Testing a Novel Multi-Temporal Multidimensional Assessment of Cooling Performance for Blue, Green, and Grey Parks: A Case Study in Wuhan, China

Abstract

Urban parks are “cool islands” for mitigating urban heat, yet most snapshot-based assessments overlook intraday cooling dynamics and divergent mechanisms across park typologies. This study examines 52 parks in Wuhan, a humid city with routine park irrigation, using thermal data from Landsat 9 and ECOSTRESS across morning, noon, and nightfall. Through stepwise analysis and blue–green classification, we quantify diurnal cooling dynamics and their drivers. While previous studies have examined diurnal (within-day) cooling, multidimensional indicators, or scale effects separately, our contribution lies in establishing a multi-temporal assessment framework that integrates temporal dynamics with blue, green, and grey park typologies to reveal how cooling patterns diverge across blue, green, and grey parks throughout the day. Results show park cooling intensity (PCI) and gradient (PCG) peak at noon, while cooling area (PCA) remains stable. Elevated cooling efficiency (PCE) at nightfall is driven not by ecological cooling, but by the rapid thermal response of impervious surfaces with low thermal inertia. Area, greenspace proportion, and building height are primary drivers, shifting from scale dominance in the morning to vegetation and building at noon, with a preliminary transition range of approximately 14–16 hm² identified for this regime shift, though this finding warrants further validation with larger samples. Based on blue–green composition, parks are categorised as blue, green, or grey, with divergent cooling dynamics due to thermophysical properties. Blue parks cool steadily all day, green parks peak at noon, while grey parks’ elevated PCE at nightfall is an apparent thermal response, not ecological cooling. Typological heterogeneity weakens models that pool all parks together, as water storage, vegetation evapotranspiration, and impervious thermal response vary across types and cancel out when pooled. Findings show that single-time-phase or full averaging insufficiently captures park cooling dynamics, underscoring the value of considering both diurnal and typological variations in climate-adaptive planning for dense cities.

1. Introduction

Rapid urbanisation has expanded impervious surfaces and intensified built-up density, substantially modifying the thermophysical properties of urban surfaces. These changes exacerbate the urban heat island (UHI) effect and increase the frequency of summer heatwaves, affecting residents’ thermal comfort and urban ecological security [1,2]. As a megacity located on the middle and lower Yangtze River, Wuhan has a high concentration of population and buildings within its Third Ring Road. Summer thermal stress and the demand for blue–green space cooling are particularly pronounced, further intensifying surface heat accumulation [3,4]. Blue–green infrastructure offers a key spatial strategy for mitigating urban heat stress and represents a critical research frontier in landscape architecture for climate change adaptation. Urban parks, as a core component of blue–green infrastructure, regulate local microclimates through vegetation shading and transpiration, as well as evaporative cooling from water bodies. These processes provide a low-cost, environmentally sustainable approach to mitigating the UHI effect and improving urban liveability [5,6]. Quantifying the park cooling effect in Wuhan can offer quantitative evidence to inform climate-adaptive park planning in high-density cities [7,8].

Land surface temperature (LST) is a key parameter characterising the thermal state of the land surface. LST inversion using thermal infrared remote sensing is widely used to quantify the cooling effect of urban parks. Using medium-resolution thermal infrared data such as Landsat, researchers have examined the park cooling effect. Evaluation indicators have been developed, with a focus on cooling intensity and spatial extent. Through the analysis of internal landscape features and external built environment characteristics, key factors have been identified as important regulators of the cooling effect. These factors include park size, vegetation cover, water body proportion, and building morphology. These findings provide a foundation for further research into the key drivers of park cooling performance [9,10].

Constrained by data source characteristics and traditional analytical methods, existing studies still leave three scientific questions unresolved. This hinders efforts to capture the diurnal dynamics of park cooling and to provide robust quantitative evidence for climate-adaptive park planning. First, a temporal bias persists in existing assessments. Owing to the fixed overpass times of polar-orbiting satellites and the reliance on cloud-free imagery, most studies are grounded in single-date, single-temporal remote sensing data. This static approach fails to capture the diurnal dynamics of park cooling under fluctuating solar radiation and varying surface energy exchange [11,12,13]. Consequently, the assessment results are highly time-dependent and may yield divergent conclusions depending on the timing of observations. Second, current quantification systems tend to rely on single-dimensional indicators, such as cooling intensity and maximum influence range, while lacking a multidimensional framework that integrates spatial extent, unit efficiency, cumulative effect, and attenuation gradient. This narrow scope obscures the overall characteristics and inter-park variability of cooling performance, fostering a one-sided understanding that falls short of the multi-objective demands of urban green space planning [8,14]. Finally, existing analyses of driving factors remain largely confined to pairwise correlation, with limited systematic identification of spurious correlations arising from multicollinearity. This makes it difficult to isolate the core dominant factors and their independent contributions from the complex interrelationships. Furthermore, most studies treat parks as a homogeneous sample, without distinguishing the divergent driving pathways associated with different blue–green space compositions. Under mixed modelling, these pathways may cancel each other out, thereby masking key mechanisms. Consequently, the resulting conclusions offer limited practical guidance for differentiated planning and design [15,16,17].

Wuhan is a high-density waterfront city in the middle and lower Yangtze River, historically known as the “City of a Hundred Lakes”. Many of its urban parks integrate both water bodies and green spaces, forming a complex land surface composition that differs markedly from the predominantly green parks in northern China. This makes Wuhan an ideal empirical site for park cooling effect research. Wuhan’s subtropical monsoon climate, coupled with rapid urbanisation, exacerbates hot, humid summers and a pronounced UHI effect. The cooling capacity of parks and blue–green spaces therefore constitutes a key ecological strategy for alleviating urban heat stress. The city is currently advancing its “Wetland and Flower City” park city model. Policy frameworks such as the Wetland and Flower City Implementation Plan (2023–2027) and the 14th Five-Year Climate Plan have identified enhancing park cooling performance as a core objective of urban thermal environment optimisation, along with assessment and design requirements for park cooling [3]. However, existing local studies in Wuhan remain confined to the conventional single-phase, single-dimensional analytical paradigm [3]. This impedes the identification of diurnal cooling dynamics, which are essential for guiding park area management and typology-based design.

To address these gaps, this study investigates how the four cooling metrics evolve across morning, noon, and nightfall, which factors independently drive these metrics and how their contributions shift diurnally, and whether parks with different blue–green compositions exhibit divergent cooling patterns. We hypothesise that PCI and PCG peak at noon due to enhanced vegetation transpiration and thermal contrast, while PCA remains relatively stable and the elevated PCE at nightfall reflects impervious surface thermal response rather than ecological cooling; that the dominant drivers shift from park area in the morning to vegetation and building factors at noon, with a potential transition range marking this shift; and that blue, green, and grey parks exhibit distinct diurnal patterns driven by their differing thermophysical properties. The novelty of this study is its integration of diurnal thermal dynamics with blue–green park typologies, an approach that has been relatively underexplored in previous park cooling research. To this end, we propose a multi-temporal assessment framework that moves beyond conventional single-phase evaluations by capturing diurnal (within-day) cooling dynamics across different park typologies.

Accordingly, this study used 52 urban parks in central Wuhan to build a multi-temporal framework using multi-source remote sensing data, quantifying diurnal cooling dynamics. Partial correlation analysis with covariate control identified the independent contributions of core drivers and their diurnal shifts. Parks were further grouped by blue–green composition and size to compare cooling patterns across typologies. The study aimed to reveal the diurnal dynamics and typological heterogeneity of park cooling in high-density settings, informing critical size thresholds and blue–green space allocation [18,19].

