Spatial Differentiation of Thermal–Ecological Environmental Responses in High-Density Central Subway-Hub Blocks and Their Associations with Built-Environment Characteristics

Abstract

Subway-hub blocks are critical areas where the pressures of metropolitan populations and environmental quality are closely interconnected. This study constructs a “pressure–context–carrier–response” (PCRC) framework (F1–F7) to systematically reveal the correlations between built-environment characteristics and environmental performance. The results demonstrate that resource allocation (F7) and comprehensive response (F5) display notable “asymmetric differentiation”. The socio-economic environment (F2, F3) considerably influences the concentration of green-space resource allocations (F7) (p < 0.01), with affluent blocks demonstrating a clear advantage in resource distribution. The thermo-ecological composite response (F5), which includes NDVI and LST, demonstrates “statistical convergence” (p = 0.894) across various block types, indicating that resource inputs cannot be linearly transformed into environmental efficiency. This disconnection is ascribed to two physical limitations: firstly, the stochastic nature of spatial distribution (Global Moran’s I ≈ 0) restricts the scale effects of green spaces; secondly, the nonlinear limitations of the physical medium indicate that under conditions of high pressure load (F1) and elevated spatial capacity (F6), the regulatory effectiveness of greening demonstrates a significant diminishing marginal return effect. Therefore, intervention planning must shift from controlling macro-level indicators to optimising micro-level accuracy to address ecological performance constraints in densely populated metropolitan areas.

1. Introduction

As global urbanisation increasingly transitions into a phase characterised by stock renewal and quality enhancement, transit-oriented development (TOD) has gained recognition as an essential spatial model for alleviating urban sprawl [1], fostering intensive land use [2], and promoting low-carbon mobility [3]. In Asia’s densely populated cities, subway-hub blocks serve as spatial units where rail transit networks and urban activities are intricately linked. These regions primarily facilitate extensive population migration, serve as transportation hubs, and support construction projects [4], demonstrating spatial operational attributes that are significantly different from those of conventional residential or commercial districts.

This intense spatial structure also imposes significant environmental pressure [5]. As a national-level pilot project, the Master Plan for Chengdu’s Construction of the Park City Demonstration Zone Practicing the New Development Concept mandates the deep integration of high-quality ecological environments into high-intensity urban functionalities. Nonetheless, a substantial body of empirical data demonstrates that, relative to adjacent regions, transit-oriented development (TOD) zones exhibit heightened urban heat island effects [6,7], ecological patch fragmentation [8], and increased impervious surface area [9]. Significantly, these environmental conditions are not evenly distributed within TOD areas: even among blocks with comparable location conditions and development intensities, substantial disparities in thermal environment and ecological performance exist across various transit-oriented districts [10]. This occurrence demonstrates that dependence on singular variables, such as density or accessibility, inadequately accounts for the spatial differentiation characteristics in environmental responses observed in subway-hub blocks [11].

Current research has made considerable progress in examining the correlation between the built environment and environmental performance, while research priorities display some discrepancies. On one side, in the domain of urban climatology, research has consistently clarified how characteristics of the built environment affect land surface temperature (LST) and local climate, employing indicators such as building density, sky view fraction (SVF), impervious surface ratio, and landscape patterns [12,13,14]. On the other hand, conventional planning and transportation research have primarily concentrated on the socio-economic impacts of transit-oriented development (TOD), including land premiums, commute behaviour, work-residence equilibrium, and gentrification [15,16]. The Chengdu Park City Green Space System Plan (2023–2035) highlights the strategic development of “micro-green-space” networks and “pocket parks” within the urban core to alleviate the urban heat island (UHI) effect. While recent research has initiated the exploration of microclimate attributes and the advantages of blue–green infrastructure near rail transit stations [17,18], the systematic quantitative characterisation of environmental responses remains relatively limited at the level of complex spatial units such as subway-hub blocks. This disparity is especially apparent in the lack of type recognition and comparative analysis grounded in multi-dimensional feature combinations. This presents a considerable obstacle to the accurate execution of refined governance mandated by the Design Guidelines for Urban Residual Space Regeneration in the “Zhongyou” (Central Area Optimisation) Region of Chengdu.

At the theoretical framework level, the conventional pressure–state–response (PSR) model and its derivatives, such as DPSIR, primarily facilitate macro-scale evaluations of ecological security or regional carrying capacity [19,20], highlighting the influence of population or economic pressures on environmental conditions while neglecting the moderating role of micro-spatial forms within these systems. This constraint impedes research from tackling essential planning enquiries at the block scale, such as why varying architectural forms and functional combinations produce disparate environmental response states under similar human activity constraints [21]. Thus, the conversion of ecological assessment results into spatial design and management methods is still limited. This study presents the PCRC (pressure–carrier–response–context) analytical framework, using the “policy context” as its theoretical underpinning.

Methodologically, current studies primarily utilise linear models, such as ordinary least squares regression (OLS) and geographically weighted regression (GWR), to analyse the correlation between the built environment and environmental performance [22,23,24]. However, urban systems exhibit significant nonlinearities and threshold effects, characterised by intricate interactions among parts of the built environment and their responses to external factors [25,26]. For example, the cooling effect of green areas may exhibit diminishing marginal returns at larger scales [27,28], and the influence of building height on ventilation and air quality may reveal significant changes across distinct critical thresholds [29,30]. In this context, conventional linear approaches encounter constraints in accurately representing intricate coupled interactions, which may result in estimation biases [31]. Therefore, evaluating the efficacy of current policy provisions on greening targets (such as green-space ratio objectives) in fulfilling their intended thermal and ecological aims in high-density environments necessitates a nonlinear analytical methodology.

This study utilises a multi-source data-driven analytical methodology, inspired by the technical indicators of Chengdu’s recent spatial planning policies, to examine the correlation between built-environment characteristics and thermal–ecological responses at the block level. Taking 357 subway-hub blocks within Chengdu’s central urban district as the study subject, it integrates remote-sensing imagery, dynamic population big data, and socio-economic data. Principal component analysis and hierarchical clustering methods are employed to identify block types, analysing the nonlinear interconnected characteristics between built-environment types and environmental responses.

