. H.MarquesJ. L. (2024). Operationalizing spatial justice in urban planning: bridging theory with practice. Urban Res. and Pract.17 (5), 720–736. 10.1080/17535069.2024.2341254

  • CrossRef
  • Google Scholar

• 10

  GaoY.JiangP.LiM. (2021). Spatial planning zoning based on land-type mapping: a case study in Changzhou City, Eastern China. J. Land Use Sci.16 (5-6), 498–521. 10.1080/1747423X.2021.2011968

  • CrossRef
  • Google Scholar

• 11

  GaoY.WangZ.ChaiJ.ZhangH. (2024). Spatiotemporal mismatch of land use functions and land use efficiencies and their influencing factors: a case study in the middle reaches of the Yangtze River, China. J. Geogr. Sci.34 (1), 62–88. 10.1007/s11442-024-2195-1

  • CrossRef
  • Google Scholar

• 12

  GaoJ.GongJ.SchwarzN.SliuzasR. (2026). A hierarchical cross-departmental agent-based approach to explore the impacts of policy interplay on land use dynamics. Comput. Environ. Urban Syst.123, 102355. Article 102355. 10.1016/j.compenvurbsys.2025.102355

  • CrossRef
  • Google Scholar

• 13

  HeP.WangQ.-C.WangH.XueJ.BaoH. X.ShenG. Q.et al (2026). From participation to partnership: a systematic review of public engagement in sustainable urban planning. Environ. Impact Assess. Rev.117, 108187. Article 108187. 10.1016/j.eiar.2025.108187

  • CrossRef
  • Google Scholar

• 14

  HuJ.LiuD.ZhengX. (2024). Research on multi-scenario simulation of urban expansion for beijing–tianjin–hebei Region considering multilevel urban flows. Land13 (11), 1830. 10.3390/land13111830

  • CrossRef
  • Google Scholar

• 15

  HuangJ.LiuX.LinY.GeL. (2024). Spatio-temporal pattern analysis of coastal zone in Nansha based on remote sensing technology. Remote Sens. Appl. Soc. Environ.35, 101222. 10.1016/j.rsase.2024.101222

  • CrossRef
  • Google Scholar

• 16

  HuangH.XuS.FangK. (2025). Exploring the coupling of land development intensity and green economy development in the Pearl River Delta: patterns, challenges, and strategic pathways. Land14 (1), 105. 10.3390/land14010105

  • CrossRef
  • Google Scholar

• 17

  LaiZ.ChenC.ChenJ.WuZ.WangF.LiS. (2022). Multi-scenario simulation of land-use change and delineation of urban growth boundaries in county area: a case study of xinxing county, Guangdong province. Land11 (9), 1598. 10.3390/land11091598

  • CrossRef
  • Google Scholar

• 18

  LiL.HuangX.YangH. (2023). Scenario-based urban growth simulation by incorporating ecological-agricultural-urban suitability into a future land use simulation model. Cities, 137, 104334. 10.1016/j.cities.2023.104334

  • CrossRef
  • Google Scholar

• 19

  LiX.ZhuJ.LiuT.YinX.YaoJ.JiangH.et al (2023). Quota and space allocations of new urban land supported by urban growth simulations: a case study of guangzhou City, China. Land12 (6), 1262. 10.3390/land12061262

  • CrossRef
  • Google Scholar

• 20

  LiJ.TianG.WuZ.ZhaoQ.MengY.XiaQ. (2024). Urban sustainable development: unraveling regional land development intensity dynamics and future trends in China. Sustain. Dev.32 (5), 4413–4427. 10.1002/sd.2917

  • CrossRef
  • Google Scholar

• 21

  LiJ.CaiX.CuiX.ZhaiJ.ChenL.WangG.et al (2026). From evaluation to planning: integrating quantity–quality in the spatial equity assessment and optimization of urban green spaces. Humanit. Soc. Sci. Commun.13, 1068. Article 1068. 10.1057/s41599-026-07357-y

  • CrossRef
  • Google Scholar

• 22

  LinW.LuQ.ShangY.LeH.LiuJ.MaY.et al (2012). “New urban district design in a transformation time in China: the case of Nansha new Urban District, Guangzhou,” in Civil Engineering and Urban Planning 2012 (American Society of Civil Engineers), 411–415. 10.1061/9780784412435.074

  • CrossRef
  • Google Scholar

• 23

  LiuS.YuQ.WeiC. (2019). Spatial-temporal dynamic analysis of land use and landscape pattern in Guangzhou, China: exploring the driving forces from an urban sustainability perspective. Sustainability11 (23), 6675. 10.3390/su11236675

  • CrossRef
  • Google Scholar

• 24

  LiuX.WeiM.LiZ.ZengJ. (2022). Multi-scenario simulation of urban growth boundaries with an ESP-FLUS model: a case study of the Min Delta region, China. Ecol. Indic.135, 108538. 10.1016/j.ecolind.2022.108538