2. Materials and Methods

2.1. Study Area

Wuhan (113°41′ E–115°05′ E, 29°58′ N–31°22′ N) is located in the eastern Jianghan Plain, where the Yangtze and Han rivers converge. This convergence has formed a dense network of waterways and lakes, creating a typical “river city” landscape. The study area has a low-lying and flat terrain, with an average elevation of approximately 23 m. This natural foundation supports urban spatial development.

Wuhan has a typical subtropical monsoon climate, with hot, humid summers and heavy rainfall. According to the Chronicles of Wuhan, the city’s mean annual temperature ranges from 16.2 to 16.7 °C, while July average temperatures range from 28.2 to 29.1 °C [20]. Rapid urbanisation and high-density development have led to extensive impervious surfaces in the central urban area. The UHI effect exhibits pronounced spatial heterogeneity, with particularly strong heat island aggregation within the Third Ring Road core [21,22].

2.2. Park Sample Selection

Wuhan has a humid subtropical monsoon climate, with hot, humid summers and abundant summer rainfall. To ensure accurate cooling capacity assessment and sample independence, parks within Wuhan’s Third Ring Road were selected from the Statistical Table of Major Urban Parks (Green Spaces) in Wuhan (2023), published by the Wuhan Bureau of Landscaping and Forestry, using Google Earth Pro 7.3.6 imagery. The selection followed three criteria. First, parks adjacent to the main channels of the Yangtze and Han rivers or to large lakes were excluded to prevent large-scale, cool, moist air masses from masking or interfering with the cooling signals of individual green spaces. Second, a minimum area threshold of 900 m² was applied, based on the 30 m spatial resolution of the Landsat thermal infrared band, to reduce mixed-pixel noise. Third, affiliated and low-lying green spaces were excluded to ensure that each sample had an independent thermal exchange boundary.

The study focused on central Wuhan and ultimately selected 52 urban park samples, with 41 within the Third Ring Road and 11 in the contiguous built-up area along its perimeter. All samples met the uniform criteria and exhibited no significant inter-group differences in surrounding built environment characteristics, confirming homogeneity. Figure 1 shows the spatial distribution of LST and park sampling points across the study area. The samples cover the core urban districts within the Third Ring Road, including Jianghan, Jiang’an, and Qiaokou, where LST is highest, thereby demonstrating the representativeness of the sample distribution.

Given Wuhan’s humid subtropical monsoon climate with abundant summer rainfall and routine park irrigation during dry spells, the midday cooling peak of green parks observed in this study is context-specific and may not hold in arid or unirrigated settings.

2.3. Data Acquisition and Processing

Table 1. Data descriptions and sources.

Data Category	Details	Source	Usage Notes
Spatial data	Boundaries and basic attributes of 52 urban parks	Wuhan Municipal Bureau of Landscaping and Forestry; Tianditu (https://www.tianditu.gov.cn)	Park selection and sample identification
Land surface temperature	Midday land surface temperature	Landsat 9 Collection 2 Level-2 Surface Temperature (ST) Product, USGS EarthExplorer, 30 m spatial resolution	Quantification of midday park cooling metrics
	Morning and evening land surface temperature	ECOSTRESS Level 2 Land Surface Temperature and Emissivity (LSTE) products, 11 August 2022 (09:23 CST, morning) and 9 August 2023 (17:30 CST, evening), NASA Earthdata, 70 m original spatial resolution	Quantification of morning and evening park cooling metrics for intraday dynamic analysis
Landscape attribute data	DEM, land cover (green space, water body, impervious surface), and vegetation structure metrics (NDVI, FVC, LAI)	Land cover and vegetation metrics are derived from the same Landsat 9 Level-2 imagery (30 m) via supervised classification, band ratio calculation, a dimidiate pixel model and an empirical vegetation index model; DEM is obtained from the SRTM 30 m digital elevation product and spatially co-registered with Landsat imagery	Calculation of internal park elevation, land cover proportions and vegetation structure indicators for driving factor system construction
Built environment data	Building outlines and heights, road network vectors	OpenStreetMap (OSM) Public Datasets	Calculation of external built environment metrics (e.g., average building height, road density) within buffer zones around parks

Notes: ① All datasets are projected to the WGS_1984_UTM_50N coordinate system and clipped to the scope of Wuhan’s central urban area. ② The original 70 m ECOSTRESS LST data were resampled to 30 m using spline interpolation to match the spatial resolution of Landsat 9 data. ③ Constrained by frequent summer cloud cover and satellite orbital characteristics, cloud-free scenes with similar summer vegetation phenology were selected for multi-temporal analysis to ensure comparability of underlying surface conditions across periods.

details the categories, sources, and applications of the multi-source spatial data. All data were standardised through preprocessing and applied to the multi-temporal quantification of the park cooling effect and its driving factors. All datasets were projected to the WGS 1984 UTM Zone 50N coordinate system and clipped to Wuhan’s central urban area to ensure consistency in spatial reference and analytical extent. To address spatial resolution mismatches among surface temperature products, ECOSTRESS data were resampled to 30 m using spline interpolation, ensuring thermal data comparability across the three periods. To match satellite overpass times, three daytime periods were defined: morning (09:23), noon (10:55), and nightfall (17:30), corresponding to the overpass times of ECOSTRESS and Landsat 9, respectively. Owing to Wuhan’s cloudy summer weather and Landsat’s fixed overpass cycle, the images selected for the three periods were not acquired on the same day. Nevertheless, they represent the best available cloud-free observations under the current data conditions [23,24]. This strategy registers multi-source images to a single Landsat overpass date. While it entails a trade-off in temporal consistency, it ensures land surface comparability across periods and outperforms the random selection of images from different dates. Previous studies have shown that multi-temporal ECOSTRESS data can be used to analyse urban thermal diurnal dynamics [25,26].

To verify the comparability of meteorological conditions across the three image dates, daily data for the Wuhan station were obtained from the China Meteorological Data Network (

Table 2. Comparison of key meteorological parameters for the three image dates.

Image Date	Daily Mean Air Temperature (°C)	Sunshine Duration (h)	Daily Mean Wind Speed (m/s)
11 August 2022	33.3	11	1.5
5 July 2024	32.3	11.8	3.5
9 August 2023	31	9.4	2.6

Note: Data were obtained from the China Meteorological Data Service Centre (https://data.cma.cn) for Wuhan Station (Station ID: 57494).

). The daily mean temperature range was 2.3 °C, well below Wuhan’s typical summer diurnal range of 6–10 °C [20]. Sunshine exceeded 9 h per day under clear skies, and average wind speeds of 1.5–3.5 m/s were classified as light wind. These results confirm meteorological consistency and the suitability of the imagery for comparative daytime analysis.

2.4. Research Methods

The technical approach of this study is illustrated in Figure 2 and comprises four main stages. First, multi-source remote sensing data were pre-processed, and LST was retrieved. Second, the significance of the park cooling effect was tested and quantified using multidimensional indicators. Third, stepwise statistical analysis was applied to identify driving factors. Fourth, parks were classified by type and their cooling patterns were compared.

The four analytical stages are detailed in Figure 2 and further described in the following subsections. Here, “type” refers to blue, green, and grey parks, and “cooling patterns” refer to the diurnal variations in PCA, PCE, PCI, and PCG across the three periods.

2.4.1. LST Derivation for Multiple Daytime Periods

To obtain LST for the three daytime periods, three periods were defined according to satellite overpass times: morning, noon, and nightfall. For the midday period, the Landsat 9 Collection 2 Level-2 product was used, providing atmospherically corrected LST in Kelvin from pixel brightness values. Morning and nightfall LST data were sourced from the ECOSTRESS L2 product (70 m) and resampled to 30 m using spline interpolation to match Landsat 9 [27,28]. All images were projected to WGS 1984 UTM Zone 50N and clipped to Wuhan’s central urban area, covering the Third Ring Road and its periphery.