This study primarily examines the following three questions: Does the thermal–ecological response at the block scale of subway-hub blocks exhibit stable differentiation characteristics? What variations arise in the combinations of built-environment attributes among distinct block types? Which characteristics of the built environment exhibit significant discriminatory power in type differentiation? This research delineates the spatial attributes of environmental response differentiation in high-density subway-hub blocks at the statistical-correlation level, using the preceding investigation. It provides an empirical reference for understanding the formation situation of different environmental performances under TOD conditions and supplies data support for tailored planning interventions. By addressing these questions, this study empirically evaluates spatial policies for Chengdu’s subway-hub blocks, offering crucial data support for future planning interventions.

2. Theoretical Framework and Research Methods

This study employs the “pressure–carrier–response–context (PCRC)” framework to elucidate the intricate coupling mechanisms among human activities, spatial configurations, and environmental responses in subway-hub blocks, representing an enhancement of the traditional PSR model. The PCRC paradigm highlights spatial form as the primary conduit between pressure and environmental reaction, facilitating a systematic examination of the nonlinear, multi-dimensional interactions at play.

Chengdu is a national centre city in southwestern China. It demonstrates the typical features of a transition from a single-centre to a multi-centre model. The city possesses a swiftly advancing rail transport network. The choice of 357 subway-hub blocks in Chengdu as study subjects ensures data availability and quality. It also ensures the representativeness and typicality of the research subjects. This study utilised a high-dimensional indicator system derived from multi-source data. It performed systematic data cleaning, dimensionality reduction, classification, and environmental response analysis. Thus, it elucidates the environmental disparities and underlying mechanisms among various types of blocks.

2.1. Analytical Framework: “Pressure–Carrier–Response–Context” (PCRC) Framework

This work iteratively modified the conventional pressure–state–reaction (PSR) model to clarify the interconnected interactions among anthropogenic activities, spatial form, and environmental responses in high-density urban blocks. A novel analytical framework, pressure–carrier–response–context (PCRC), was developed. In contrast to conventional correlation analyses utilising multi-dimensional indicator systems, the principal advantage of the PCRC framework is in its hierarchical analytical structure. Conventional approaches typically do parallel correlation analyses of many environmental parameters, which frequently conceal the intricate structural discrepancies within high-density urban systems. By conceptualising the “carrier” as a spatial mediator, the PCRC framework effectively deconstructs the link between the “allocation of green space” and “environmental performance”. The importance of this deconstruction lies in its capacity to clearly delineate the asymmetric differentiation between the two; specifically, it elucidates why an increase in physical inputs (such as F7 allocation of green space) has not resulted in a notable improvement in environmental performance (such as F5 thermal–ecological response) within the statistical distribution. This shift from “state description” to “transformation efficacy analysis” enables this study to uncover distributional discrepancies obscured by global statistical trends, thereby offering more precise empirical tools for enhanced urban regeneration.

In the PCRC analytical framework, environmental response is understood as the composite state characteristics that emerge in conjunction with built-environment features, under the influence of anthropogenic pressures, and within specific socio-economic and locational contexts. The “carrier” dimension delineates the form and functional organisation of the built environment at the block scale, highlighting its essential spatial attributes in the interplay between stress and environmental response.

This framework comprises four key elements:

(1)

Pressure: Refers to the concentration of intense human activity and developmental requirement at the block scale, including the intensity of population activity, changes in population during commuting periods, and the degree of facility clustering.

(2)

Carrier: Refers to the built-environment characteristics that carry and organise the above pressures, encompassing development intensity, variances in architectural form, road infrastructure, and the integration of land use and function combination.

(3)

Response: Refers to the observable environmental-state characteristics at the block scale, mainly reflected in the thermal environment and ecologically related indicators, such as surface temperature and vegetation indices.

(4)

Context: Pertains to the external variables influencing the co-occurrence of the aforementioned factors, including socio-economic level, locational accessibility, and macro policy. Contextualised by specific policy directives (e.g., “Chengdu Park Urban Green Space System Planning (2023–2035)”), mandatory spatial development constraints (such as maximum plot ratio limits) and ecological response objectives (including cooling and greening mandates) are established, thereby regulating the “pressure–carrier–response” coupling process at a fundamental level.

The urban environment system can be viewed as a multivariable, nonlinearly coupled process. Here, the environmental response (R) is not only a direct linear product of external pressure (P) but also a comprehensive effect of pressure under the constraint of specific spatial and temporal context (X), which is physically modulated by the spatial form carrier (C). This relationship can be shown with structural equations:

$$\mathrm{C}=\mathrm{g}(\mathrm{P},\mathrm{X})$$

$$\mathrm{R}=\mathrm{f}(\mathrm{P},\mathrm{C},\mathrm{X})$$

The spatial carrier (C) results from adaptation to pressure and context, whereas the environmental response (R) is the composite manifestation of the combined effects of the aforementioned multi-dimensional elements. The aforementioned formula seeks to clarify the interaction logic among the different dimensions within the PCRC framework, providing a theoretical foundation for subsequent dimensionality reduction (PCA) and type identification (cluster analysis), rather than serving as a parametric estimation model of causal pathways.

This paper adheres to a research trajectory of “data-driven characterization–type identification based on the built environment–asymmetric differentiation characteristics of built-environment type and environmental response.” It establishes a thorough analysis flow chart based on the PCRC approach (Figure 1). The framework aims to compare state characteristics of the environmental response of different types of subway-hub blocks in the built environment from the block scale, including the following three stages:

(1)

Phase 1: Data collection and establishment of an indicator system. During this phase, we initially gathered multi-source data, encompassing population distribution (LandScan), points of interest (POI), commuting data (BaiduMap), and remote sensing images (SensingData), in accordance with pertinent policy objectives (Step 1-1). Utilising these many data sources, the study developed a PCRC analytical framework encompassing four dimensions: pressure, carrier, response, and context. To mitigate multicollinearity across variables, the study employed the Pearson correlation coefficient to evaluate the basic framework, eliminating superfluous indications and eventually identifying 20 essential indicators (Step 1-2). In the process of selecting indicators, this study integrates the ecological performance requirements outlined in the “Chengdu Park Urban Green Space System Planning (2023–2035)” with the technical focus on microform optimisation specified in the “Chengdu Zhongyou Regional Urban Remaining Space Renewal Planning and Design Guidelines”. This logical congruence ensures that empirical variables (such as NDVI and LST) can accurately reflect the fundamental aims of policy issues (such as enhancing green spaces and lowering temperatures), thereby providing a scientific foundation for the meticulous implementation of policies.

