Cooling Effects of Urban Park Green Spaces in Downtown Qingdao

Abstract

Global climate warming and rapid urbanization have intensified the urban heat island effect (UHI). Previous studies indicated that urban parks could effectively mitigate the UHI and improve urban thermal environments. This study aimed to quantify the cooling effect of 64 urban park green spaces in downtown Qingdao. Park cooling intensity (PCI), park cooling gradient (PCG), park cooling area (PCA), and park cooling efficiency (PCE) were selected as indicators to quantify the cooling effect of park green space. These four indicators comprehensively assessed park cooling effects in terms of the maximum value and cumulative value, respectively. Key factors influencing cooling factors and their relative importance were analyzed. The results showed that the mean PCI, PCG, PCA, and PCE were, respectively, 0.02 °C, 0.71, 63.72 ha, and 10.71 °C for the 64 park green spaces. The average temperature reduction and cooling distance were, respectively, 3.35 °C and 211.53 m. Correlation analysis revealed that park area, park perimeter, and NDVI (Normalized Difference Vegetation Index) were significantly positively correlated with PCA, PCI, and PCG. Conversely, these factors presented a significant negative correlation with PCE. Additionally, water body ratio and green space ratio were positively correlated with PCA, while green space ratio was also positively correlated with PCI. The threshold value of efficiency (TVoE), which was calculated based on PCA, was 30.24 ha. TVoE represented the minimum area of urban park green space required to maximize cooling benefits. By means of Ward’s hierarchical clustering method, the 64 park green spaces were classified into four cooling clusters, which were dominated by differentiated cooling metrics and characterized by distinct internal landscapes and surrounding environments. Cluster 1 accounted for 43.75% of the 64 park green spaces, and it was dominated by PCI and PCG. These findings would provide crucial insights for optimizing urban thermal environments, enhancing livability, and promoting sustainable urban planning.

1. Introduction

China’s level of urbanization has increased rapidly over the past 40 years. According to the Chinese National Bureau of Statistics, the urbanization rate has increased from 17.92% in 1978 to 66.16% by the end of 2023. The continuous influx of population into cities and the rapid expansion of urban scale have brought about a series of complex ecological and environmental challenges. In the process of rapid urbanization, the urban landscape has undergone fundamental changes, characterized by the swift spread of impervious surfaces. High-density, multi-layered building complexes and transportation networks, along with the large-scale conversion of natural and semi-natural land-use types into impervious surfaces, have created complicated built environments, which differ from the natural ecosystem. These factors have led to significantly higher land surface temperatures (LSTs) in urban areas in contrast with the surrounding rural regions, and this phenomenon is called the urban heat island effect (UHI) [1]. The high-temperature urban environment impacts human comfort, increases the risk of cardiovascular and respiratory diseases in residential areas, particularly affecting vulnerable groups such as young children, the elderly, and low-income populations. In addition, it also enhances energy and water consumption, aggravates air pollution, and hinders urban sustainable development [2,3]. Consequently, effectively mitigating the UHI and regulating the urban thermal environment has become a critical issue in urban development and planning. Due to the ability of expansive natural woodlands and wetlands in suburban regions to mitigate increasing temperatures, extensive studies have suggested that the urban heat island (UHI) phenomenon can be tackled through the implementation of green and blue infrastructure (GBI), considering the current urban land-use patterns [4,5].