  • CrossRef
  • Google Scholar

• 25

  LiuJ.HuC.KangX.ChenF. (2023). A loosely coupled model for simulating and predicting land use changes. Land12 (1), 189. 10.3390/land12010189

  • CrossRef
  • Google Scholar

• 26

  LiuC.ZhongM.HeM.WangX.GanH.CaoP.et al (2025). Research on Co-Creation of community public cultural spaces through generative dynamic workflows. Systems13 (5), 316. 10.3390/systems13050316

  • CrossRef
  • Google Scholar

• 27

  LiuX.ZhangR.GuoJ.XuH.MiaoY.NiuF.et al (2025). Analysis of the spatiotemporal dynamics of grassland carbon sinks in Xinjiang via the improved CASA model. Ecol. Indic.170, 113062. 10.1016/j.ecolind.2024.113062

  • CrossRef
  • Google Scholar

• 28

  LuJ.JiaoS.ChenX. (2026). Spatial morphology optimization for reconciling urban expansion with ecological integrity based on a multi-level ecological network framework. J. Geogr. Sci.36, 399–420. 10.1007/s11442-026-2453-5

  • CrossRef
  • Google Scholar

• 29

  MaheshwariA.ChopraS. S. (2025). Actionable insights for equitable urban sustainability transition through building-level assessment of sustainable development goals. Npj Urban Sustain.5, 63. 10.1038/s42949-025-00238-4

  • CrossRef
  • Google Scholar

• 30

  MahmoudI. H.MorelloE.LudlowD.SalviaG. (2021). Co-creation pathways to inform shared governance of Urban living labs in practice: lessons from three European projects. Front. Sustain. Cities, 3, 690458. 10.3389/frsc.2021.690458

  • CrossRef
  • Google Scholar

• 31

  MarondedzeA. K.MutangaO.ChoM. A. (2024). Promoting inclusion in urban land use planning using participatory geographic information system (PGIS) techniques: a systematic review. J. Environ. Manag.370, 123099. 10.1016/j.jenvman.2024.123099

  • CrossRef
  • Google Scholar

• 32

  Molinero-ParejoR.Aguilera-BenaventeF.Gómez-DelgadoM.ShurupovN. (2023). Combining a land parcel cellular automata (LP-CA) model with participatory approaches in the simulation of disruptive future scenarios of urban land use change. Comput. Environ. Urban Syst.99, 101895. 10.1016/j.compenvurbsys.2022.101895

  • CrossRef
  • Google Scholar

• 33

  NourianP.AzadiS.BaiN.de AndradeB.Abu ZaidN.RezvaniS.et al (2024). EquiCity game: a mathematical serious game for participatory design of spatial configurations. Sci. Rep.14, 10912. 10.1038/s41598-024-61093-4

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 34

  PusinghaP.MarchantR. A.FriendR. M.HanR. (2026). Emerging lessons on how participatory scenario planning can support sustainable land management and climate resilience. Geogr. Sustain.7 (1), 100405. Article 100405. 10.1016/j.geosus.2025.100405

  • CrossRef
  • Google Scholar

• 35

  SaraswatA.PipraliaS.KumarA. (2025). Advances in urban sustainability assessment: a systematic literature review of methods, frameworks, and future directions. Front. Urban Rural Plan.3, 15. Article 15. 10.1007/s44243-025-00063-4

  • CrossRef
  • Google Scholar

• 36

  SeoB. K. (2022). Co-creation of knowledge in the urban planning context: the case of participatory planning for transitional social housing in Hong Kong. Cities, 122, 103518. 10.1016/j.cities.2021.103518

  • CrossRef
  • Google Scholar

• 37

  ShenL.ZhangL.BaoH.WongS.DuX.WeiX. (2023). An empirical Study on the mismatch phenomenon in utilizing urban land resources in China. Land12 (6), 1196. 10.3390/land12061196

  • CrossRef
  • Google Scholar

• 38

  ShiH.LiQ.ZhangS.ChenW.WangY.ShenY. (2024). Does urban construction land in China achieve sustainable allocation and utilization?Land Use Policy, 144, 107223. 10.1016/j.landusepol.2024.107223

  • CrossRef
  • Google Scholar

• 39

  SongJ.HuangB.LiR. (2018). Assessing local resilience to typhoon disasters: a case study in Nansha, Guangzhou. PLOS ONE13, e0190701. 10.1371/journal.pone.0190701

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 40

  WangM. (2020). Polycentric urban development and urban amenities: evidence from Chinese cities. Environ. Plan. B Urban Anal. City Sci.48, 239980832095120–239980832095416. 10.1177/2399808320951205