2.4.2. Park Cooling Effect Measurement

The buffer averaging method quantified the park cooling effect. As shown in Figure 3, ten buffer zones were established at 30 m intervals from the park boundary, with a total buffer distance of 300 m. This spacing matches the 30 m resolution of the Landsat 9 thermal infrared band and captures temperature gradients near park boundaries. A 300 m buffer has been validated in high-density urban studies as the optimal range, balancing model accuracy with spatial independence, and was therefore adopted directly [29,30]. To determine the presence of an observable cooling effect, each park was tested across the three periods. Using buffer distance (r, m) as the independent variable and mean LST within each buffer (T(r), °C) as the dependent variable, a simple linear regression model was constructed:

$$T\left(r\right)=\alpha +\beta r$$

(1)

If β > 0 and is statistically significant (p < 0.05), LST increases significantly with distance, confirming a marked cooling effect on the surroundings. Parks meeting this criterion are retained for subsequent analyses; those that do not are excluded.

For these parks, a third-order polynomial was fitted to the variation in LST with buffer distance using the following equation:

$$Tr = a{r}^3+ b{r}^2+ cr + d$$

(2)

where r is the buffer distance (m) and T(r) is the mean LST within the buffer (°C). The first turning point, where the first derivative is zero, defines the maximum cooling distance L, marking the park’s effective cooling range. Beyond L, the cooling effect is negligible. Based on the fitted curve and L, four indicators were defined:

$$PCA=Smax$$

(3)

$$PCE={\displaystyle \frac{Smax}{Spark}}$$

(4)

$$PCI={\displaystyle \frac{L\times T_L-{\int }_0^{L}T\left(r\right)dr}{L\times T_L}}$$

(5)

$$PCG={\displaystyle \frac{L\times T_L-{\int }_0^{L}T\left(r\right)dr}{L}}$$

(6)

In the equation, r is the buffer distance (m), and T(r) is the mean LST within each buffer zone (°C). The first turning point, where the first derivative of T(r) is zero, defines the cooling distance L and the background LST (TL, °C) beyond the cooling range. Smax is the buffer area within L (m²), representing the maximum spatial extent of the cooling effect, while Spark is the park area (m²). Four indicators were derived from these parameters: PCA for spatial extent, PCE for cooling area per unit area, PCI for the cumulative cooling effect on the surroundings, and PCG for the attenuation of cooling with distance. PCA and PCE are maximum-perspective indicators, capturing spatial extent and unit-area efficiency. PCI and PCG are accumulation-perspective indicators, reflecting total intensity and attenuation. Together, these four indicators evaluate cooling performance across four dimensions: spatial extent, efficiency, intensity, and attenuation gradient.

2.4.3. Factors Influencing Park Cooling Effects

To identify the key drivers of the park cooling effect and their influence patterns, influencing factors were selected from internal landscape features and the external built environment [31,32]. The definitions, formulas, and mechanisms for each variable are given in Table 3. All external built environment indicators were computed within a 300 m buffer around each park to ensure consistent spatial scale and comparability across samples [33].

To identify the key drivers of the park cooling effect and their mechanisms, a stepwise statistical analysis was conducted using valid park samples, with the four cooling indicators as dependent variables and candidate factors as independent variables. First, Spearman’s rank correlation was used to screen factors against cooling indicators across time periods (p < 0.05), eliminating invalid factors and retaining key variables. Collinearity was then assessed using a threshold of VIF < 5 to ensure that factors retained for subsequent analysis had sufficient statistical independence. Following core factor identification through initial screening and collinearity diagnosis, partial correlation analysis was applied to control for confounders, and net correlation coefficients were calculated to rule out spurious associations and validate independent contributions [34]. Subsequently, multiple linear regression models were constructed with core factors as independent variables and PCA and cooling PCI as dependent variables. Standardised regression coefficients were calculated to quantify the relative contribution of each factor. Parks were then divided into large and small groups based on the median park area for each time period [35]. Within-group partial correlations between core factors and cooling area were computed, and inter-group comparisons were used to identify critical intervals marking structural shifts in driving patterns. Given the limited sample size after grouping, this analysis focused on trend identification rather than precise threshold estimation.

Table 3. Influencing factors affecting park cooling effects.

Variable Category	Variable	Formula	Description	Mechanism of Influence
Internal Geometry
	Park Area (AREA)	${\mathrm{A}}_{\mathrm{p}}\left(\mathrm{h}{\mathrm{m}}^2\right)$	The park’s area	Larger areas provide more vegetation and water surfaces, enhancing cooling [9,10].
	Park Perimeter	$\mathrm{P}\left(\mathrm{m}\right)$	Total length of the park’s perimeter	The longer the perimeter, the larger the interface between the park and its surrounding environment, which facilitates heat exchange [16,17].
	Shape Index	${\displaystyle \frac{\mathrm{P}}{2\sqrt{\mathsf{\pi }\mathrm{A}}}}$	The Complexity of Park Shapes	Complex shapes can increase the park’s contact with its surroundings, promoting the dispersion of cool air, but may reduce the efficiency of the internal heat island [16,17].
Internal Landscape Composition
	Green Space Percentage	$\left({\displaystyle \frac{A_{green}}{A_p}}\right)\times 100\mathrm{\%}$	Percentage of green space within the park	Green spaces lower temperatures via transpiration and shading; higher proportions enhance cooling [5,6].
	Water Body Percentage	$\left({\displaystyle \frac{A_{water}}{A_p}}\right)\times 100\mathrm{\%}$	Percentage of the park covered by water bodies	Water’s high heat capacity buffers temperatures via evaporation and absorption, especially at noon [35].
Internal Vegetation Structure
	Fractional Vegetation Cover (FVC)	${\displaystyle \frac{NDVI-NDV{I}_{soil}}{NDV{I}_{veg}-NDV{I}_{soil}}}$	The proportion of the ground covered by vegetation canopy per unit area	Higher FVC strengthens transpiration and shading, lowering surface temperature [5,6].
	Leaf Area Index (LAI)	${\displaystyle \frac{A_{leaf}}{A_{ground}}}\left({\mathrm{m}}^2·{\mathrm{m}}^{-2}\right)$	Density of vegetation foliage	Higher LAI enhances canopy radiation interception and transpiration, amplifying cooling [5,6].
Internal Terrain
	Average Elevation	${\displaystyle \frac{{\sum }_{i=1}^{n}DE{M}_i}{n}}\left(\mathrm{m}\right)$	Average elevation within the park	Higher elevation lowers ambient temperature, enlarging the thermal gradient between the park and surroundings and amplifying the cooling effect [4,36].
External Built Environment
	Average Building Height	$\left({\displaystyle \frac{{\sum }_{i=1}^n{H}_i}{n}}\right)\left(\mathrm{m}\right)$	Average height of buildings within the 300 m buffer zone surrounding the park	High-rise buildings may block sunlight, but they also impede airflow; the impact on thermal comfort is complex [33].
	Impervious Surface Percentage	$\left({\displaystyle \frac{A_{impervious}}{A_{buffer}}}\right)\times 100\mathrm{\%}$	Percentage of impervious surfaces within the 300 m buffer zone surrounding the park	High impervious cover exacerbates UHI, but its rapid diurnal cooling may produce apparent cooling in remote sensing [26,37].
External Transport
	Road Density	$\left({\displaystyle \frac{L_{road}}{A_{buffer}}}\right)\left({\displaystyle \frac{\mathrm{k}\mathrm{m}}{\mathrm{k}{\mathrm{m}}^2}}\right)$	Road network density within the 300 m buffer zone surrounding the park	High road density increases anthropogenic heat and UHI intensity, but may improve ventilation [16].

Note: ① DEM, FVC, LAI, impervious surface percentage, and land cover types derived from Landsat 9 Level-2 imagery (30 m); ② road data from OSM; ③ symbols in formulas correspond to variables in the table.

Table 3. Influencing factors affecting park cooling effects.

