Multi-Scale Analysis of the Mitigation Effect of Green Space Morphology on Urban Heat Islands

Abstract

Urban green spaces (UGS) serve as critical mitigators of urban heat islands (UHIs), yet the scale-dependent mechanisms through which UGS morphology regulates thermal effects remain insufficiently understood. This study investigates the multi-scale relationships between UGS spatial patterns and cooling effects in Macao, employing morphological spatial pattern analysis (MSPA) to characterize UGS configurations and geographically weighted regression (GWR) to examine city-scale thermal interactions, complemented by patch-scale buffer analyses of area, perimeter, and landscape shape index effects. Results demonstrate that high-UGS-integrity areas significantly enhance cooling capacity (area with proportion of core ≥35% showing optimal performance), while fragmented elements (branches, edges) exacerbate UHIs, with patch-scale analyses revealing nonlinear threshold effects in cooling efficiency. A tripartite classification of UGS by cooling capacity identifies strong mitigation types with optimal shape metrics and cooling extents. These findings establish a tripartite UGS classification system based on cooling performance and identify optimal morphological parameters, advancing understanding of thermal regulation mechanisms in urban environments. This research provides empirical evidence for UGS planning strategies prioritizing core area conservation, morphological optimization, and seasonal adaptation to improve urban climate resilience, offering practical insights for sustainable development in high-density coastal cities.

1. Introduction

Urban heat island (UHI) intensification has emerged as one of the most pressing environmental challenges faced by contemporary cities under the backdrop of global climate change [1]. The elevated urban temperatures caused by UHI have been linked to adverse impacts on human health, air quality degradation, and vegetation stress, particularly in densely built environments [2]. Given these challenges, Nature-based solutions (Nbs), including optimizing green spaces are increasingly recognized as a strategy for addressing climate challenges, achieving carbon neutrality, improving biodiversity and enhancing social well-being [3,4]. Urban green spaces (UGS), comprising both artificial and natural greening (e.g., parks, street trees, and forest), play a vital role in regulating the microclimate through evapotranspiration, shading, and photosynthesis, thereby offering natural cooling benefits [5,6,7]. As such, the strategic integration of UGS into urban planning is essential, not only for alleviating UHI effects but also for reducing heat-related health risks and air pollution, and promoting environmental justice [8]. Beyond their climatic functions, UGS also support a biophilic urban vision, fostering connections between people and nature that enhance psychological well-being and social cohesion [9]. Therefore, the planning and design of UGS are fundamental to building climate-resilient, equitable, and livable cities.

The spatial configuration of UGS has been increasingly recognized as a determinant of its ecological and thermal performance [7,10]. Studies have shown that not just the size, but also the shape, connectivity, and distribution of UGS, significantly influence their cooling effects on urban environments [6,11]. Larger, aggregated, patchy and complex-shaped patches tend to be effective in cooling, while no consensus was reached based on patch-level metrics [12]. Furthermore, it has been acknowledged that the relationship between UGS areas and its cooling capacity is nonlinear, meaning that when exceeding a certain threshold, the cooling effects will decline. [11,13]. Given the tension in urban land use and the fading cooling effects, the design of UGS cannot be inefficient, especially in high-density cities. It is thus necessary to deepen the understanding of the relationship between the geometric morphology and cooling effect of UGS at patch level.

Although landscape metrics are widely used to quantify the configuration and composition of UGS, these indices fall short in visualizing spatial morphology and ecological connectivity, especially in complex urban contexts [14,15,16]. Furthermore, with ongoing urbanization, UGS is becoming increasingly fragmented and spatially isolated, leading to reduced cooling potential, particularly at broader spatial scales [10,17,18]. However, most studies have focused on high-temperature zones, despite the fact that UHI intensity varies dynamically and that low-intensity areas may evolve into or merge with higher-intensity zones over time [7]. Therefore, while landscape metrics can derive morphological indicators of UGS patches for cooling effect analysis, they provide limited insight into spatial relationships between patches and their cumulative effects on urban thermal environments. This limits their utility for formulating systematic ecological planning strategies to mitigate UHI at the city scale.

Compared to landscape metrics, morphological spatial pattern analysis (MSPA) offers a more informative alternative, as it not only quantifies the size and shape of patches but also visualizes their spatial layout [19]. As a pixel-based image analysis technique, MSPA identifies explicit landscape elements (e.g., core zones, branches, and bridges) using mathematical morphology operations [20], enabling a refined classification of UGS components that are both ecologically functional and spatially coherent [19,21]. Its application across multiple spatial scales, ranging from block-level to regional contexts, has been demonstrated in prior studies [22]. MSPA has also been widely applied in the construction of ecological and thermal networks, helping to extract ecological sources [21] or delineate “cold islands” and “heat islands” in urban settings [10,14,23,24,25,26,27]. Such research illustrates the potential of MSPA-based network perspectives to identify spatial morphologies that enhance or weaken UHI mitigation. Studies recognized that core zones (area with large and cohesive green patches), deliver the most substantial cooling effects, whereas fragmented units such as branches or islets often lack sufficient ecological function [23,25,28]. However, these approaches often fail to fully specify patch-level design parameters, such as area, shape, or fragmentation, that are necessary to guide practical interventions. Additionally, a systematic understanding of how different UGS configurations, particularly core zones (relatively large green spaces patches), contribute to UHI mitigation across seasons remains underdeveloped. Prior studies have shown that seasonal variations in solar radiation, hydrothermal conditions, and vegetation phenology significantly affect the relationship between surface temperature and UGS characteristics [29,30].

Previous research has predominantly focused on either local patch-scale or broader city-wide analyses, rarely integrating both levels. Specifically, most of studies addressing individual intervention without a systematic evaluation to interpret multi-scale cooling mechanism [31]. This overlooks potential scale-dependent effects and cross-scale interactions between regional UGS morphology and localized cooling performance. Overall, there are two key knowledge gaps regarding the relationship between UGS morphology and LST: First, there is a lack of research examining how different configurations of UGS, especially core zones, perform in mitigating UHI across distinct seasonal conditions. Second, few studies have jointly analyzed UGS–LST relationships at both city-wide and patch levels, despite the inherent scale sensitivity of thermal processes.