  • CrossRef
  • Google Scholar

• 41

  WangR.HeW.WuD.ZhangL.LiY. (2021). Urban land expansion simulation considering the diffusional and aggregated growth simultaneously: a case Study of Luoyang City. Sustainability13 (17), 9781. 10.3390/su13179781

  • CrossRef
  • Google Scholar

• 42

  WangZ.FuH.ZhouL. (2023). Multiple urban resilience evaluation of resource-based cities’ sustainable transformation effect. Resour. Conservation Recycl.191, 106912. 10.1016/j.resconrec.2023.106912

  • CrossRef
  • Google Scholar

• 43

  WangS.SunH.WeiL.PiP.ZengM.PanY.et al (2024). Integrated assessment of coastal subsidence in Nansha District, Guangzhou City, China: insights from SBAS-InSAR monitoring and risk evaluation. Remote Sens.16 (2), 248. 10.3390/rs16020248

  • CrossRef
  • Google Scholar

• 44

  WeiH.HanQ.MaY.JiW.FanW.LiuM.et al (2024). Multi-Scenario simulating the effects of land use change on ecosystem health for rural ecological management in the zheng–bian–luo rural area, central China. Land13 (11), 1788. 10.3390/land13111788

  • CrossRef
  • Google Scholar

• 45

  WuH.HuangY.LiuG.LuoL.LuoY. (2024). Synergizing stakeholder collaboration for value co-creation in China's prefabricated decoration diffusion: a tripartite evolutionary game perspective. Heliyon10 (15), e34775. 10.1016/j.heliyon.2024.e34775

  • CrossRef
  • Google Scholar

• 46

  XuN. (2025). Lying flat city and rat race city: chinese cities’ land development strategies. Urban Gov., 5(2), 155–168. 10.1016/j.ugj.2025.05.010

  • CrossRef
  • Google Scholar

• 47

  XuX.DingD.LiuX. (2024). A three-dimensional future land use simulation (FLUS-3D) model for simulating the 3D urban dynamics under the shared socio-economic pathways. Landsc. Urban Plan.250, 105135. 10.1016/j.landurbplan.2024.105135

  • CrossRef
  • Google Scholar

• 48

  YangJ.WangZ.SunY. (2025). Multi-scenario simulation of urban land expansion modes considering differences in spatial functional zoning. ISPRS Int. J. Geo-Information14 (4), 138. 10.3390/ijgi14040138

  • CrossRef
  • Google Scholar

• 49

  YazarM.HaarstadH.ElfvingJ. (2025). Cities incorporate equity in their climate policies but overlook procedural justice in decision-making. Nat. Cities2, 17–27. 10.1038/s44284-024-00167-w

  • CrossRef
  • Google Scholar

• 50

  YuH.YangY.ZhaoJ.CaiM.WangR.ChenG.et al (2025). Dynamic urban morphology mapping in Chinese cities based on local climate zone approach. Sci. Data12, 181. 10.1038/s41597-025-04478-y

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 51

  YuJ.GuoS.WangS.LuoY. (2025). Multi-scenario simulation of land use change and ecosystem health assessment in chengdu metropolitan area based on SD-PLUS-VORS coupled modeling. Sustainability17 (7), 3202. 10.3390/su17073202

  • CrossRef
  • Google Scholar

• 52

  ZhangL.LiaoZ. (2024). The prediction of urban growth boundary based on the ANN-CA model: an application to guangzhou. Heliyon, 10(19), e38052. 10.1016/j.heliyon.2024.e38052

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 53

  ZhangY.WangQ.HuY.WangW.MaoX. (2025). Analysis of the evolution of non-agriculturization arable land use pattern and its driving mechanisms. Land14 (5), 968. 10.3390/land14050968

  • CrossRef
  • Google Scholar

• 54

  ZhaoL.LiuG.XianC.NieJ.XiaoY.ZhouZ.et al (2022). Simulation of land use pattern based on land ecological security: a case study of Guangzhou, China. Int. J. Environ. Res. Public Health19 (15), 9281. 10.3390/ijerph19159281

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 55

  ZhaoM.YaoY.IndraprahastaG. S. (2023). Reflection on Guangzhou’s strategic spatial planning: current status, conflicts, and dilemmas. Land12 (11), 1996. 10.3390/land12111996

  • CrossRef
  • Google Scholar

• 56

  ZhengZ.XianS. (2023). Considering regional connectivity and Policy factors in the simulation of land use change in new areas: a case study of nansha new district, China. J. Geogr. Res.6, 42–60. 10.30564/jgr.v6i3.5814

  • CrossRef
  • Google Scholar

• 57

  ZhengH.HsiaoC. Y.-L.HoH.-W.JiC.ZhangS. (2024). How the spatial functional division of labor in urban clusters affects urban population size? Evidence from 19 urban clusters in China. Humanit. Soc. Sci. Commun.11 (1), 1704. 10.1057/s41599-024-04183-y

  • CrossRef
  • Google Scholar