(2)

Phase 2: Identification of subway-hub block types based on built-environment variables. Utilising the matrix of 20 key indicators developed in Phase 1, the study applied principal component analysis (PCA) for feature extraction and dimensionality reduction, revealing seven mutually independent latent factors (F1–F7). Consequently, utilising the scores of these criteria, hierarchical clustering was employed to categorise the 357 blocks into four distinct built-environment types: R1–R4. Simultaneously, Global Moran’s I and LISA spatial statistical analyses were utilised to clarify patterns of the heterogeneous distribution of the sample concerning attribute characteristics and geospatial distribution, thus establishing the classification basis for subsequent environmental response evaluations.

(3)

Phase 3: Assessment of environmental response status and policy recommendations. The study identified an association between plot types (R1–R4) and environmental response factors (F5 and F7) through spatial overlay analysis. The study clarifies the actual performance and reaction characteristics of ecological resources within several spatial designs under high-density situations, highlighting the substantial disparities in environmental performance scores among different types of subway-hub blocks. The study establishes the correlation between green-space distribution and thermal–ecological performance across various block types, offering evidence-based support for the iterative revision of the “Guidelines for Refined Governance of Urban Renewal in Chengdu”, thus enabling a shift in planning from “macro-level indicator control” to “micro-level morphological intervention”.

Figure 1 illustrates the comprehensive research methodology and the logical connections among the different phases, emphasising the closed-loop process from policy goals, data collection, dimension construction, and type identification to environmental response analysis, thereby showcasing the practical significance of the PCRC framework in the examination of intricate urban systems.

2.2. Research Method

This study uses a data-driven PCRC framework to develop a systematic analytical workflow that includes indicator selection, principal component analysis (PCA), the identification of block types, and the quantitative analysis of type characteristics to clarify the coupling mechanisms between built-environment drivers and environmental responses.

This method identifies multicollinearity among indicators as a preliminary step for variable selection. Using the Pearson correlation coefficient matrix, indicators with |r| > 0.7 are discarded, preserving variables that are information-independent to ensure robustness in subsequent models.

Due to the intricate relationship between urban morphology and environmental indices, direct clustering is vulnerable to the influence of data noise. The PCA approach employs an orthogonal transformation to convert high-dimensional, correlated variables into linearly independent, low-dimensional principal components. This research identifies principal components that account for approximately 65% of the cumulative proportion in ANOVA as new feature vectors, thereby facilitating dimensionality reduction while optimally retaining original information.

Cluster analysis is used to identify typical patterns in types of subway-hub blocks. This study employs a hierarchical clustering algorithm using the Ward method to minimise intra-class variance, with principal component scores from PCA as input. By analysing dendrograms alongside the contour coefficient, the appropriate cluster count is determined, thereby distinguishing specific subway-hub block types defined by notable morphological and environmental characteristics.

R1–R4 type labels were assigned to specific block units after type identification. To measure differences in environmental responses across various built-environment typologies, statistical features of environmental response factors (F5, F7), including the mean, median, and outlier distributions, were computed.

3. Research Area and Data Samples

3.1. Research Area

This research designated Chengdu, a national central metropolis in southwestern China, as its empirical instance (Figure 2). Chengdu, a typical Asian metropolis that transitioned from a high-density monocentric to a polycentric structure, has experienced the fast development of its rail transit system in recent years. The regions around various subway-hub blocks exhibit significant differences in development intensity, geographical configuration, and functional organisation. This spatial heterogeneity, arising within a relatively stable macro-context, creates optimal settings for comparing the environmental response characteristics of different subway-hub blocks.

The research area includes the twelve administrative districts of the central urban zone as defined in the Chengdu Land Space Master Plan (2021–2035): Jinjiang District, Qingyang District, Jinniu District, Wuhou District, Chenghua District, High-Tech Zone, Longquanyi District, Shuangliu District, Wenjiang District, Xindu District, Pidu District, and the core area of Tianfu New Area, encompassing an approximate total area of 880 km². This research takes the operating subway station as the centre and establishes a buffer zone based on an 800 m walking service radius (about a 10 min walk). Overlapping regions are segmented and fused using Thiessen polygons, taking physical boundaries into account, including urban trunk roads and rivers. Finally, 357 unique metro-hub/subway-hub block units are constructed as the primary analytical samples.

3.2. Data Source and Processing

This research uses the PCRC framework to develop a multi-source geographic database (

Table 1. Multi-source geospatial data variable description statistical table.