Understanding the thermal dynamics of UGS under varying spatial configurations and seasonal contexts is essential for climate-resilient planning, particularly in compact, high-density cities vulnerable to extreme heat. To address these gaps, this study proposes a dual-scale analytical framework to assess the cooling effects of UGS in Macau, a subtropical island city characterized by intensive development and limited ecological space. Specifically, this study aims to: (1) investigate the seasonal relationship between distinct MSPA-derived green space components and land surface temperature (LST) at the city scale; (2) analyze how UGS patch morphology (area, perimeter, and shape) affects cooling performance at the patch level; and (3) identify and classify UGS patches based on their observed summer cooling performance and extract the key morphological traits associated with high- or low-performing patches. By integrating city-scale spatial decomposition with localized patch-level evaluation, this research contributes to a systematic understanding of UGS cooling dynamics and provides actionable insights for optimizing green infrastructure in subtropical high-density urban environments.

2. Materials and Methods

2.1. Study Area

Macau is a coastal city located at the land–sea interface on the western shore of the Pearl River Estuary, bordering mainland China and facing the South China Sea (Figure 1). As one of the most densely populated urban areas globally, Macau is characterized by a compact urban form, intensive land use, and limited green spaces [32,33], making it a typical high-density Asian cities exposing to extreme heat risk [34]. The city experiences a humid subtropical monsoon climate (Cwa) under the Köppen classification, with pronounced thermal and hydrological seasonality due to strong monsoonal influences. Summers (May–September) are hot, humid, and characterized by frequent rainfall, while winters (December–March) are mild, dry, and generally sunny. This sharp seasonal contrast provides a natural experimental setting for investigating urban heat island dynamics across different climate conditions.

2.2. Data

Land surface temperature (LST) data were derived from Landsat 8 OLI_TIRS imagery with a spatial resolution of 30 m, acquired on 11 March and 3 September 2022. These dates were carefully selected to represent typical spring and summer conditions in Macao, based on both meteorological criteria and image quality. Heat fields are classified using the mean (μ) and standard deviation (std) method [30]. As shown in

Table 1. LST level classification based on mean-standard deviation method.

Level	Classification Criteria
High-Temperature Zone	LST > u + std
Sub-High-Temperature Zone	u + 0.5std < LST < u + std
Medium-Temperature Zone	u − 0.5std < LST < u + 0.5std
Sub-Low-Temperature Zone	u − std < LST < u − 0.5std
Low-Temperature Zone	LST > u − std

.

According to the long-term daily temperature records from 2012 to 2022 (Figure 2) and the reports from Macao Meteorological and Geophysical Bureau, Macao exhibits a clear seasonal thermal gradient with short transitional spring and autumn, while summer is characterized by persistently high temperatures and humidity. Based on the 25th–75th percentile threshold method for seasonal division [35], June to September was defined as summer, the statistically warmest period. Accordingly, September 3 falls within the summer, capturing high temperatures and high solar radiation intensity. In contrast, March 11 lies near the end of the cooler season and prior to the steep springtime temperature rise, effectively representing a relatively low thermal baseline. These two time points thus serve as effective thermal contrasts, enabling comparative assessment of UGS cooling performance under distinct seasonal conditions. Their selection was also based on the availability of high-quality, cloud-free imagery, ensuring reliable LST retrieval. The images were sourced from the U.S. Geological Survey (USGS) (https://landsatlook.usgs.gov/, accessed on 12 March 2024), specifically from Landsat Collection 2 Level-2 (LC2L2) products. These products provide surface temperature data retrieved from thermal infrared bands and have been pre-processed with radiometric calibration and atmospheric correction, enabling direct calculation of LST values [36]. As the temperature data are distributed in 16-bit integer format, spectral digital numbers were converted to Kelvin using scale factors and subsequently transformed to Celsius by subtracting 273.15.

Land use/land cover (LULC) data were obtained from Gaofen-1 imagery with a spatial resolution of 2 m, captured on 5 September 2022. The image was sourced from the China Centre for Resources Satellite Data and Application (https://data.cresda.cn/#/mapSearch, accessed on 29 August 2024). LULC classification was performed on the ENVI 5.6 platform, using a combination of supervised classification, visual interpretation, and field validation. The classification employed image-based analysis of spectral and spatial characteristics, resulting in seven LULC categories: arbor forest, shrubland, grassland, built-up areas, water bodies, roads, and bare soil. The accuracy of final output was assessed using the Kappa coefficient, which yielded a value of 0.94, indicating high reliability.

UGS patches were extracted manually based on the green facilities published by the Macau Nature (https://nature.iam.gov.mo/e/, accessed on 18 March 2024). The initial dataset included all publicly designated urban parks across Macao, covering a broad spectrum of park typologies such as country parks, municipal parks, and neighborhood parks. From this comprehensive set, patches were screened based on the following criteria: (1) an area larger than 900 m² (30 m × 30 m) to ensure detectability in Landsat-derived LST imagery [37]; (2) dominance of tree, shrub, or grassland vegetation types, as water can generate great impact on cooling performance and cause incorrect estimation [37,38]; and (3) clear and continuous spatial boundaries. The final selection of 32 patches represents vegetation-dominated UGS with relatively rich and dense plant cover, which are more likely to exhibit meaningful thermal regulation effects and are closely associated with residents’ well-being. These patches are spatially distributed across Macao’s administrative divisions, reflecting diverse park functions and morphological contexts. Consequently, they provide a reasonable and ecologically relevant basis for assessing patch-level cooling performance and morphological variation in a compact urban environment.

2.3. Methods

This study adopted a dual-scale analytical framework to investigate the relationship between UGS morphology and cooling effects. First, UGS morphological types were extracted city wide using MSPA, based on reclassified land use data. At the city scale, a grid-based approach was implemented to ensure spatial consistency, where LST values and the proportions of MSPA components were calculated for each grid. At the patch scale, geometric attributes of 32 representative UGS patches were quantified. Buffer analysis was then applied to derive four cooling effect indexes from LST data. Finally, statistical methods including Pearson correlation, Geographically Weighted Regression (GWR), and regression analyses were employed to assess the relationships between UGS morphology and LST patterns at both the city and patch scales.