Variable Category	Variable	Formula	Description	Mechanism of Influence
Internal Geometry
	Park Area (AREA)	${\mathrm{A}}_{\mathrm{p}}\left(\mathrm{h}{\mathrm{m}}^2\right)$	The park’s area	Larger areas provide more vegetation and water surfaces, enhancing cooling [9,10].
	Park Perimeter	$\mathrm{P}\left(\mathrm{m}\right)$	Total length of the park’s perimeter	The longer the perimeter, the larger the interface between the park and its surrounding environment, which facilitates heat exchange [16,17].
	Shape Index	${\displaystyle \frac{\mathrm{P}}{2\sqrt{\mathsf{\pi }\mathrm{A}}}}$	The Complexity of Park Shapes	Complex shapes can increase the park’s contact with its surroundings, promoting the dispersion of cool air, but may reduce the efficiency of the internal heat island [16,17].
Internal Landscape Composition
	Green Space Percentage	$\left({\displaystyle \frac{A_{green}}{A_p}}\right)\times 100\mathrm{\%}$	Percentage of green space within the park	Green spaces lower temperatures via transpiration and shading; higher proportions enhance cooling [5,6].
	Water Body Percentage	$\left({\displaystyle \frac{A_{water}}{A_p}}\right)\times 100\mathrm{\%}$	Percentage of the park covered by water bodies	Water’s high heat capacity buffers temperatures via evaporation and absorption, especially at noon [35].
Internal Vegetation Structure
	Fractional Vegetation Cover (FVC)	${\displaystyle \frac{NDVI-NDV{I}_{soil}}{NDV{I}_{veg}-NDV{I}_{soil}}}$	The proportion of the ground covered by vegetation canopy per unit area	Higher FVC strengthens transpiration and shading, lowering surface temperature [5,6].
	Leaf Area Index (LAI)	${\displaystyle \frac{A_{leaf}}{A_{ground}}}\left({\mathrm{m}}^2·{\mathrm{m}}^{-2}\right)$	Density of vegetation foliage	Higher LAI enhances canopy radiation interception and transpiration, amplifying cooling [5,6].
Internal Terrain
	Average Elevation	${\displaystyle \frac{{\sum }_{i=1}^{n}DE{M}_i}{n}}\left(\mathrm{m}\right)$	Average elevation within the park	Higher elevation lowers ambient temperature, enlarging the thermal gradient between the park and surroundings and amplifying the cooling effect [4,36].
External Built Environment
	Average Building Height	$\left({\displaystyle \frac{{\sum }_{i=1}^n{H}_i}{n}}\right)\left(\mathrm{m}\right)$	Average height of buildings within the 300 m buffer zone surrounding the park	High-rise buildings may block sunlight, but they also impede airflow; the impact on thermal comfort is complex [33].
	Impervious Surface Percentage	$\left({\displaystyle \frac{A_{impervious}}{A_{buffer}}}\right)\times 100\mathrm{\%}$	Percentage of impervious surfaces within the 300 m buffer zone surrounding the park	High impervious cover exacerbates UHI, but its rapid diurnal cooling may produce apparent cooling in remote sensing [26,37].
External Transport
	Road Density	$\left({\displaystyle \frac{L_{road}}{A_{buffer}}}\right)\left({\displaystyle \frac{\mathrm{k}\mathrm{m}}{\mathrm{k}{\mathrm{m}}^2}}\right)$	Road network density within the 300 m buffer zone surrounding the park	High road density increases anthropogenic heat and UHI intensity, but may improve ventilation [16].

Note: ① DEM, FVC, LAI, impervious surface percentage, and land cover types derived from Landsat 9 Level-2 imagery (30 m); ② road data from OSM; ③ symbols in formulas correspond to variables in the table.

3. Results

3.1. Diurnal Dynamics of Park Cooling Effects

Following a significance test (slope > 0, p < 0.05), 36 of the 52 parks exhibited an observable cooling effect in at least one period, while the remaining 16 parks did not pass the test in any of the three periods (Figure 4). A total of 48 valid observations were obtained across the three periods (21 in the morning, 17 at noon, and 10 at nightfall).

To ensure sample homogeneity for dynamic comparison, the four cooling metrics (PCA, PCE, PCI, and PCG) were quantified for the three periods. Their distributions are shown in Figure 5. The descriptive statistics of key park attributes and surrounding built environment factors are summarized in

Table 4. Descriptive statistics of key park attributes and surrounding built environment factors (n = 52).

Variable	Mean	Median	SD	Min	Max
Park Area (hm2)	16.84	13.88	12.81	1.39	62.91
Green Space Percentage (%)	69.61	74.73	21.78	8.11	98.98
Water Body Percentage (%)	13.83	3.2	21.24	0	69.41
Average Building Height (m)	10.35	9.19	2.18	6.92	18
Impervious Surface Percentage (%)	20.87	18.82	9.02	2.19	39.15

Note: The data in the table are rounded to two decimal places. SD = standard deviation; Min = minimum; Max = maximum.

. From the maximum perspective, PCA remained relatively stable across periods, with a slight contraction at noon, while PCE increased steadily throughout the day, with the most pronounced rise at nightfall and widening inter-sample differences. From the accumulation perspective, PCI and PCG exhibited highly consistent diurnal patterns, both peaking at noon and remaining lower in the morning and at nightfall, showing a typical unimodal diurnal pattern.

3.2. Driving Factor Analysis of Park Cooling Effects

3.2.1. Dual-Perspective Correlation and Independent Validation of Driving Factors

To verify the independent driving effects of each factor, Spearman’s rank correlation was applied to 17 valid midday samples to identify core drivers from two analytical perspectives: the maximum perspective (PCA, PCE) and the accumulation perspective (PCI, PCG) (Figure 6). Partial correlation analysis further validated the independent contributions of key factors.

From the maximum perspective, PCA and PCE exhibit markedly different driving mechanisms. PCA is regulated by both internal and external factors: green space proportion (ρ = 0.50, p < 0.05) acts as a positive driver, effectively expanding cooling extent, while building height (ρ = −0.60, p < 0.05) and water body proportion (ρ = −0.48, p < 0.05) serve as negative inhibitors. Park geometry and vegetation structure parameters exert relatively weak influences. PCE, however, is subject to the dual constraints of park size and the surrounding built environment. A strong negative correlation exists between park area and PCE (ρ = −0.82, p < 0.001), reflecting an inherent trade-off between cooling extent and unit-area efficiency. Building height also exerts a significant negative effect (ρ = −0.68, p < 0.01), acting as the key external constraint on cooling efficiency. Water body proportion shows a negative trend but does not reach statistical significance.

From the accumulation perspective, PCI and PCG are highly consistent (r = 0.93). Building height is the only significant driver, negatively affecting both PCI (ρ = −0.53, p < 0.05) and PCG (ρ = −0.58, p < 0.05). It weakens cooling by blocking cool air dispersion and increasing heat storage, thus representing a key built-environment constraint. Green space proportion and elevation show a positive correlation, which may intensify the thermal gradient through vegetation shading, transpiration, and elevation-related temperature differences, although their independent effects are not significant.

VIF diagnostics (threshold < 5) showed values ranging from 1.02 to 2.87, indicating no severe multicollinearity. Partial correlation analysis further validated the independent effects of each factor. After controlling for park area, the positive effect of green space proportion on PCA (r = 0.502, p < 0.05) and the negative effect of building height on PCE (r = −0.567, p < 0.05) remained significant, both independent of scale and statistically robust. In contrast, the perimeter–PCA correlation weakened substantially after area control, indicating that perimeter lacks independent driving value. After controlling for building height, the negative water–PCA correlation was no longer significant (r = −0.453, p > 0.05), indicating spatial confounding by building height rather than water body thermal properties. Similarly, after controlling for green space, the elevation-PCI correlation was not significant (r = 0.421, p > 0.05), indicating partial mediation by green space.