Dimensions	Subclass	Indexes	Name	Unit	Data Sources
I. Anthropogenic Pressures Dimension	Population Activity Intensity	A14	Population Density	People/km2	Oak Ridge National Laboratory (ORNL): LandScan 2024 Global Population Distribution Dataset (1 km resolution)
		A15	Population Density Reclassification	Grade (1–5)
		A37	Overall Average Relative Population Value	Dimensionless	Baidu Huiyan location big data (19 August 2024, 21 August 2024, 24 August 2024, 25 August 2024, 24 h gridded population big data, accuracy 200 m)
	Commuting Tidal Pressure	A29	The Average Relative Population Value of the Morning Peak on Weekdays	Dimensionless
		A30	The Average Relative Population Value of the Peak in the Evening of the Weekdays	Dimensionless
		A31	The Average Relative Population Value for the Rest of the Weekdays	Dimensionless
		A32	Average Relative Population Value of Weekdays	Dimensionless
	Business Leisure Pressure	A33	The Average Relative Population Value of the Morning Peak on Weekend	Dimensionless
		A34	The Average Relative Population Value of the Peak in the Evening of the Weekend	Dimensionless
		A35	The Average Relative Population Value for the Rest of the Weekend	Dimensionless
		A36	Average Relative Population Value of Weekend	Dimensionless
	Facilities Agglomeration Pressure	A1	POI Density	Number/km2	Gaode Maps Open Platform 2024 POI Data
II. Space Form Carrier Dimension	Development Intensity	A2	Bus Station Density	Number/km2	Baidu Maps open platform API 2024
		A9	Building Density	%
		A10	Volume Ratio	Dimensionless
		A11	Average Building Height	m
		A12	Standard Deviation of Building Average Height	m
		A13	Road Network Density	%	Urban Road Network Vector Data in 2024
	Land Use and Function	A3	Proportion of Residential Land	%	EULUC-China 2.0 database [32]
		A4	Proportion of Commercial Land	%
		A5	Proportion of Industrial Land	%
		A6	Proportion of Road and Traffic Land	%
		A7	Proportion of Public Service Land	%
		A8	Function Mix	Dimensionless
III. Environmental Response State Dimension	Thermal Environment	A19	Average Surface Temperature	°C	GEE (Landsat-8/9 TIRS & MODIS)
		A20	Heat Island Intensity	°C
	Ecological Quality	A16	Greening Rate	%	UGS-1m Fine Green Space Dataset of Major Cities in China [33]
		A21	NDVI	Dimensionless	GEE (Landsat-8/9 TIRS & MODIS)
IV. Socio-economic and Location Environment Dimension	Economic Value	A17	Average Housing Price	yuan/m2	Shell Housing Platform 2024 Second-hand Housing Transaction Data
		A18	Average House Price Standard Deviation	yuan/m2
	Macro Location	A22	Number of Sites Reached in 30 Minutes	Number	Gaode Map API (Path Planning)
		A23	Arriving—Average Time	Minute
		A24	Number of Functional Areas in 30 Minutes	Number
		A25	Time to Chunxi Road	Minute
		A26	Time to Jinrong City	Minute
		A27	Time to Tianfu Fifth Street	Minute
		A28	Time to Chengdu East Railway Station	Minute

) that includes four dimensions: pressure, carrier, response, and context. The data period is unified for 2022–2024, projected to the WGS_1984_UTM_Zone_48N coordinate system, and aggregated to the block unit (Figure 3). To ensure consistency across these steps, all indices are standardised using the Z-score method (Figure 4), reflecting each block’s index relative to the whole-domain average.

The Tyson polygon division results in varying geographical areas for each site block unit. This study standardised the areas of all quantitative and scale-model indicators to mitigate the impact of scale effects on evaluation outcomes. Excluding ratio-based or mean-based indicators like house prices (A17), function mix (A8), and environmental response indicators (A19–A21), all other indicators pertaining to total volume (including A9 building density, A13 road network density, and A14 population density) were calculated using the actual area (km²) of each Tessen polygon unit as the denominator. Transforming absolute numbers into a scaling feature ensures the cross-sectional comparability of blocks of varying scales when evaluating built-environment intensity.

The Google Earth Engine (GEE) cloud platform was employed to obtain all Landsat 8/9 TIRS (30 m) and MODIS/VIIRS (1 km) imagery for the environmental response indicators (A19–A21) within the study area from 1 July to 31 August 2024. This timeframe coincides with the apex of the summer heatwave and the zenith of vegetation growth in Chengdu. This work employed cloud removal and median reduction to synthesise stable representative data for this time period, effectively mitigating noise from daily meteorological fluctuations and phenological variations. In regions with inadequate Landsat coverage or cloud cover, MODIS/VIIRS raster data were used for spatial interpolation to address these deficiencies, and the data were consistently resampled to 30 m resolution, ensuring a high congruence between thermal–ecological indicators across both temporal scope and spatial precision.

3.3. Construction of Index System

This study has preliminarily developed a measurement system consisting of four primary dimensions and 37 indicators, based on the PCRC analytical framework. To ensure the independence and explanatory power of the model inputs, many collinearity assessments were performed on the indicators using Pearson’s correlation coefficient (Figure 5). After removing redundant variables with high correlations (|r| > 0.7), the study identified 20 essential indicators for further examination.

3.4. Sample Characteristics and Spatial Distribution

3.4.1. Global Spatial Autocorrelation, Global Moran’s I

The Global Moran’s I statistic results from the complete sample (N = 357) (Figure 5) indicate significant spatial heterogeneity in the distribution of indicators, demonstrating a varied spatial distribution pattern overall.

(1)

Macro-level context and stress indicators demonstrated robust spatial aggregation. The Moran’s I index for average arrival time (A23) attained 0.853, the highest among all measures, succeeded by heat island intensity (A20, I = 0.77) and population density (A14, I = 0.731). These data demonstrate that the indicators exhibit significant spatial continuity across the research region, suggesting a high degree of similarity between neighbouring hub blocks.

(2)

In contrast, the spatial autocorrelation of indicators associated with the built environment was generally low. Moran’s I values for functional diversity (A8) and road land share (A6) were 0.165 and 0.160, respectively. Although still statistically significant (p < 0.01), their spatial clustering was considerably lower than that of macro-level contextual and pressure indicators, suggesting that spatial form characteristics at the neighbourhood scale exhibited higher local differences within the research area.

(3)

Moran’s I index for the greening rate (A16) was −0.009 and was not statistically significant (p > 0.05), suggesting that its spatial distribution was nearly random. This outcome indicates, at the statistical level, that the distribution of green spaces in high-density central district regions lacks unique spatial continuity characteristics.

Overall, the spatial autocorrelation analysis reveals notable disparities in spatial clustering across different indicator types: Macro-level context and pressure indicators demonstrate strong spatial continuity. In contrast, built-environment characteristics and specific ecological indicators exhibit more dispersed and localised spatial distribution patterns.