2.3.1. Identify UGS Morphology at the City and Patch Scales

MSPA was used to quantify UGS spatial morphology at the city scale. To generate the binary raster map required for MSPA, LULC classification was first conducted using a supervised random forest algorithm on Gaofen-1 imagery. The Normalized Difference Vegetation Index (NDVI) was incorporated as an auxiliary variable to enhance the classification accuracy, particularly in distinguishing vegetated areas (arbor forest, shrubland, and grassland) from built-up or bare land. Based on the LULC results, a binary mask was created using the reclassification tool in ArcGIS 10.8: arbor forest, shrubland, and grassland were designated as foreground (value = 2), while all other land cover types were defined as background (value = 1). This binary raster was then used as the input for MSPA, conducted via GuidosToolbox (version 3.0), which applies a series of mathematical morphology operations (erosion, dilation, opening, and closing) to classify the spatial configuration of green space. In this process, the grid pixel size and the edge width was set to 30 m × 30 m and 1, respectively, corresponding to LST spatial resolution. The eight-neighborhood rule was applied. The foreground (i.e., vegetated UGS) was further subdivided into seven morphological categories: core, edge, perforation, islet, bridge, loop, and branch (Figure 3 and

Table 2. The classification of the MSPA and its definitions.

Type	Definition	Ecological Meaning
Core	A set of primitives whose foreground primitives are farther away from background primitives than a certain parameter of a specified size	Large natural patches, wildlife habitats, forest reserves, etc.
Islet	Patches that are not connected to any foreground area and whose area is smaller than the minimum value of the core area	Small, isolated, fragmented natural patches that are not connected to one another, often including small urban green spaces within built-up areas.
Perforation	Holes inside the center area, composed of background	Construction land within the core area that does not have ecological benefits.
Edge	Edges outside the foreground	The transition between the core area and the construction land has an edge effect.
Bridge	At least 2 points are connected to different core areas	The strips of ecological land connecting the core areas, i.e., the corridors in the regional green space, promote the migration of species, energy flow and network formation within the region.
Loop	At least 2 points are connected to the same core area	Ecological corridors connecting the same core area are small in scale and have low connectivity with surrounding natural patches.
Branch	Only one side is connected to the edge area, bridge area or loop area	Ecological patches that are only connected to one end of the core area have poor landscape connectivity.

). These categories reflect varying degrees of ecological integrity and spatial connectivity. For instance, core patches are continuous interior regions with higher ecological integrity, while edge and branch zones indicate fragmentation and potential exposure to thermal stress. These morphological components would be used as independent variables to analyze their association with LST.

Three metrics, area (A), perimeter (P), and the landscape shape index (LSI), were used to measures UGS geometric morphology at the patch scale. These metrics were calculated based on the spatial boundaries of 32 green space patches in ArcGIS. A and P are basic yet fundamental indicators of patch size and edge complexity, commonly employed in landscape ecology, while LSI serves as a proxy for shape complexity with higher value indicating more irregular or fragmented shape [39]. The geometric configuration of urban green space patches exhibits distinct spatial differentiation patterns mediated by anthropogenic and natural drivers. Patches embedded within highly urbanized areas are predominantly constrained by built-environment determinants, including road network geometry and building footprint distributions. In contrast, peri-urban patches demonstrate stronger dependence on biophysical constraints such as topographic gradients and land resource availability.

2.3.2. Quantify UGS Cooling Effects at the City and Patch Scales

Fishnet analysis was applied at the city scale to obtain spatially consistent LST values and UGS morphology features. After testing multiple grid sizes (300 m to 900 m), a resolution of 600 × 600 m was selected as optimal for capturing thermal variability and landscape heterogeneity at the Macau scale. The study area was thus divided into 157 uniform grid cells, within which the proportion of each MSPA component (e.g., core, edge, and branch) and the mean LST for two seasons (spring and summer) were calculated. This approach ensured spatial comparability and enabled localized assessment of UGS morphology–temperature relationships.

Multi-ring buffer analysis was employed at the patch scale to extract cooling performance indicators of UGS. For each of the 32 selected green space patches, ten concentric buffers were created at 30-m intervals, consistent with the resolution of Landsat-derived LST data. Four metrics were computed to characterize the thermal response of each green patch:

(1)

Average temperature within UGSs (AT): Defined as the mean temperature within each UGS, directly reflecting the thermal condition.

(2)

Immediate Cooling Magnitude (IC): Defined as the temperature difference between the green patch and its immediate surrounding buffer zone. This metric reflects the relative cooling intensity of the green space by comparing mean LST within the patch to the surrounding zone. This metric was developed to adapt to the highly compact and heterogeneous landscape environment of Macao and can accurately characterize the local cooling intensity of UGS.

(3)

Maximum Cooling Intensity (MC): Defined as the maximum temperature difference observed between any buffer zone and the patch interior. This metric captures the peak cooling capacity of the UGS, representing its strongest temperature gradient. This metrics is commonly used in patch-based UGS cooling effects analysis, especially for parks [36,40,41].

(4)

Cooling Distance (CD): Refers to the farthest buffer ring at which the LST difference between the green patch and its surroundings remains detectable. This indicates the spatial extent of the green space’s cooling effect. It is obtained through visual interpretation method, as this method is the most accurate, reliable and direction-sensitive recommended by recent research [42].

2.3.3. Analyze UGS–LST Relationships at the City and Patch Scales

At the city scale, after excluding grid cells with a core UGS proportion of less than 5% to ensure ecological relevance, the analysis was conducted based on 85 uniform grid cells. Pearson correlation analysis was used to assess the linear relationships between the proportions of each MSPA landscape component (core, edge, bridge, branch, loop, perforation, islet) and mean LST values during spring and summer. To further explore how UGS configuration modulates these relationships, the 85 grid cells were stratified into three equal-interval groups based on core green space proportion. Correlation analyses were then repeated within each group. Subsequently, to capture the spatial heterogeneity of UGS–LST interactions, Geographically Weighted Regression (GWR) was employed for variables that showed statistically significant correlations in the last step of the Pearson analysis. GWR allows for localized regression coefficients, thereby revealing spatially varying relationships between MSPA metrics and LST patterns. We imposed p < 0.01 across GWR models to ensure robustness against Type I errors from spatial multiple testing.

At the patch scale, regression analyses were conducted to examine the relationships between geometric metrics (A, P, and LSI) and four cooling effect indexes (AT, IC, MC, and CD). Linear and logarithmic regression were tested to determine the model yielding the best fit. however, we emphasize that regression results reflect statistical associations rather than causal relationships, as unmeasured confounding variables or reverse causality may influence the observed patterns [43]. Based on the principle of diminishing marginal utility, the threshold value of efficiency (TVoE) was defined as the most cost-effective patch area, beyond which increases in area no longer produce substantial gains in cooling intensity [11]. When the cooling effect follows an optimal logarithmic curve with respect to park area, TVoE is defined as the park area for which the slope of the curve is equal to 1. Increasing green space area produces a significant cooling effect until the TVoE value is reached, but the marginal effect decreases until it approaches 0 (Figure 4). In this study, TVoE was calculated as the point at which the slope of the logarithmic regression between area and MC equals 1, following the approach of previous studies [44,45].