In summary, park area (median: 13.88 hm²), green space proportion (mean: 69.61%), and building height (mean: 10.35 m) are the core drivers. Water body proportion is a secondary factor, while morphological and topographical factors lack independent driving value and were excluded from the analysis.

3.2.2. Diurnal Variation in Independent Effects of Core Drivers

Based on the midday period, three core drivers were identified: park area, green space proportion, and building height. To further reveal the diurnal dynamics of their independent effects, second-order partial correlation analysis was applied. With the other two factors controlled, the net correlations between each factor and PCA and PCI were calculated separately for the three periods (

Table 5. Partial correlation coefficients between core drivers and cooling indicators across periods.

Driving Factors	Cooling Metrics	Morning (21)	Noon (17)	Nightfall (10)
Area	PCA	0.589 **	0.044	−0.514
	PCI	0.246	−0.004	−0.11
GreenRatio	PCA	−0.03	0.472	0.298
	PCI	−0.261	−0.097	−0.158
BuildingHeight	PCA	−0.490 *	−0.423	0.067
	PCI	0.072	−0.322	0.169

Note: * p < 0.05, ** p < 0.01; the other two core factors were controlled; results for nightfall are for reference only due to small sample size.

). PCE and PCG are derived indicators with driving mechanisms aligned with PCA and PCI, and are therefore not re-analysed separately.

The positive effect of park area on PCA exhibited marked diurnal variations. In the morning, park area and PCA showed a strong positive correlation (r = 0.589, p < 0.01), peaking during this period. At midday, however, the partial correlation dropped sharply to 0.044 and was no longer significant, indicating that the independent effect of park area on PCA virtually disappeared. During nightfall, the results fluctuated considerably, likely due to the limited sample size. The partial correlations between park area and PCI remained low across all three periods and were not significant, indicating that park area has no significant independent explanatory power for PCI.

The positive effect of green space proportion on PCA followed a unimodal diurnal pattern, peaking at midday (r = 0.472) and weakening in the morning and at nightfall. In the morning, the coefficient approached zero (r ≈ 0), indicating no significant independent effect. At nightfall, the effect was intermediate between morning and midday levels. Across all three periods, the correlation between green space proportion and PCI was not significant, and the overall effect was weak.

Building height negatively affected PCA, with effect intensity fluctuating across daytime periods. In the morning, the two variables showed a significant negative correlation (r = −0.490, p < 0.05). At midday, the negative effect persisted but weakened slightly and was not significant. By nightfall, the correlation direction became unstable. Across all three periods, building height showed no significant partial correlation with PCI.

Overall, the independent effects of the three core drivers on PCA showed clear intraday variations, while their effects on PCI were generally weak and not significant across any of the three periods. These trends are broadly consistent with the Spearman correlation results, confirming the stability of the temporal variations in factor effects. Owing to the limited sample size for nightfall, the results lack robustness and should be interpreted only as indicative of trends.

3.2.3. Quantitative Assessment of Combined Driver Contributions Using Time-Specific Regression Models

To further quantify the combined contributions of the three core drivers (park area, green space proportion, and building height) to the park cooling effect, separate multiple linear regressions were run for each period, with PCA and PCI as outcomes and the three core factors as predictors. Standardised beta coefficients were used to compare factor importance. As the nightfall sample size was insufficient for multiple regression, modelling was limited to the morning and midday periods (Figure 7).

For the morning period, the PCA model was significant (adjusted R² = 0.356, p < 0.05), explaining 35.6% of the spatial variation in the morning PCA. Park area was the dominant positive driver (Beta = 0.545, p < 0.01), while building height showed a significant negative effect (Beta = −0.422, p < 0.05), and green space proportion was negligible (Beta = −0.023, p = 0.904). The PCI model was not significant (adjusted R² = −0.036, p = 0.527), with no significant factor effects.

For the midday period, neither the PCA nor the PCI model reached significance (p = 0.112 and p = 0.672, respectively). In the PCA model, green space proportion (Beta = 0.432, p = 0.075) and building height (Beta = −0.377, p = 0.116) approached significance, while park area was near zero (Beta = 0.035, p = 0.878). In the PCI model, all three standardised coefficients were below 0.33 (p > 0.05), indicating weak contributions.

From morning to midday, the beta coefficients diverged: green space proportion increased from −0.023 to 0.432, park area dropped sharply from 0.545 to 0.035, and building height decreased slightly from −0.422 to −0.377. By midday, the absolute beta values for green space proportion and building height exceeded those of park area, signalling a clear temporal shift: park area (median: 13.88 hm²) dominated in the morning, while green space proportion (mean: 69.61%) and surrounding building height (mean: 10.35 m) prevailed at midday.

This ranking was confirmed by the multicollinearity-adjusted regression model and was broadly consistent with the direction and magnitude of the associations identified by Spearman correlation analysis.

It should be acknowledged that the blue-park category contains only six samples and that the nightfall analysis is based on only ten observations. These results should therefore be interpreted as indicative trends rather than definitive typological characterisations.

3.2.4. Driving Effects and Critical Intervals of Core Factors

To test for a critical park area threshold, the valid morning and midday samples were divided into small- and large-scale groups by median park area. Partial correlation coefficients between green space proportion and PCA, as well as between building height and PCA, were then calculated for each group to test for scale-threshold effects (

Table 6. Partial correlation coefficients of core factors with cooling areas for parks of different sizes.

Time Period	Group	Sample Size (n)	Green Ratio (Controlling for Building Height)	Building Height (Controlling for Green Ratio)
Morning	Small parks	11	−0.136	0.513
	Large parks	10	0.06	−0.698 *
Noon	Small parks	9	0.192	0.164
	Large parks	8	0.453	−0.462

Note: * p < 0.05; grouped by median park area: morning ~16.0 hm², noon ~14.2 hm².

).

The results reveal marked scale heterogeneity in core factor effects, with 14–16 hm² identified as a potential transition range in the driving pattern. Inter-group comparisons show that the positive effect of green space proportion on PCA strengthens with park size, with the largest difference at midday, when the effect was stronger in large parks than in small ones. The negative effect of building height on PCA was more stable in large parks and significant during the morning.

In small parks, the direction and significance of core factor effects fluctuated considerably, with no stable independent driving effect observed. Given the limited sample size after grouping, this analysis focused on identifying trend-based characteristics.

It should be noted that the subgroup analysis is based on relatively small sample sizes (morning groups: n = 10–11; noon groups: n = 8–9). The 14–16 hm² range identified here is therefore intended to indicate a trend rather than a precisely estimated threshold. We present this as preliminary evidence that should be confirmed or refined through future studies with expanded park samples.

The above analysis establishes a continuous core-factor framework for the full sample. However, cooling mechanisms may deviate across park types due to differences in thermophysical properties, warranting verification from a typological heterogeneity perspective.

3.3. Classification of Urban Parks Based on Blue–Green Space Composition

3.3.1. Classification Framework and Sample Characteristics of Urban Parks

Based on blue–green composition, the 36 cooling-effective parks were classified into blue, green, and grey types. Blue parks, with a water proportion of 30% or higher, are characterised by lakes or ponds with riparian vegetation and mature trees along shorelines. Green parks are predominantly vegetated, with tree cover as the dominant land cover type and understory vegetation consisting mainly of grass and shrubs; canopy structure ranges from closed forests to open woodlands, with tree maturity varying across parks. Grey parks are dominated by impervious surfaces, including plazas, paved paths, and recreational facilities, with sparse and scattered tree cover. All parks receive routine maintenance, including irrigation during dry periods. The remaining parks were assigned to green or grey categories by median impervious and green space proportions [3].

Table 7. Classification of 36 urban parks in Wuhan by blue–green space composition.