3.4.2. Local Spatial Autocorrelation Analysis, LISA

This study utilises local indicators of spatial association (LISA) to examine representative indicators, hence enhancing the characterisation of spatial distribution patterns at the local scale (Figure 6).

(1)

LISA clustering results reveal that indices associated with population activity intensity (Figure 6a) generally display a distribution pattern along rail transit corridors, characterised by several discrete “high–high” agglomeration nodes. This spatial arrangement illustrates the concentration of population activity around subway hubs rather than a continuous spread.

(2)

The functional mix indicator (Figure 6b) displays a clear spatial pattern of “core–periphery” divergence. Urban centre areas mostly exhibit “high–high” agglomerations, while peripheral zones exhibit “low–low” agglomerations, indicating substantial spatial variation in functional layouts across blocks.

(3)

The LISA results for thermal environment indicators (Figure 6c) demonstrate a pattern of consistently high-value clusters, with “high–high” zones widely scattered throughout the centre urban area, indicating the spatial continuity of thermal environmental conditions.

(4)

The local spatial autocorrelation findings for the green-space rate index (Figure 6d) predominantly indicate statistically inconsequential regions, exhibiting relatively little agglomeration in select areas. This suggests that the distribution of green space is relatively dispersed at the block level.

3.4.3. Exploratory Data Analysis: Description of Nonlinear Characteristics

This study utilises a bivariate scatter plot with superimposed LOWESS smoothing curves to examine the coupling features of key variables (Figure 7). The study seeks to delineate the overall trend of change between variables.

Firstly, no substantial linear positive association was detected between development intensity (A10) and heat island intensity (A20) (Figure 7a). The scatter plot shows dispersion, and the overall variation in the LOWESS fitted curve is finite. The link between the two is relatively complex within the sample range and may be influenced by additional built-environment elements.

Secondly, the link between the green-space ratio (A16) and heat island intensity (A20) is weak (Figure 7b). Across most value ranges, the fitting curve tends to be gentle, suggesting no stable monotonic relationship between the variation in green-space rate and the thermal–environment response at the block scale.

Commuting time (A23) demonstrates a distinct negative association with housing price (A17) (Figure 7c). The fitted curve decreases with increasing commuting time, illustrating the basic spatial correlation characteristics between location conditions and economic value.

Finally, the correlation between functional diversity (A8) and population activity intensity (A37) demonstrates a significant nonlinear distribution feature (Figure 7d). The scatter plot shows significant dispersion within the high-mixing range, and the fitting curve does not exhibit a continuous upward trend, indicating that the relationship between the two may vary across different value ranges.

Overall, this exploratory analysis reveals that the link between various built-environment indicators and environmental response variables is not merely linear; the interconnections among the variables demonstrate a notable complexity depending on the combinations of indicators used. This result provides a data-level reference for subsequent type identification and multivariate analysis.

4. Results

This section examines the fundamental factors influencing the built environment and its environmental response traits in subway-hub blocks, utilising multi-source spatiotemporal data. Initially, seven primary latent variables were identified via principal component analysis, elucidating the overarching disparities and inherent connections across the dimensions of population activity pressure, spatial configuration, socio-economic context, and environmental response. Subsequently, hierarchical clustering was utilised to categorise four distinct block types based on built-environment parameters, delineating the spatial structures and functional configurations of each. Ultimately, the similarities and disparities among the different block types were analysed regarding green-space resource distribution and thermal–ecological efficacy. We observed that the allocation of green-space resources (F7) showed clear typological variation, whereas the thermal–ecological response (F5) exhibited intricate, nonlinear, and homogeneous uniform traits.

4.1. Analysing Spatial Dimensions: PCRC Factor Extraction and Semantic Designation

Utilising the 20 key indicators selected through Pearson correlation coefficient screening, seven orthogonal latent variables were effectively retrieved through principal component analysis (PCA) (KMO = 0.741, p < 0.001). Utilising the criterion of eigenvalues greater than 1 and examining the explained cumulative variance, seven orthogonal main components were recovered, representing 65.04% of the total variance. This study adheres to the notion that the absolute value of factor loading must exceed 0.6 to identify core contributing factors in the semantic definition of principal components, hence ensuring the objectivity and scientific rigour of factor nomenclature.

According to the factor loading structure, this study divides it into four dimensions—pressure, carrier, context and response—and names them (Figure 8):

Pressure layer (F1): Defines the pressure of population activities and the intensity of construction, indicating the essential load-bearing capability of the block.

Context layer (F2, F3): Analyses the social and economic context and locational resources of the block. F2 (value context): Mainly driven by the proportion of house prices and commercial services, representing the level of land rent and economic potential energy. F3 (traffic background): Amalgamates site accessibility and road network density to illustrate the accessibility of blocks within the urban system.

Carrier layer (F4, F6): Characterise the physical space form inside the block. F4 (function carrier): Mostly influenced by the degree of land-use integration, indicating the function of the sample and its spatial arrangement. F6 (form carrier): Defines the vertical shape and height characteristics of the building, indicative of its spatial capacity.

Response Layer (F5, F7): Defines environmental response states.

F7 (Green-Space Resource Allocation): Reflects the abundance of green spaces and physical resources.

F5 (Thermal–Ecological Response Performance): Combines NDVI (positive influence) and heat island intensity (negative influence). Positive F5 values signify exceptional performance (cool and highly verdant), whilst negative values reflect subpar conditions (hot and deficient in greenery).

The results reveal that the drivers of the built environment (F1–F4, F6) and the outcomes of environmental responses (F5, F7) demonstrate a statistically significant orthogonal separation, thereby facilitating the following classification of block types based exclusively on the drivers.

4.2. Block-Type Identification of Subway Hub Based on Built-Environment Factors

This study identified five primary components relating to the built environment at subway-hub blocks—F1, F2, F3, F4, and F6—as input variables. Hierarchical cluster analysis was performed utilising Ward’s least squares approach. The integration of the dendrogram structure with the contour coefficient data indicated that the ideal number of clusters was four (Figure 9). The type-identification procedure was exclusively based on the built environment and location-related parameters, omitting any thermal or ecological response variables. The results delineate the geographical, structural, and land-use disparities among subway-hub blocks. Clustering results show that the four types of blocks show significant differences in the combination of built-environment factors, and their classification characteristics and thermal–ecological performance (

Table 2. Summary table of classification characteristics and thermal–ecological performance of subway-hub blocks.