The K-means clustering algorithm was proposed by MacQueen in 1967. As a fundamental partitioning method in clustering techniques, it has gained widespread acceptance due to its efficiency and maturity, offering relatively high computational performance. The algorithm works by first randomly selecting K points from the dataset as initial cluster centers. It then calculates the distance between each sample and the cluster centers, assigning each sample to the nearest cluster based on this distance. A key characteristic of the algorithm is that it evaluates the classification accuracy of each sample during every iteration. If misclassified, the sample is reassigned. After all samples have been processed, the cluster centers are updated, and the algorithm proceeds to the next iteration. This process continues until all samples are correctly classified, with no further adjustments needed, and the cluster centers stabilize [46]. In this study, based on normalized cooling metric data from 32 parks, we employed K-means clustering on the SPSS 26 platform to explore the categorization of parks’ cooling effect indexes.

3. Results

3.1. UGS Spatial Morphology and Seasonal LST Distribution at City Scale

The LULC distribution (Figure 5a) shows that the green spaces in the study area (including trees, shrubs, and grasslands) accounts for 35.51%. These green spaces are primarily located in the southern region, where they are more concentrated and cover larger areas. In contrast, the green spaces in the northern part are more scattered and cover smaller areas. Comparing the three types of land use, it is clear that the proportions are grasslands (14.6%) > arbor forests (13.03%) > shrublands (7.88%), indicating that grasslands are predominant in Macau’s urban development, while arbor forests are more widely distributed in natural green spaces.

LST analysis (Figure 5b,c) reveals notable seasonal and spatial variation in Macao’s thermal landscape. In spring, high-temperature zones are mainly concentrated in the central Taipa. The proportion of sub-high-temperature and sub-low-temperature areas is relatively high, while the high-temperature and low-temperature zones are relatively small. In summer, the proportion of high-temperature areas is larger, appearing in extensive clusters in the southern region and concentrated in the central area, with temperatures spreading outward from the center. On the northern main island, high-temperature areas are scattered and small, but compared to spring, the area of high-temperature zones has significantly increased. Low-temperature areas are mainly concentrated in the southern mountainous regions and northern water bodies, occupying a large proportion, while sub-low-temperature and sub-high-temperature areas occupy a smaller proportion. The proportion of medium-temperature areas has significantly increased. Comparing the land use types in Macau, their spatial distribution generally aligns with urban planning directions, indicating that urban development activities and socio-economic factors play an important regulatory role in LST.

The spatial morphology of UGS, derived from MSPA (Figure 3b) exhibits a typical “large patches with scattered fragments” pattern. At the city scale, the MSPA classification shows the following proportions: core (68.47%) > edge (20.71%) > branch (3.56%) > perforation (2.49%) > islet (2.33%) > bridge (1.69%) > loop (0.75%). Ecologically, core areas represent large, contiguous green patches with high vegetative density and strong internal cooling potential, such as country parks and forested hills, which are critical for generating stable cooling centers. Edge and branch areas, in contrast, are exposed to built-up surroundings and prone to heat intrusion, reducing their local cooling capacity. The small shares of loop and bridge suggest limited connectivity between green patches, which may restrict the propagation of cooling effects across urban areas. Meanwhile, the presence of numerous perforation and islet features indicates fragmented vegetation, which has been shown to offer lower thermal regulation due to reduced evapotranspiration and insufficient canopy continuity.

These MSPA-derived patterns reflect not only the physical structure of green space but also their potential to mitigate urban heat. In line with UHI theory, larger and more cohesive green cores contribute more effectively to ambient temperature reduction, while fragmented and disconnected elements tend to exacerbate local heat accumulation [10,16].

3.2. Spatial Morphology and Seasonal City-Wide Cooling Effects

The average temperature on the study day was analyzed for its correlation with the overall proportion of seven MSPA elements, resulting in 157 research units. After excluding areas within the units with less than 5% green space, 85 research units and two seasonal temperature maps were obtained. A correlation analysis was conducted between the average temperature within the research units and the proportion of each MSPA element, revealing their correlations and changes in strength (

Table 3. Correlation between the proportion of each MSPA indicator and average temperature across different seasons (Pearson coefficient).

MSPA Account	T (Spring)	T (Summer)
Core	−0.340 **	−0.537 **
Islet	0.303 **	0.352 **
Loop	0.006	−0.111
Bridge	0.342 **	0.388 **
Perforation	0.068	0.077
Edge	0.248 **	0.203
Branch	0.425 **	0.505 **

Note: ** p < 0.01.

).

In spring, only the proportion of the core area has a certain mitigating effect on urban heat islands. The overall average temperature of the 85 research units and the proportion of the core area show a significant negative correlation at the 0.01 level, while the proportions of the isolated area, edge area, bridge area, and branch area show a significant positive correlation at the 0.01 level. However, the correlation analysis between the proportion of the perforated area, loop area, and average temperature (T) indicates no significance at the 0.01 and 0.05 levels, with p-values close to 0. Therefore, it can be considered that changes in these two MSPA indicators do not affect the average temperature (T) in spring The strength of the correlation between spring MSPA indicators and T is expressed as branch > bridge > core > islet > edge > perforation > loop.

In summer, the UGS core area plays a better role in mitigating urban heat islands. The overall average temperature of the selected 85 research units and the proportion of the core area show a significant negative correlation at the 0.01 level, while the proportions of the branch area, isolated area, and edge area show a significant positive correlation at the 0.01 level. However, the correlation analysis between the proportions of the perforated area, loop area, and bridge area with the average temperature shows no significance at the 0.01 and 0.05 levels. Therefore, it can be considered that changes in these three MSPA indicators do not affect the average temperature. The correlation strength between summer MSPA indicators and average temperature is ranked as core > branch > bridge > islet > edge > loop> perforation.

Different spatial forms of the same type of MSPA elements alleviate urban heat islands to varying degrees. This study combined with the results of Table 3 and the characteristics of uniform data distribution. Following the principle of equal division to reduce error, based on the proportion of the core area with the strongest heat mitigation capability, the 85 study units are divided into three types: high UGS integrity, grid cells with core area proportion ≥35%, totaling 28 units; moderate UGS integrity, grid cells with core area proportion between 12–35%, totaling 29 units; and low UGS integrity, grid cells with core area proportion ≤12%, totaling 28 units.