Park Type	Number	Percentage	Park Name
Blue parks	6	16.7%	Lianhua Lake Park, Xingfu Bay Park, Wufeng Gate Wetland Park, Huashan Wetland Park, Baodao Park, Ziyang Park
Green parks	13	36.1%	Wuhan Landscape Science Park, Wuhan Garden Expo Park, Optics Valley Third Road Wetland Park, Xunsi River Flowing Water Park, Xunsi River Sports Park, New District Park, Qingshanji Park, Tuanjie Park, Baiyu Park, Baibuting Garden, South Main Canal Garden, Qiaokou Park, Simeitang Park
Grey parks	17	47.2%	Wuhan’s Xunsi River Scenic Park, Dijiang Park, Guishan Park, Hongshan Park, Chu Wangtai Ruins Park, Guanshan Park, Changqing Park, Houxianghe Park, Yellow Crane Tower Park, Jiefang Park, Jinyintan Park, Lanting Park, Qingshan Park, Shouyi Square, Wangjiadun Park, Xibeihu Green Space, Zhang Zhidong Sports Park, Zhongshan Park

summarises the counts and basic attributes of each type.

3.3.2. Spatial Differences in Cooling Effects Among Park Types

The static differentiation of cooling effects across the three park typologies was examined at midday from both the maximum and accumulation perspectives, revealing the regulatory role of internal landscape composition (Figure 8).

From the maximum perspective, PCA and PCE exhibited inverse differentiation across the three park types. For PCA, blue parks had the highest median (1,015,098 m²) and the lowest dispersion, indicating the most stable spatial extent among water-dominated cool islands. Green parks showed a slightly lower median (761,214 m²) but a broader distribution, while grey parks had the lowest median (287,503 m²) and greatest heterogeneity. In contrast, PCE followed the opposite pattern: green parks achieved the highest median efficiency, reflecting vegetation-driven cooling per unit area. Grey parks had a median PCE comparable to that of blue parks but with more pronounced extreme highs, predominantly associated with small-scale parks, consistent with the scale-efficiency trade-off. Blue parks exhibited the most concentrated PCE distribution with minimal fluctuation.

From the accumulation perspective, PCI and PCG distributions overlapped considerably, with limited differentiation across park types. Median levels were similar across the three types, with sporadic extremes only in grey parks. After outlier removal, the distribution ranges of the three types largely coincided, indicating that the cumulative cooling effect at midday was weakly regulated by landscape type. High-intensity and high-gradient characteristics were primarily driven by a few parks with specific configurations, rather than representing type-typical patterns.

In summary, blue–green space composition exerts a dimension-specific influence: it strongly and divergently shapes spatial extent and unit-area efficiency, while its effects on cumulative intensity and attenuation gradient are relatively indirect.

3.3.3. Diurnal Cooling Patterns Across Park Types

To reveal diurnal variations in cooling patterns across park types, PCA, PCE, PCI, and PCG were range-standardised and plotted as radar charts for each period (Figure 9). The geometric evolution of these charts revealed distinct cooling patterns across the three park types.

Across all three periods, blue parks exhibited a spindle-shaped distribution dominated by PCA and PCG, with only proportional scaling and no axial reorganisation. The pattern expanded in the morning, contracted slightly at midday, and re-expanded by nightfall, demonstrating strong temporal consistency.

Green parks exhibited a marked time-dependent shift in configuration. In the morning, the four axes were relatively balanced. At midday, PCG expanded significantly, forming a cone-shaped pattern dominated by a single axis. By nightfall, the configuration returned to a balanced state with a slightly reduced scale. At midday, radar line convergence was strongest, and the cooling pattern was most clearly directional.

In grey parks, the cooling effect was suppressed at midday but rebounded markedly by nightfall, revealing a distinct diurnal contrast. In the morning, samples exhibited strong heterogeneity with pronounced inter-park variability. At midday, the cooling effect was generally suppressed, and the overall pattern contracted. By nightfall, the PCE axis expanded sharply outward, forming a dual-axis pattern dominated by PCE and PCA.

Overall, the three park types exhibited marked differences in diurnal cooling patterns. Blue parks maintained stable patterns throughout the day, green parks showed pronounced midday characteristics, and grey parks displayed a distinct nightfall reversal. These diurnal variations correspond to the core driving mechanisms identified earlier and provide an empirical basis for subsequent analyses of typological heterogeneity in cooling mechanisms.

4. Discussion

4.1. Diurnal Asynchrony of Cooling Metrics and Its Driving Mechanisms

The results indicate that PCI and PCG peak at midday, while PCA remains relatively stable throughout the day, and PCE increases markedly at nightfall. These three metrics do not evolve synchronously, suggesting that different dimensions of the park cooling effect are governed by distinct mechanisms. This asynchrony can be attributed to the combined effects of diurnal solar radiation variation and surface thermal response.

PCI and PCG are plausibly associated with vegetation transpiration and shading, which peak with solar radiation at midday [38]. Meanwhile, rapid warming of impervious surfaces amplified the thermal contrast between the park and its surroundings, bringing the temperature difference to its diurnal maximum at midday [39]. Using multi-temporal ECOSTRESS data, Wei et al. found that cumulative cooling metrics (PCI and PCG) peaked in the early afternoon in Beijing’s 97 parks [11]; Kong et al. further examined diurnal variations in the cooling effect of blue–green spaces in the same city, finding that vegetation contribution peaked at 32.30% at midday and dropped to 13.86% at nightfall, confirming that vegetation cooling is significantly enhanced during the midday period [37]. In this study, PCI peaked around 10:55, earlier than the early afternoon peak reported for Beijing [11]. Wuhan lies at a lower latitude, with a summer solstice day length about 53 min shorter than Beijing’s and an earlier sunrise [40]. Solar radiation patterns also differ: Beijing shows a single peak, while Wuhan shows a double peak. This earlier solar onset in Wuhan may explain why PCI peaks earlier there than in Beijing [41]. Despite the midday peak in thermal contrast, this does not imply that the same cooling intensity persists throughout the day. Unlike PCI, PCG, and PCE, PCA remained relatively stable, suggesting that spatial and intensity attributes are governed by distinct mechanisms. Cool air dispersion depends primarily on park area and surrounding built-environment configuration, which are static conditions unaffected by diurnal solar radiation [42]. Wei et al. found that PCA fluctuated diurnally far less than PCI [11]; Wang et al. further demonstrated that LST drivers exhibit significant spatio-temporal non-stationarity, shifting systematically with urban development stage and spatial location [43]; Zhang et al. confirmed that urban morphology affects LST with marked seasonal and diurnal variations, showing strong spatiotemporal heterogeneity [44]. Compared with Beijing, PCA in Wuhan exhibited smaller diurnal fluctuations, likely due to higher water coverage. The thermal buffering effect of water bodies, resulting from their high heat capacity, helps maintain stable cooling boundaries throughout the day. Thus, PCI and PCA reflect different mechanisms and are not interchangeable; a single indicator cannot fully capture park cooling performance.

4.2. Scale Thresholds and the Shift from Area Dominance to Quality Constraints in Park Cooling

The above group analysis reveals that when parks are divided by median area (approximately 14–16 hm²), the partial correlation between green space proportion and PCA is weaker or less stable in small parks but significantly stronger in large parks, while the negative effect of building height is more stable in large parks. This indicates that core driver effects exhibit significant scale heterogeneity, marking a potential transition range in the driving pattern. The formation of this transition range can be understood as a dynamic equilibrium between park cooling capacity, horizontal cool-air dispersion, and external thermal disturbance.

Small parks have limited cooling capacity; cool air is dissipated by thermal radiation from the surrounding high-density built environment before horizontal diffusion can fully take effect [45]. The independent contribution of internal landscape composition to cooling is difficult to realise in small parks. Consequently, the driving effects of qualitative factors, such as green space proportion, are not significant. Only when park area increases and cooling capacity accumulates to a level sufficient to offset external thermal disturbances do differences in internal blue–green space configuration begin to emerge. Parks with a high green space proportion can generate a more stable cool air flux, while the blocking effect of surrounding building height on cool air dispersion is simultaneously amplified [33].