Type	N	F5 Factor Performance Interval	F7 Factor Performance Interval	Morphological and Response Characteristics
R1	47	[−1.84, +2.12] Dispersion is extremely high.	[−0.76, +0.57] Fluctuations are significant.	Unstable shadow effect: Exceptionally high vertical intensity (F6: 1.94) induces considerable shadow cooling; yet, the canyon effect results in heat accumulation, significantly influenced by local topography.
R2	164	Concentrated on the negative interval (the mean is about − 1.0).	Concentrated near 0.	Subpar ecological edges: The majority are underdeveloped zones (F2, F3, F4 < 0), characterised by low construction intensity and minimal functional diversity. Despite the limited number of structures, there is a lack of structured, high-quality green landscape design.
R3	93	Best performance (the highest positive frequency).	[−1.11, +0.99] Fluctuation is large and the upper limit is high.	High-quality ecological: The substantial concentration of capital and transportation (F1, F2, F3 > 0) is coupled with the investment in superior micro-green spaces or landscape engineering, resulting in enhanced thermal regulation within a high-density context.
R4	53	Two-polarisation (scattered distribution).	Overall low, mostly negative values.	Ecological support lag: typical transitional plots exhibit significant functional mixing (F4: 1.35) but possess poor economic value (F2: −0.99). Numerous rigid surfaces exist, and ecological amenities have not been maintained.

) are as follows:

(1)

R1 High-Form–Low-Accessibility Type: It is characterised by an extremely high vertical form (F6: 1.94) but poor accessibility (F3: −0.80). These are often high-density housing areas at the city’s edge with few roads and low spatial connectivity.

(2)

R2 Low-Intensity–Low-Mixed Type: All categories exhibit low scores, mainly distributed in immature development zones or single functional groups on the edge of the city.

(3)

R3 High-Density–High-Value–High-Accessibility Type: It achieved exceptionally high scores across pressure, value, and accessibility. The usual representative is the city’s central hub, distinguished by a significant concentration of capital, pedestrian traffic, and vehicular activity.

(4)

R4 High-Hybrid Renewal Type: Defined by significant functional hybrid characteristics (F4: 1.35) but comparatively diminished economic value (F2: −0.99). Denotes brownfield redevelopment areas in transition, signifying potential growth hubs for urban vibrancy.

In order to further reveal the differentiation degree of built-environment characteristics to block types, the study analysed factor loadings (Figure 8a) and factor scores for each type (Figure 9). Empirical data demonstrate that the scores of indicators with significant loadings directly influence the classification attributes of blocks. Within the “carrier” dimension, A11 (average building height, 0.844), which possesses a markedly high loading, predominates Factor F6, resulting in type R1 displaying extremely high vertical form attributes (F6 score of 1.9378), while concurrently demonstrating a notable reduction in Factor F3 (the “accessibility” dimension) (−0.7933), thereby illustrating its asymmetric differentiation defined by “high vertical form and low traffic support”. In the aspects of “pressure”, “carrier”, and “context”, the R3 type achieved the greatest overall scores for F1 (0.4761), F2 (0.7335), and F3 (0.8483). This confirms that the indicator combination of A37 (overall average relative population value), A17 (average housing price), and A22 (number of sites reached in 30 min) acts as the principal distinguishing criterion for identifying “high-density–high-value–high-accessibility” subway-hub blocks.

Moreover, A8 (function mix, 0.795), with a substantial loading value, predominates the F4 factor, leading to R4 types demonstrating considerable functional diversity (F4 score: 1.35) and precisely characterising their typical characteristics of transition period as “highly mixed, low economic value (F2: −0.9942)”. This feature mapping logic, grounded in fundamental indicators (such as A11, A37, and A8), determines which elements of the built environment exhibit the highest discriminatory power, thus establishing a coherent closed loop from indicator screening to spatial classification.

4.3. Asymmetric Differentiation Characteristics of Green-Space Resource Allocation and Environmental Response

This study utilised Tukey’s IQR method (Tukey, 1977) [34] for the outlier treatment to ensure that statistical results accurately represent general patterns across various block types and to mitigate the impact of extreme outliers, such as exceptionally large ecological parks, on overall averages. The criterion entailed the exclusion of data points that lie beyond the specified range. Tukey’s IQR test showed minor sample size variations for each group due to differences in indicator distributions. There were 351 valid samples for the environmental responsiveness dimension (F5) and 354 for the green-space allocation dimension (F7). The minimal sample exclusion rate (both below 2%) ensures that the study’s conclusions are broadly applicable to high-density TOD blocks. After this modification, the data distribution within each group is more closely aligned with the assumption of normality, thereby substantially improving statistical robustness. The Levene test showed that the variances of F5 and F7 were not equal among the groups (p < 0.05). Therefore, this study used Welch’s ANOVA, which handles unequal variances well, to assess the significant differences among groups.

4.3.1. Type-Homogeneity Characteristics of Thermal–Ecological Environment Response (F5)

As illustrated in Figure 10, the score distributions of metro-hub block types across the F5 dimension exhibit significant overlaps and homogeneity across different built environments. The medians for all four types converge near zero, with no discernible separation in the longitudinal distribution of box plots. The median for type R3 is numerically slightly higher than the other types, whilst types R1 and R4 exhibit relatively greater dispersion. However, no stable gradient ranking structure emerges between the different types.

Welch’s ANOVA test showed no statistically significant differences between the built-environment types in the F5 dimension (p = 0.8940 > 0.05). This extremely high p value strongly reveals that, within the context of subway-hub blocks, the thermal–ecological comprehensive environmental performance is affected by the built environment in a more complex way. Moreover, it does not exhibit regular differentiation characteristics based on land-use type classifications.