Based on Pearson correlation coefficient analysis, significant MSPA indicators are further extracted and input into ArcGIS 10.8 for geographically weighted regression (GWR), mapping the significant indicators onto spatial locations.

3.2.1. High UGS Integrity (Grid Cells with Core Area Proportion ≥ 35%)

In high-UGS-integrity areas, the impact of MSPA elements on the urban heat island effect during summer is more pronounced. In the summer high-UGS-integrity areas, the proportion of core areas in UGS spatial patterns shows a significant negative correlation with T at the 0.01 level. The proportion of edge areas shows a significant positive correlation with T at the 0.01 level, and the proportion of branch areas shows a significant positive correlation with T at the 0.05 level, with no correlation to other MSPA indicators. In spring only the proportion of perforated areas shows a significant positive correlation with T at the 0.05 level, with no significant correlation to other MSPA indicators (

Table 4. Correlation between MSPA indicators with varying core area proportions and T (Pearson coefficient).

MSPA Account	T
	High UGS Integrity (≤12%)		Moderate UGS Integrity (12% < 35%)		Low UGS Integrity (>35%)
	Spring	Summer	Spring	Summer	Spring	Summer
Core	−0.020	−0.181	−0.463 *	−0.429 *	−0.120	−0.491 **
Islet	0.391 *	0.472 *	0.255	0.296	−0.025	0.145
Loop	0.158	0.150	−0.188	−0.135	0.404 *	0.256
Bridge	0.594 **	0.552 **	0.280	0.263	0.180	0.485 **
Perforation	−0.072	0.085	0.030	0.047	0.216	0.188
Edge	0.445 *	0.210	0.069	0.018	0.177	0.251
Branch	0.513 **	0.481 **	0.384 *	0.484 **	0.242	0.423 *

Note: * p < 0.05; ** p < 0.01.

). Analyzing the correlation results across the two seasons reveals that in areas with a larger proportion of core areas, when summer temperatures are generally higher, more complex UGS types have a more significant impact on surface temperature.

Further application of ArcGIS 10.8 to conduct GWR model regression analysis verifies how various MSPA elements significantly affect surface temperature. In high-UGS-integrity areas, there are more MSPA indicators significantly correlated with T during summer, and the significance is stronger. To enhance the accuracy of spatial regression analysis, this study selects three indicators for the geographically weighted regression analysis (GWR) model with T: the proportion of core areas, which has a negative correlation, and the proportions of edge and branch areas, which have positive correlations.

In high-UGS-integrity areas (core area proportion ≥ 35%), the GWR model fit (R²) for significant MSPA elements—core, branch, and edge areas—with T is relatively good (

Table 5. Significance of MSPA elements and T’s GWR analysis.

Classification	R2	Adjusted R-Square	AICc	SSE
High UGS integrity	0.43	0.29	120.66	65.00
Moderate UGS integrity	0.45	0.35	132.65	106.01
Low UGS integrity	0.39	0.34	102.85	45.57

), indicating that the GWR model results are relatively accurate, and the impact of MSPA elements on T is more significant in this type.

Relatively complete green spaces can effectively mitigate urban heat islands. Observing the relationship and distribution of the scale of core areas, edge areas, and branch areas on the thermal environment as shown by the GWR model (Figure 6a–c): If the area of the core area is proportionally increased in the southeastern coastal region and the central mountainous and grassland areas of high UGS integrity, the cooling effect near the southeast is more significant compared to the central area. The proportion of edge areas and branch areas is positively correlated with urban heat islands, with the correlation intensity of edge areas decreasing from northwest to southeast toward the center. Therefore, in the study area of high UGS integrity, increasing the scale of the core area while reducing the branch area in the southeast and the proportion of the edge area on the west side of the center will be beneficial for overall cooling. In high UGS integrity, most land types are forest land, with the southern forest land being larger and more contiguous compared to the northern built-up area, and the cooling effect of the core area in the south is better than that of the northern built-up area. In summary, in regions with larger forest land scales, the integrity of forest land has a significant impact on mitigating urban heat islands.

3.2.2. Moderate UGS Integrity (Grid Cells with Core Area Proportion Between 12% and 35%)

Complex land use types have a certain heat mitigation effect on MSPA elements. Within moderate UGS integrity, there are complex land use types, and the overall proportion of UGS is relatively low. The core area proportion in this region ranges from 12% to 35%, with a total of 29 research units identified. In this area, the correlation analysis between the proportion of MSPA elements and surface temperature T (Table 2) shows that in both spring and summer, the core area proportion has a significant negative correlation with urban heat islands at the 0.05 level. In spring, the proportion of branch areas is significantly correlated with T at the 0.05 level, whereas in summer, the proportion of branch areas shows a significant correlation with T at the 0.01 level. Comparing the correlation results of the two seasons with the land use types within moderate UGS integrity, it is evident that the land use types within moderate UGS integrity are more complex, with a larger scale of construction land. The heat mitigation capability of MSPA elements has certain limitations, but heat mitigation can also be achieved by adjusting some indicators.

Based on the analysis results in Table 4, further GWR model regression analysis was conducted to verify the relationship between the proportion of branch areas and T in summer (Figure 6e). The results indicate that in the moderate-UGS-integrity region, the areas with a significant positive correlation between the proportion of branch areas and urban heat islands are concentrated on both sides of the southern part of the southern island near the sea, showing an increasing trend from north to south. This suggests that in areas with a certain scale of green space construction, reducing the scale of branch areas can effectively enhance the heat mitigation capability of the region.

3.2.3. Low UGS Integrity (Grid Cells with Core Area Proportion ≤ 12%)

The type of MSPA elements has a negligible effect on mitigating urban heat island effects. This area is primarily located along the boundaries of large core zones and within smaller core zones, featuring extensive urban construction areas where the core zone accounts for less than 12%, significantly less than non-natural ecological elements. In this area, 28 research units were identified. Using SPSS 26 to analyze the correlation between the proportion of edge areas and branch areas in two seasons and T, the results showed that both the proportions of edge areas and branch areas exhibited a significant positive correlation at the 0.01 level, with low correlation with other MSPA elements. This indicates that in regions where core areas are significantly fewer, the scale proportion of various MSPA elements cannot explain the changes in surface temperature across different seasons. However, altering the proportion of edge areas and branch areas can help alleviate the thermal environment throughout the year.