The 14–16 hm² range marks the shift from area-dominance to quality-constraint control, with the marginal cooling benefit of park area expansion exhibiting diminishing returns [7], consistent with the threshold efficiency concept [15].

The transition range is not a universal constant across cities, but is jointly modulated by background climate and local blue–green space composition. In a cross-climate comparison, Wei et al. identified a threshold of approximately 24 hm² in Beijing [11], considerably higher than the 14–16 hm² observed in Wuhan in the present study. A comparative analysis by Tian et al. across four climatic regions in China, covering 108 parks, further revealed that the optimal cooling areas for Shenyang, Zhengzhou, Wuhan, and Nanning were 7.27 hm², 8.08 hm², 9.69 hm², and 44.42 hm², respectively [46]. Wuhan’s value ranks in the lower-middle range in cross-city comparisons, partly attributable to the high background humidity of its subtropical monsoon climate, which sustains vegetation transpiration and evaporation, reducing energy loss during cool-air dispersion [36]. This is also linked to Wuhan’s high water body coverage. The 9.69 hm² reported by Tian et al. [46] denotes the size that maximises cooling efficiency per unit area, whereas the 14–16 hm² identified in this study marks the transition in dominant driving factors. Although defined differently, both estimates point to the same conclusion: the critical scale for Wuhan is substantially smaller than that for Beijing. This discrepancy cannot be explained by background climate alone; differences in water body coverage may also be a significant contributing factor.

Cao et al. found that the cool island intensity of Wuhan parks remained largely unchanged when water body proportion varied between 30% and 60%, but increased significantly with water proportion once this range was exceeded [3]. This U-shaped response suggests that low water proportions limit cold storage, while excessive water raises humidity and suppresses heat exchange, reducing efficiency [47]. In contrast, Du et al. found that in Shanghai, larger water bodies were associated with a more stable cooling effect [48]. This discrepancy may arise from the generally low water body proportion in the Shanghai sample, which had not reached the efficiency-inhibiting threshold. In Wuhan, most urban parks have water body proportions concentrated between 30% and 60%, a range where water bodies exert a stable cooling contribution and their thermal buffering effect is fully realised. This allows Wuhan’s parks to sustain a stable low-temperature core even at relatively small sizes, explaining why their critical scale is lower than that of predominantly green parks in northern cities such as Beijing [3,36]. The 30% water body threshold and the 14–16 hm² area threshold pertain to distinct analytical dimensions: the former captures the relationship between water proportion and cooling intensity, whereas the latter identifies the area at which the dominant driving factor shifts.

In summary, the cross-city variation in critical intervals cannot be attributed solely to climatic zone background. The fact that differences persist even among cities sharing the same humid subtropical climate suggests that local blue–green space composition plays a moderating role in determining the actual position of these thresholds. Moreover, the threshold efficiency concept proposed by previous studies provides a theoretical framework for understanding these scale effects [7]. However, this concept alone cannot explain why the driving mechanisms differ on either side of the critical interval. The partial correlation group analysis conducted in this study further reveals that this interval marks a transition in the driving regime, shifting from total scale dominance to allocation quality and external constraints, rather than simply the point at which cooling gains begin to diminish.

4.3. Cooling Patterns Shaped by Surface Thermophysical Properties Across Park Typologies

The three park types are classified by internal blue–green space composition, corresponding to dominant surface types of water bodies, vegetation, and impervious surfaces. With their differing thermal capacity and thermal inertia, these surface types exhibit systematic diurnal variations in cooling patterns. Such differences in response due to dominant surface type are precisely why a full-sample analysis would dilute the effects of certain factors [49]. Previous studies have also shown that internal landscape features outweigh external factors in driving park cooling performance [50]. However, this conclusion is based on the average effect across the full sample and does not account for structural differences in blue–green space composition across park types. Whether the decoupling of spatial and intensity attributes, and the scale thresholds identified above, hold across different park types is closely tied to their internal blue–green composition.

The stable cooling characteristics of blue parks are likely attributable to the thermal buffering effect of water bodies, whose high heat capacity buffers temperature fluctuations. As a result, water bodies are less affected by diurnal solar radiation than vegetation or impervious surfaces, enabling a stable cool core throughout the day. A cross-climatic comparison by Geng et al. further indicates that background climate exerts regionally variable modulation on the dominant drivers of park cooling, with the release of latent heat from evaporation being more sustained in humid low-latitude regions [36]. The water-dominated cooling effect not only extends over a larger spatial extent but also exhibits the smoothest diurnal fluctuations. The highest median PCA and lowest dispersion are consistent with the thermal buffering mechanism of water bodies. Similarly, Du et al. found that parks with larger water bodies in Shanghai exhibited a more stable cooling effect, a pattern consistent with the findings in Wuhan [48].

The midday peak in green parks corresponds to the unimodal diurnal rhythm of vegetation physiological activity [51,52]. Vegetation cooling relies on shading and transpiration, the effectiveness of which rises and falls with solar radiation, peaking at midday and diminishing in the morning and at nightfall. This midday peak, however, is contingent on moisture availability to sustain transpiration. In Mediterranean cities such as Valencia, limited summer precipitation and irrigation reduce green space cooling efficiency by noon due to soil moisture depletion and diminished transpirational cooling [53,54,55]. Under such water-limited conditions, canopy shading alone is less effective. In contrast, Wuhan’s humid subtropical climate and routine park irrigation maintain vegetation activity throughout summer, allowing both shading and transpiration to peak at midday and producing the observed unimodal PCI and PCG peaks. This contrast suggests that the midday cooling peak of green parks is not a universal phenomenon, but a context-dependent outcome shaped by local climate and management practices. Consequently, their diurnal fluctuations are greater than those of blue parks. Zhou et al., in a cross-climatic study of 276 urban parks across China, found that internal landscape factors such as vegetation cover are key drivers of park cooling efficiency [50]. Shi et al. found that canopy shade is strongest at midday and weakens substantially in the morning and at nightfall [45,56]. Green parks exhibit the highest median PCE, reflecting their peak shading and transpiration at midday. However, the spatial extent of vegetation-driven cooling depends heavily on vegetation structure and distribution, resulting in greater PCA variability than in blue parks. Similarly, Wei et al. found that parks with high vegetation cover in Beijing exhibited the greatest cooling intensity at midday [11], confirming the cross-city consistency of the midday peak in green parks.

The evening reversal in grey parks is plausibly associated with the low thermal inertia of impervious surfaces. Owing to this low thermal inertia, impervious surfaces warm rapidly under intense midday radiation, suppressing the cooling effect. At nightfall, once radiation subsides, they cool quickly, while surrounding high-thermal-capacity buildings may lag in heat release. This could widen the apparent thermal contrast between the park and its surroundings. However, it should be acknowledged that LST alone cannot fully distinguish the contributions of thermal inertia, building shading, heat storage release, and ecological cooling processes. Other factors—such as shadows cast by surrounding buildings during low-sun periods—may also contribute to the observed LST reductions in grey parks, particularly in dense urban contexts [12,19]. Therefore, while the low-thermal-inertia explanation is consistent with our observations, it should be interpreted as one plausible mechanism among several. Moreover, this apparent phenomenon is predominantly concentrated in grey parks. Pan et al. found that grey parks exhibit high cooling efficiency in the morning and at nightfall, but that this efficiency virtually disappears at midday [12]. Sun et al. found that building shadows contribute significantly to cooling in the morning and at nightfall, but this effect weakens sharply at midday [19]. Kim et al. also found that artificial and hybrid shade structures can contribute to local cooling, particularly during low-sun periods [55].Grey parks had the lowest median PCA and the greatest heterogeneity. Extreme high values were predominantly found in small-scale parks surrounded by dense development, where cooling was an apparent effect of shade at dawn and dusk rather than a genuine scale or ecological response. This aligns with the 14–16 hm² transition range identified earlier: small grey parks below this threshold are more prone to apparent evening cooling, an effect amplified by Wuhan’s high-density built environment.