4.3.2. Type-Heterogeneity Characteristics of Green-Space Resource Allocation (F7)

In contrast, notable structural variances arise across various built-environment types regarding the green-space resource allocation factor (F7) (Figure 10). The R3 box position is significantly elevated, with its median and interquartile range primarily situated in a positive area. This indicates a relatively ample provision of green spaces within this category of blocks (high-density, high-value type). In contrast, the R4 box position is markedly diminished, concentrated in negative values with a mean of roughly −0.5. This highlights the challenge of green-space scarcity facing this category of renewal blocks.

Welch’s ANOVA test showed very significant differences among the types in the F7 dimension (p = 0.0011 < 0.01). This research demonstrates that built-environment typologies have significant explanatory powers and discriminatory abilities regarding the amounts of green-space resource allocation. It affirms that land-use functions and development intensity at the planning level have a direct and substantial impact on the availability of green spaces.

The analysis indicates that different types of built environment show significant asymmetrical variation in environmental response dimensions. The green-space resource allocation factor (F7) reveals notable intergroup disparities across built-environment types, whereas the thermal–ecological response factor (F5) shows neither synchronous nor distinct differentiation patterns in response to changes in built-environment types.

From the standpoint of the response dimension, F7 shows significant statistical distinctiveness among typologies, with pronounced differences in the median values and distribution centres across built-environment types. This suggests that type classifications derived from morphological characteristics can, to some degree, directly correlate with changes in levels of green-space provision. Conversely, F5 shows significantly overlapping distributions across types, with comparable medians and dispersion ranges across categories. This suggests that the thermo-ecological environmental reaction does not establish a permanent gradient sequence at the type level.

This finding suggests that in high-density blocks near subway hubs, built-environment typologies exhibit varying explanatory powers across different environmental response indicators: they successfully distinguish environmental factors at the resource-allocation level but do not yield corresponding type differentiations at the integrated thermal–ecological performance level.

5. Discussion

5.1. Theoretical Logic of PCRC Framework

The “pressure–carrier–response–context” (PCRC) analytical framework introduced in this study is an enhancement and modification of the traditional “pressure–state–response” (PSR) model [19,20], specifically within the urban spatial context, addressing the research trend of “form–process–function” interrelation in urban ecology. In the analysis of intricate urban systems, the conventional PSR model frequently emphasises a linear depiction of the causal chain, neglecting the fundamental function of physical spatial shapes as a “physical intermediary”. The PCRC framework regards the block as a material entity that mediates anthropogenic pressures and environmental responses by introducing the “ carrier “ dimension (i.e., the morphological and functional characteristics of the built environment). The empirical findings demonstrate that the framework can effectively elucidate the nonlinear relationship between green-space resource allocation (F7) and environmental response (F5) in high-density environments, offering a more spatially targeted analytical tool for examining the coupling of urban complex environments at the micro scale and addressing the oversight of spatial form in macro-environmental assessments.

5.2. Asymmetric Differentiation Characteristics of Built-Environment Types and Environmental Responses

This study, informed by the empirical data from Section 4.3, demonstrates notable asymmetrical differentiations in environmental response dimensions across several built-environment types (R1–R4). This phenomenon elucidates the intricate asymmetry between physical spatial configuration and environmental performance in high-density urban blocks [35,36]:

(1)

Rigid Mapping of Resource Allocation: The green-space supply factor (F7) demonstrates considerable variation among different categories (p < 0.01), suggesting that built-environment characteristics (land-use type, development intensity) have a substantial explanatory impact on the spatial distribution of green-space resources. It demonstrates the significant constraint that planning involvement imposes on resource allocation.

(2)

Statistical Convergence in Environmental Responses: The thermal–ecological response factor (F5) exhibited no significant differences among categories (p = 0.8940 > 0.05); this statistical outcome indicates the inadequacy of macro-level classifications to account for micro-level performances. An investigation of global spatial autocorrelation (Global Moran’s I = 0) suggests that the random geographical distribution of green-space resources may be a fundamental mechanism driving this occurrence. This differentiation characteristics indicate that enhancements in environmental performance are shaped by micro-scale spatial configurations rather than a mere linear accumulation of physical inputs [37]. This phenomenon aligns with existing research findings on the non-monotonic nature of thermal responses in high-density urban areas [38], suggesting that macro-level typological classifications have inherent limitations in elucidating variations in the urban thermal environment.

The aforementioned data confirm that in high-density environments, the density difference no longer corresponds to the stable thermal environment difference [39]. Macro-level built-environment types influence the spatial distribution of urban resources and activities, but they do not directly dictate environmental performance. This is consistent with findings that the built environment has spatial heterogeneity on microclimates [22,40].

5.3. Potential Mechanism: Spatial Randomness and Nonlinear Correlation Trend

In conjunction with the spatial autocorrelation and exploratory analysis in Section 3.4, this section further clarifies the deep mechanism behind the lack of type difference in F5:

(1)

The randomness of green-space distribution: The LISA clustering map (Figure 6d) offers direct empirical evidence that the green rate (A16) does not constitute a continuous, spatially cohesive network. Notable ”high–high” clusters are present only in a limited number of isolated cases in the city centre, whereas the remainder of the research area (grey regions) is categorised as “not significant”, exhibiting widespread “low–low” distributions on the periphery. Current studies validate that this spatial randomisation considerably restricts the synergistic impact of green patches on microclimate regulation [41].

(2)

Nonlinear correlation characteristics of environmental responses: The LOWESS fitting trajectories reveal a clear nonlinear relationship between built-environment variables and thermal–environmental responses. Specifically, in high-density TOD blocks, the increase in local green spaces demonstrates significant declining marginal effects. This indicates that enhancements in environmental performance show nonlinear responsiveness to physical interventions. This adequately elucidates why the ultimate total thermal–ecological response of R1-R4 tends to converge gently, despite the fact that the allocation of green-space resources is different. Consistent with the conclusion that the influence of the built environment on temperature exhibits “non-monotonicity” and “spatial heterogeneity” in current research [25], it underscores the intricate nature of the environmental response mechanisms in high-density urban areas, rather than a straightforward linear aggregation.