Based on the analysis results, the more significant spring indicators were selected for further GWR model regression analysis in ArcGIS 10.8 to verify the relationship between the proportion of spring edge areas and branch areas and T (Figure 6f). The results indicate that in low-UGS-integrity areas, the positive correlation strength between edge areas, branch areas, and the thermal environment increases from north to south, and their relationship is strongly influenced by spatial heterogeneity. This suggests that in areas with a lower proportion of green space, the size of edge areas and branch areas has a relatively poor effect on the annual temperature influence.

3.3. Geometric Morphology and Seasonal Patch-Scale Cooling Effects

Based on the LULC data obtained in Section 3.1 (Figure 5a), data from 32 green spaces (Figure 7) indicates that their average area is 21.6 hectares, with 75% of the parks being smaller than 5.44 hectares. Four parks are larger than 59 hectares, and all of these are classified as Type A. Perimeter values ranged from 187.42 to 8089.25 m, with an average of 1438.6 m, and 50% of the parks had a perimeter less than 800 m. LSI values ranged from 1.06 to 2.52 (mean 1.34 ± 0.29), with 93.8% of the patches concentrated between 1.06 and 1.7, and only 1 patch (2.52) deviating significantly. This highly concentrated distribution pattern may be related to regularized artificial planning in the city (such as rectangular plots and grid-like road systems), resulting in most patches having shapes close to simple geometric figures. In contrast, suburban patches are influenced by topography, leading to more complex boundaries and larger areas, thereby limiting LSI variability.

3.3.1. Correlation Between Patch Characteristics and Cooling Metrics

Based on the geometric morphology indicators (A, P, LSI) of 32 parks, with a focus on the impact of internal and external thermal environments, a regression analysis was conducted between the three park shape indices and four cooling metrics (AT, IC, MC, and CD).

This study found that in both seasons, park area exhibited a significant logarithmic relationship with all four sets of cooling metrics (p < 0.01), indicating a nonlinear association between park size and thermal environment indicators (Figure 8). Specifically, there was a negative correlation with average surface temperature, and the cooling efficiency decreased as the area expanded. This phenomenon reveals a scale threshold effect in the thermal regulation efficiency of vegetation, where increasing park area beyond a certain point results in diminishing effects on cooling, suggesting that merely expanding the area is not an effective way to continuously enhance cooling benefits. There was a positive correlation with cooling magnitude, maximum temperature difference, and cooling range, indicating that as park area increases, there is a significant upward trend in cooling magnitude and range, suggesting that parks may expand the spatial range of cooling. Park area was negatively correlated with average surface temperature, meaning that as the area increases, there is a significant downward trend in surface temperature, suggesting that green spaces may lower surface temperature through evapotranspiration. Comparing the curves of summer and spring, it can be observed that the R² in summer is significantly higher than in spring, indicating that park area has a stronger explanatory power on cooling effects in summer, with park area being the primary factor influencing cooling metrics in summer, while other factors might be more influential in spring.

Statistical analysis indicates that park perimeter has a significant logarithmic relationship with cooling metrics (cooling magnitude, maximum temperature difference, and cooling distance) (p < 0.01). This suggests that although the cooling effect improves with the increase in edge length, due to diminishing marginal returns, this improvement will gradually stabilize. The negative logarithmic correlation between perimeter and average temperature further confirms this saturation effect, indicating that although increasing the edge contact area aids in heat exchange, the cooling efficiency significantly decreases once the perimeter exceeds a critical value. Notably, the R² value in summer is significantly higher than in spring, suggesting that the park perimeter has a stronger explanatory power for the cooling effect in summer, while in spring, other factors may have more influence.

The shape of the park has a varied relationship with the cooling effect. Compared to area and perimeter, LSI has a less effective cooling impact. Observing Figure 8c, it is evident that there is no strong correlation in either season, with the r-squared value approaching 0. The data indicate that the scatter distribution between LSI and surface temperature is randomly dispersed, the linear regression slope is nearly zero, and there are no significant seasonal differences.

The results indicate that the metrics of area and perimeter, which describe the shape of the park, are more effective than the LSI in determining the relationship between parks and temperature. They play a stronger role and better represent the relationship between parks and the surrounding and internal temperatures. The closer the area is to 0.48 ha, the better the cooling effect of the park. Beyond this threshold, the cooling effect is not significant. Compared to parks with more complex shapes, parks with area and perimeter within this threshold have a more pronounced effect on mitigating the heat island effect.

3.3.2. Classification of Patches Based on Cooling Performance

Based on the analysis in Section 3.3.1, the heat mitigation capability of parks is stronger in summer than in spring. Therefore, this section focuses on summer heat mitigation data. Through K-means clustering analysis, the normalized averages of cooling effect indicators and patch shape metrics were used to classify green spaces based on their cooling performance. The within-cluster sum of squares (WCSS) was calculated across a range of K values. A distinct “elbow point” was observed at K = 3, indicating that beyond this point, the rate of WCSS reduction significantly decreased. This suggests that K = 3 effectively captures the primary structure within the data. Furthermore, at K = 3, the average silhouette score (S = 0.7205) was the highest compared to other K values, confirming that the clusters were well-separated and internally cohesive at this partitioning. Consequently, the parks were categorized into three classes: Type A (Strong Mitigation), Type B (Weak Mitigation),and Type C (No Mitigation). Data for each type of green space are shown in

Table 6. The average value and name of various types of public parks reduce temperatures.

Types	Amount	Average Temperature (°C)	Coolong Range (°C)	Maximum Temperature Difference (°C)	Cooling Distance (m)
A	11	35.94	1.94	8.16	234.55
B	13	38.83	0.26	2.75	140.77
C	8	40.09	−0.48	1.91	105.00

.

In Type A (strong mitigation), 11 green spaces were selected. Spatially, these parks are mainly distributed in the southern natural mountainous areas and some northern mountains. This type of green space has the lowest overall normalized average temperature (<0.4), the largest cooling range, the highest maximum temperature difference, and the longest cooling distance, indicating the strongest cooling capability. Compared to other types of patches, this type of green space has a significant advantage in area, with a notably larger average size (44.75 ha). This indicates a significant positive correlation between the size of green spaces and their cooling effect (Figure 9A).