There were no significant differences among the three park types in PCI and PCG at midday, consistent with the mechanism analysis above. Cumulative metrics capture the total park-surrounding temperature difference; at midday, radiative forcing dominates, amplifying it to its daily maximum across all types and masking inter-type differences.

The heterogeneity of these three driving pathways explains the reduced explanatory power of the full-sample linear model [15,50]. When modelled together, water cooling, vegetation evapotranspiration, and impervious thermal response offset one another, weakening the linear relationship. The temporal sensitivity differences among the three park types carry clear planning implications: blue parks maintain stable all-day cooling and serve as reliable all-weather sources; green parks peak at midday and are well-suited to mitigating extreme afternoon heat; however, the high evening PCE of grey parks is an apparent phenomenon and should not inform planning decisions. The asynchronous diurnal evolution of the cooling effect, the critical size range, and the threefold pattern differentiation all originate from fundamental differences in thermal capacity and response rates among water bodies, vegetation, and impervious surfaces. Cooling mechanisms vary substantially across landscape types and resist generalisation. A single full-sample model cannot simultaneously capture the independent contributions of all three mechanisms; this is precisely the motivation for the typological decomposition in this study.

Given the limited number of blue parks (n = 6) and nightfall observations (n = 10), these typological comparisons require confirmation through larger-scale studies.

4.4. Limitations and Future Research Directions

Although this study has established a multi-temporal framework for assessing the park cooling effect and revealed its diurnal dynamics and driving mechanisms in central Wuhan, several limitations remain due to data sources, sample size, and study scale. Further work will help deepen the understanding of park cooling dynamics.

Temporal resolution and data consistency remain inherent constraints. The LST data come from two sensors—Landsat 9 at midday and ECOSTRESS in the morning and at nightfall—and from different acquisition dates. Validation studies indicate that Landsat 9 LST has a bias of approximately 0.24 K and an RMSE of 3.42 K relative to in situ measurements, with good agreement against Landsat 7 and 8 [23]. To ensure comparability, we applied strict co-registration, phenological filtering, and spline interpolation to resample ECOSTRESS to 30 m [27,28]. Nevertheless, these sensor differences may affect cross-period comparisons of absolute cooling values. We therefore focus on relative diurnal patterns and rank-order changes rather than absolute LST differences. Given that the reported bias is substantially smaller than the observed diurnal variations in cooling metrics [23], our main conclusions are unlikely to be overturned. The nightfall results, however, are based on only 10 observations and should be treated as indicative. Future work could integrate hourly geostationary LST data or apply temperature cycle modelling to merged ECOSTRESS-Landsat datasets [54,55,56] for more temporally consistent daytime cooling analysis. This strategy has been increasingly adopted in urban climate studies, as demonstrated by recent applications integrating both sensors for diurnal thermal analysis across multiple cities and heatwave periods [12,54,56], supporting the methodological approach of this study.

Sub-sample robustness requires improvement. Only six blue parks were included, due to the exclusion of waterfront parks adjacent to the main channels of the Yangtze and Han rivers and large lakes. This limits the representativeness of blue park cooling patterns. The nightfall sample is also limited to only 10 cases, yet the all-day stability of blue parks and the apparent evening reversal in grey parks are among the core findings of this study; the small sample size may affect their robustness. Future research could broaden coverage to include more types and sizes, or enhance subgroup analysis power through cross-city data integration. The sample size also varies across periods, with 21 valid parks in the morning, 17 at noon, and 10 at nightfall. Only two parks met the criterion for a common-subset sensitivity analysis across all three periods, precluding meaningful statistical inference. This limitation should be considered when interpreting the diurnal comparisons.

The evaluation framework remains one-dimensional. All cooling indicators are derived from LST, which captures only surface radiant heat and differs from near-surface air temperature and actual human thermal perception. Thus, these indicators cannot fully represent park contributions to thermal comfort [57]. In the future, thermal comfort indicators such as the Universal Thermal Climate Index (UTCI) and Physiological Equivalent Temperature (PET) could be incorporated, supplemented by on-site microclimate observations, to achieve a multidimensional upgrade from physical cooling to human-centred thermal perception assessment.

Beyond individual park design, our findings inform broader spatial planning. The 14–16 hm² transition range and the divergent cooling patterns across park typologies suggest that cooling performance depends not only on park size but also on landscape composition and surrounding urban morphology. Strategic allocation of park types—rather than uniform area expansion—may therefore enhance heat mitigation at the district scale. Blue parks provide stable all-day cooling and are suited to high-density residential or commercial areas; green parks, with their midday peak, benefit zones with high daytime pedestrian activity; grey parks, however, exhibit elevated evening PCE as an apparent thermal response rather than ecological cooling, and should not be prioritised solely on that basis. Recent advances in fine-grained urban simulation offer potential for integrating these typology-specific cooling performances into planning frameworks to guide the spatial prioritisation of blue–green spaces across urban functional zones [58]. Future planning frameworks should also integrate human-centric thermal comfort indicators to complement LST-based physical cooling assessments [59].

The multi-temporal framework can serve as a methodological reference for other cities facing similar heat mitigation challenges. Although specific cooling magnitudes and threshold values may vary with local climate, vegetation, and urban morphology, the methodological protocol—including the integration of thermal data from multiple sources, the quantification of cooling metrics across multiple dimensions, and the classification based on park typologies—can be adapted to different urban contexts with minimal modification. Future work could further simplify data requirements by using more accessible satellite products or by developing analytical toolkits that are easy to use, thereby facilitating broader application of the proposed framework beyond individual case studies.

Overall, the multi-temporal framework developed in this study can serve as a methodological reference for similar research. Further extensions will refine the theoretical understanding of park cooling dynamics and provide a more robust scientific basis for urban climate adaptation planning.

5. Conclusions

Based on multi-source remote sensing data from Landsat 9 and ECOSTRESS, this study constructed a novel three-period analytical framework (morning, noon, and nightfall) to assess the cooling effect and driving mechanisms of 52 urban parks in central Wuhan. The main findings are as follows:

(1)

Multidimensional indicators of the park cooling effect exhibit asynchronous diurnal variations. Both PCI and PCG follow a unimodal distribution, peaking at midday. The marked increase in PCE at nightfall reflects the low-thermal-inertia response of impervious surfaces rather than enhanced ecological cooling, while PCA remains relatively stable throughout the day. These findings suggest that single-period assessments cannot fully capture park cooling performance.

(2)

Park area, green space proportion, and building height are the three key driving factors, with their contributions shifting systematically over the course of the day. In the morning, park area dominates; by noon, its role diminishes, while green space proportion and building height become the primary regulators. These three factors consistently influence PCA but have only weak explanatory power for PCI across all periods. The contribution pattern exhibits a potential transition within the approximate range of 14–16 hm², though this should be interpreted cautiously given the limited subgroup sample sizes. Below this range, factor effects appear unstable, while above it, patterns become clearer and more predictable.

(3)

The three park types (blue, green, and grey) exhibit fundamental thermophysical differences in cooling mechanisms, and this typological heterogeneity is the primary reason for the weak explanatory power of the full-sample linear model. A water body proportion threshold of approximately 30% is also identified for cooling contribution.

These findings confirm that single-time-phase or full-sample averaging cannot fully capture park cooling dynamics. The multi-temporal framework and typological analytical approach developed in this study can serve as a reference for park area management and typology-based design in Wuhan and similar cities. Future research could incorporate multi-seasonal data and thermal comfort indicators to further validate the generalisability of these conclusions.