5.4. Strategic Shifts in Planning Practice: From Macro-Control to Micro-Intervention

Conventional planning solutions focused on “land-use modifications” or “development intensity zoning” risk failure in enhancing microclimate performance, demonstrating ineffectiveness in directly improving thermal conditions [13]. Environmental optimisation must transition from macro-management and -control to micro-scale interventions, emphasising an improved connection of green areas, the creation of micro-ventilation corridors, and the advancement of vertical greening to alleviate the urban heat island effect [42,43].

In this context, the discovery of the R3 type (high-density, high-value, high-accessibility) offers empirical validation at the micro-level for the “compact city” idea [44,45]. Although the R3 type experiences considerable development intensity and population pressure, its integrated thermal–ecological response (F5) has not greatly worsened, and it surpasses other kinds in green- space resource allocation (F7). This suggests that refined morphological design and the reasonable allocation of green-space resources can enable dense urban areas to attain a dynamic equilibrium between development intensity and environmental quality. Consequently, this conclusion affirms the sustainability potential of compact cities and underscores that environmental optimisation should transition from simple density regulation to modifications in spatial morphological specifics. This closely aligns with policies presently enacted in Chengdu, such as the “Chengdu Park City Green Space System Plan (2023–2035)”, which prioritise the improvement in urban ecological performance via sophisticated spatial governance. In parallel, the micro-morphological optimisation suggested in this study, grounded in the PCRC framework, offers a technological approach to realising these overarching policy goals.

This study concentrates on high-density subway-hub blocks in Chengdu, a city with a typical humid subtropical climate. The proposed planning strategies—green-space connectivity, micro-ventilation corridors, and vertical greening—are based on research and empirical evidence relevant to this specific climate and urban context. However, urban environmental characteristics and ecological management requirements differ significantly across various climatic conditions (e.g., drought, cold, or temperate regions), which in turn affect the outcomes of these methods. Therefore, to address these variations, forthcoming research ought to incorporate varied climatic contexts and perform cross-regional comparative analyses. This approach would help investigate the adaptability and scalability of these planning strategies across different environments, thereby supporting more precise and systematic urban environmental optimisation initiatives.

6. Conclusions

6.1. Main Conclusions

(1)

Identification of multi-dimensional built-environment features: Seven principal components were extracted by PCA, categorising four typical built-environment types (R1–R4) and demonstrating the efficiency of the PCRC framework in assessing high-density, complex-block structures.

(2)

Spatial heterogeneity across varying scales: Global spatial autocorrelation analysis indicates that macro-level contexts and stress indicators display significant spatial clustering, while micro-level morphology and green-space ratios reveal pronounced fragmentations and random distributions, illustrating intricate spatial heterogeneity at the block scale.

(3)

Asymmetric differentiation between resources and performances: The built-environment type strongly affects green-space resource allocation (F7) (p < 0.01) but does not significantly influence thermal–ecological response factors (F5) (p > 0.05). This modifies the linear presumption that “form dictates performance”, suggesting that many nonlinear regulating systems influence environmental performance in high-density blocks.

6.2. Policy Implications

This study identifies actionable insights to improve governance under Chengdu’s “park city” plan, based on findings from the PCRC framework analysis.

(1)

Transitioning from scale growth to performance enhancement: Chengdu’s present policies have shifted from volume regulation to quality enhancement. Enforce access criteria for “building-integrated shading” and “vertical greening” for type R3. By optimising the characteristics of the “carrier”, enhance passive cooling efficacy. This fosters climatic resilience in high-density development.

(2)

Enhancing the spatial connection of ecological areas: For type R4, it must be combined with the “micro-green-spaces” plan. Through the execution of strategic planning initiatives, rectify the disorganised allocation of green spaces, mitigate heat retention in densely populated areas, and guarantee that the advantages of these policies are fairly distributed to vulnerable communities.

(3)

Resource integration and spatial aggregation: For R1/R2 classifications, consolidate underutilised land using “urban renewal” initiatives. Convert fragmented idle land parcels into cohesive green infrastructure to improve environmental response stability.

6.3. Research Limitations and Prospects

While this study has achieved some breakthroughs in methodology, it is important to note the following limitations should be addressed in future research:

(1)

The dynamics of the time dimension: Contemporary studies predominantly utilise section data. Future research may utilise multi-period sensing imagery and long-period population data to explore the dynamic trajectories of block types across seasonal changes and urban evolution.

(2)

This study used statistical methods to examine asymmetric divergence and nonlinear correlation trends among environmental indicators. However, the constraints of satellite remote sensing limit these indicators to reflecting mainly regional-scale surface energy balance patterns. They do not adequately represent three-dimensional thermal conditions at the pedestrian level in subway-hub blocks. In complex transit-oriented developments, building shadows, street canyon effects, and micro-scale surface features significantly affect pedestrian thermal comfort, such as PET or UTCI. Remote-sensing metrics with a 30 m resolution cannot capture these micro-scale phenomena. Thus, the statistical connections remain physically opaque. To improve this, future studies should integrate micro-scale numerical simulations like CFD or ENVI-met and ground-based field measurements. These methods enable a comprehensive analysis of how block morphology shapes the wind–heat environment. This approach will strengthen the link between the statistical findings and their physical causes and better connect regional thermal indicators with individual comfort.

(3)

The expansion of social dimension: Present definitions of “context” predominantly depend on real estate values and traffic. Future research may incorporate crowd portrait data derived from mobile phone signals to further investigate social equity concerns related to environmental exposure.

This study’s significance lies in the creation and implementation of the PCRC analytical framework. This framework systematically delineates the nonlinear relationship between the distribution of green-space resources and environmental performances in high-density areas. The empirical identification of the “random spatial distribution of green spaces” offers data support and a conceptual foundation for enhanced governance. This supports the establishment of “park cities” and “compact cities” within megacities. The study facilitates a shift from macro-level indicator management to micro-level morphological interventions. This is significant for cities globally facing similar high-density challenges, giving the study considerable theoretical and practical value.