In Type B, 13 weak mitigation green spaces were selected, mainly distributed in the northern built-up areas. As shown in Figure 9B, the normalized average temperature value of this type of green space is between 0.5 and 0.8, indicating a certain degree of cooling. However, its maximum temperature difference is relatively high, and the cooling distance is relatively long, suggesting a certain cooling capability, though not significant. The cooling range is relatively low, between 0.2 and 0.4, indicating that the heat mitigation capability of this type is limited. By comparing land use, it can be observed that this type of green space is mainly distributed within built-up areas and is significantly influenced by surrounding buildings and human activities. Compared to other types of patches, this type of green space has a certain scale, with a larger area and perimeter.

Type C has identified a total of 8 non-mitigating green spaces. By examining the distribution in Figure 5, it can be found that the distribution is relatively scattered and mostly located near Type A. Comparing the normalized shape index and the normalized cooling index (Figure 9C), it is evident that the normalized average temperature index of green spaces in this type is generally above 0.5, indicating that it is close to or higher than the surrounding built-up areas, failing to form a significant cooling island effect. The normalized cooling range and maximum temperature difference are both below 0.2, indicating that this type of green space not only fails to effectively alleviate heat environmental pressure but may even lead to local temperature increases due to insufficient vegetation cover and anthropogenic heat emissions. This suggests that the surrounding environment has a more significant impact on temperature, and the green space does not perform its cooling function. The normalized cooling distance exhibits considerable volatility, indicating that its cooling capability is unstable, likely influenced by the density of surrounding buildings, surface materials, and the internal structure of the green space. Its area and perimeter are the smallest, with some sample indices approaching 0, indicating that its spatial scale is extremely small. This fragmentation and miniaturization characteristic prevent it from forming an effective cooling core area.

Research shows that the cooling effect of green spaces is closely related to their scale and the degree of urbanization of the surrounding area. Larger, naturally formed green spaces provide superior cooling effects, whereas small, regular artificial green spaces have limited impact in densely built areas.

4. Discussion

4.1. Impacts of Green Space on the Thermal Environment at the Regional Scale

Among the land cover types in cities, green space is often considered one of the most effective factors in mitigating urban heat island regulation [47]. Due to the complexity of land cover, this study classified green spaces into seven MAPS categories and found significant differences in the effect of different green space morphology types on UHI intensity. Only the core in the MSPA classification reflects mitigation of the heat island. Islet, edge, bridge and branch showed significant positive correlation with UHI. This is in contrast to other studies that have found green space core, perforation, and loop to have an impact on UHI in larger, inland cities [16,48]. The findings of this study differ from others on the green space cooling characteristics of high-density island cities.

In this study, the core area was found to have the strongest negative correlation with UHI intensity. As the core area of the green space is larger and has an advantage in size, it is the most effective type of green space form to mitigate the UHI effect. Consistent with the understanding of other scholars [49]. It also shows that the heat island mitigation effect exerted by the core area in a high-density island city exhibits consistent findings with other cities.

This study found that other types of green space did not play a significant role in mitigating the urban heat island, and from the results, some types may even exacerbate the urban heat island. This is consistent with the findings of Lin et al. [19] and Hong et al. [50]. In fact, it is not so much that these types exacerbate the urban heat island, but rather that the urban heat island is more influenced by other surrounding land cover. Smaller green spaces are generally surrounded by built-up land, such as street trees and low shrubs alongside or in the middle of traffic roads. Impervious paving, such as concrete or asphalt paving, has an extremely high LST and less perimeter shade, creating a more pronounced heat island environment. Therefore, other types of green space in the MSPA do not demonstrate mitigation for UHI. However, fewer studies have found similar conclusions, which may also be one of the differences between high-density island cities and other cities. The unused heat island mitigation effect caused by the scale or land use configuration of high-density urban green spaces should be of concern.

4.2. Cores Exhibits Greater Cooling Capacity in the Summer Months

We found that the effect of green space morphology type on UHI was more significant in the summer, with the cores showing greater cooling capacity in the summer. It is possible that this is due to the seasonal growth differences of plants [51]. The main vegetation type in Macao is subtropical broad-leaved evergreen forest. Plants, especially trees, grow better in summer and provide better transpiration shade. At the same time, heat island intensity is higher during summer months, providing greater potential for UGS to mitigate the heat island effect. Consequently, UGS generates stronger localized cooling effects in summer compared to March, more effectively reducing UHI. In March, plants are not yet fully foliated, and canopy growth remains limited. Additionally, Macao’s relatively cool and comfortable climate during this period results in low heat island intensity, preventing the cooling capacity of UGS from being effectively demonstrated [52]. This difference aligns with the seasonal variation in UHI intensity, demonstrating the intrinsic relationship between heat island intensity, green space characteristics, and seasonal climatic conditions. Therefore, seasonal variations in UHI and seasonal differences in plant-based heat mitigation performance should be considered when configuring vegetation in UGS during urban planning.

4.3. Differences in the Share of Cores Will Produce Different Heat Island Mitigation Effects

This study found that among all MSPA landscape components, only the “core” category demonstrates a significant cooling effect on urban heat. In contrast, fragmented elements such as branch, edge, and islet are positively correlated with higher LST, potentially exacerbating local UHI conditions. The correlation coefficients for the “core ratio” were consistently stronger than those for other MSPA metrics, underscoring the dominant role of intact vegetated cores as effective cooling sources in dense urban environments [19]. This is in line with previous research that suggests expanding core areas can amplify the overall urban cooling capacity [25,52].

To further explore spatial heterogeneity, the study classified all 85 grid cells into three typologies based on core area proportions. Distinct seasonal differences in the relationship between UGS configuration and thermal environment were observed across these types:

High-UGS-integrity regions are characterized by a high share of core UGS. These are typically large parks dominated by tree cover, often with topographic advantages that enhance cooling effects. For example, A6 (Seac Pai Van Park), the largest country park and statutory nature reserve in Macao, features dense vegetation and hilly terrain, contributing significantly to ecological, educational, and recreational functions. In such regions, preserving existing large core areas or incorporating adjacent mountainous areas further strengthens thermal regulation, particularly during summer.

Moderate-UGS-integrity regions contain moderate core proportions, mostly associated with medium-sized community parks. In these areas, cooling performance was weaker and less stable across seasons. For instance, B1 (Areia Preta Urban Park) mainly serves the surrounding residential population with multifunctional recreational spaces. Here, the influence of MSPA components on UHI was marginal in spring and limited to a weak positive association with the branch component in summer. The combination of limited ecological land and high built-up density may reduce the efficacy of green infrastructure in mitigating urban heat.

Low-UGS-integrity regions exhibit the lowest core proportions and are often smaller parks situated in compact built-up areas. For example, C3 (Central da Taipa Park) is a typical municipal parks in a high-rise residential zone with relatively sparse vegetation and flat topography, prioritizing urban amenities and active uses over ecological structure. These patches showed consistent positive correlations between edge/branch components and LST in both seasons, while the core had no significant impact. The concentrated built-up environment likely overrides the cooling potential of fragmented green spaces. In regions with greater built-up land coverage, particularly in high-density cities where building clusters are more concentrated, urban microclimates are more strongly influenced by dense built-up areas, resulting in higher UHI intensity. Similar results and interpretations were reached by Lin et al. [19].

These findings offer valuable insights for investigating the role of MSPA elements in UHI mitigation. Specifically, this research enables planners to prioritize both the structural characteristics and strategic placement of green spaces to maximize their cooling benefits.

4.4. Non-Linear Threshold Effects Between Various Metrics and Cooling Efficiency at the Patch Scale

Prevalent research has found that the cooling intensity of parks increases with park size [40]. In this study, the quantitative analysis of the cooling effect of green space area and perimeter found that it has a positive logarithmic relationship with the cooling amplitude, maximum temperature difference and cooling distance, indicating that with the expansion of the green space scale, the cooling benefit per unit area shows marginal decreasing characteristics. This results in separate TVoEs for summer and winter, which are larger in summer than in winter, indicating that the winter UGS achieves optimal cooling when the area is smaller. The cooling efficiency was significantly better in summer than in winter, mainly due to the higher ambient temperatures and solar radiation intensity in summer, which enhanced the transpirational cooling and shading effects of the vegetation. One study, through a review of the literature, found that the average TVoE for subtropical monsoon climate zones was 0.53 ha, and for temperate monsoon climate zones it was 0.43 ha [53]. The average TVoE for the subtropical monsoon climate zone is similar to that derived from this study, which is about 0.48 ha in summer and 0.3 ha in winter. However, one study found that the TVoE for a city park area in a subtropical climate zone was 1.08 ha [45]. The comparatively small urban extent of our study area may partially explain why the TVoE for UGS in this study differs from findings in other contexts. Our results suggest that UGS TVoE exhibits significant spatial variability across cities and climate zones. Beyond climatic factors, the spatial distribution of UGS and the morphological characteristics of the built environment may substantially influence urban ventilation patterns [53,54,55].

This study further reveals a significant negative logarithmic relationship between the mean internal temperature of green spaces and their area or perimeter metrics. While spatial expansion consistently lowers internal temperatures, the cooling efficiency exhibits a decay trend with increasing scale—a pattern aligned with the law of diminishing marginal utility. Additionally, our findings corroborate prior research [56,57] indicating that UGS perimeter optimization contributes measurably to localized temperature reduction. However, we did not find a significant relationship between LSI and maximum temperature difference and cooling distance. This is inconsistent with the conclusions reached by some studies [58,59]. This may be due to the limited variation in UGS shapes in Macao, which constrains the explanatory power of LSI. As shown in Figure 9, the compact urban morphology, characterized by high-density development and limited land availability, especially on the Macao Peninsula, results in relatively uniform patch configurations; Although patches in the less developed Coloane region show greater shape variability, their sample size is small. Furthermore, the city’s hybrid and tightly clustered urban blocks [60] may further restrict the geometric diversity of green spaces. These morphological constraints underscore the need for future studies in cities with more diverse UGS forms.

Our findings establish a critical theoretical foundation for identifying the scale-efficiency threshold in UGS planning. When designing urban parks, it is essential to incorporate cooling-performance thresholds, though UGS serves multiple functions beyond UHI mitigation. Specifically, urban green space planning should prioritize area and perimeter optimization, as increasing these dimensions can significantly enhance cooling efficiency. However, diminishing marginal returns must be evaluated to ensure cost-effectiveness and avoid resource overutilization.

4.5. Different Types of UHI Have Different Mitigating Capacities

Our findings demonstrate that expansive, ecologically oriented UGS, such as country parks or ecological landscape green spaces, constitute the optimal configuration for UHI mitigation. In contrast, small-scale, geometrically regular, and fragmented urban parks exhibit limited cooling efficacy. Notably, under specific conditions characterized by insufficient vegetation cover and elevated anthropogenic heat emissions, these fragmented green spaces may paradoxically function as localized heat sources.

Nevertheless, urban planning necessitates a multi-scalar approach rather than relying solely on a single green space typology. Consequently, complementary landscape pattern metrics—such as the leaf area index (LAI) [61]—should be incorporated to enhance the thermal regulation capacity of UGS. Future research will focus on elucidating the micro-scale landscape features within parks to quantify their nuanced effects on UHI.

4.6. Limitations of This Study

Our study has some limitations. First, although many studies have used LST as an analogue to UHI [62], LST is not exactly equal to air temperature. Due to the difficulty in obtaining air temperature, our study used LST as an analogue to air temperature, which may deviate from the actual situation. Especially in high-density cities, building shadows may have a greater impact on LST.As some studies have found, the large shaded area of a building can have a significant impact on LST, potentially optimizing shading and evapotranspiration benefits [6,63]. We are trying to break through this technical challenge of obtaining more air temperature in our subsequent research through the combination of field measurements and numerical simulations. Second, our study is no way to analyze the thermal environment during the hottest months due to satellite image availability. In the future, more image data will be collated to the point where we can study the relationship between green space and UHI in hot conditions.

5. Conclusions

This study establishes a multi-scale assessment framework to evaluate the thermal effects of urban green spaces (UGS) using multi-source remote sensing data. We demonstrate that core areas significantly mitigate urban heat islands (UHIs), particularly in summer, while fragmented patches (e.g., edges, branches, and isolated greens) may exacerbate warming. Park size and perimeter strongly influence cooling capacity, with larger UGS exhibiting more consistent cooling effects, whereas shape complexity shows negligible impact. Optimizing UGS layout—prioritizing large, contiguous, and structurally compact core patches—can enhance urban resilience. Our findings provide actionable insights for high-density city planning, though higher-resolution thermal data and mechanistic analyses are needed for further refinement. The proposed framework offers a transferable strategy for sustainable urban development in warming climates.