GBI is a biological habitat network system that integrates natural, semi-natural, and artificial green vegetation and water bodies [6]. As a typical element of urban GBI, urban park green space not only provides recreational spaces and critical ecological functions but also delivers the cooling effect [7], carbon sequestration function [8], and biodiversity conservation [9]. The cooling mechanism of urban parks operates through vegetation shading, evapotranspiration, and enhanced albedo to mitigate local thermal environments [10]. Previous studies on the cooling effect of urban park green space primarily focused on three aspects: evaluation of cooling efficiency [8], mechanisms influencing cooling effects [11], and park planning and design based on cooling performance [12]. In the past two decades, methodologies for studying park cooling effects have evolved from field measurements to remote sensing combined with geographic information systems (GIS) and microclimate models. Early empirical studies began in the 1990s, investigating temperature differences between Chapultepec Park in Mexico City and its surrounding built-up areas to explore park cooling effects [13]. In the early 21st century, microclimate models, e.g., TAS and Envi-met, were employed to analyze energy consumption patterns of typical commercial buildings near Bukit Batok Nature Park and compare the thermal conditions with and without Clementi Woods Park. These studies revealed that the cooling effects of urban green spaces are not only significant within vegetated areas but also extend to the surrounding built environment [14]. Currently, retrieving LST from satellite data has become a research hotspot, offering highly efficient and in-depth evaluation results. For example, LST derived from Landsat 8 OLI/TIRS remote sensing data was employed as a proxy for air temperature, as well as estimated LST in Fuzhou City through the radiative transfer equation (RTE) algorithm [15]. Another study employed the Split-Window algorithm to quantify the cooling island intensity of urban parks across different spatial gradients in Beijing [16].

In general, existing research has primarily focused on using inversion results to measure cooling effect indicators. To be precise, earlier studies quantified cooling effects through maximum-value perspectives. For instance, Feyisa et al. defined cooling intensity as the temperature difference between a park and its surrounding environment [17]. Some scholars have defined the cooling intensity as the difference between the land surface temperature (LST) within a 500 m buffer zone outside the park and the average LST inside the park [18]. A study exploring the cooling potential of urban parks in Belgium was focused on two key indicators: cooling intensity (maximum temperature reduction) and cooling range (maximum horizontal distance with a minimum cooling of 0.1 °C). And, the results showed that the maximum cooling intensity during the day was 3.4 °C, with a maximum cooling range of 498 m [19]. Although these calculations are relatively simple, there is some limitation in accuracy because the cooling effect of parks is spatially nonlinear and continuous. Relying solely on maximum cooling indices cannot fully reflect the spatial variation in surface temperature around urban parks [15]. Consequently, recent studies have proposed the park cooling gradient (PCG) and park cooling intensity indicators to enhance the dynamic understanding of the spatial continuity of the cooling process in urban parks [15,20,21].

With regard to urban sustainable development, it is crucial to guide urban planning by analyzing factors influencing the cooling effect [22]. Previous studies have assessed the impact of park area, perimeter [23], Normalized Difference Vegetation Index (NDVI) [24], proportion of green space and water bodies within parks [25], and landscape pattern indices on park cooling effects [26]. Most current research focuses on the internal structures of parks and their impact on thermal environments. However, this study also considers external environmental factors. Previous studies extensively employed statistical models (e.g., correlation analysis and linear regression) to explain the relationship between park cooling effects and influencing factors, and identified the positive or negative effects of potential factors on the cooling effect. Alternatively, urban parks are categorized into typical clusters based on combinations of cooling indices to compare differences in park characteristics. Due to uncertainties in dominant factors and efficiency thresholds (TVoE) of urban park cooling effects under different local climates, there are distinct regional variations in cooling performance.

Qingdao, a major coastal economic and population center in western China, has adopted an active population aggregation strategy, aiming for a permanent population of 11 million by 2025. Industrial and urban expansion has intensified pressure on urban green space ecosystems. To enhance urban ecosystem quality, the government proposed building a “park city” and enacted the Three-Year Action Plan for Park City Construction in Qingdao (2022–2024).

Prior to the present study, the authors acknowledge the limitations in evaluation of cooling effects of urban green spaces, i.e., (1) a predominant focus on single-dimensional indicators, lacking a comprehensive, multi-metric evaluation system; (2) insufficient understanding of the synergetic mechanisms among park morphology, vegetation structure, and surrounding environments; and (3) a limited investigation of cooling thresholds. Hence, the aim of this study is (1) to comprehensively quantify urban park cooling performance from both the perspective of maximum cooling intensity and the cumulative cooling effect; (2) to identify and rank the relative importance of internal and external factors influencing park cooling performance; (3) to determine the threshold of efficiency (TVoE) to reveal the optimal park scale for maximizing synergistic cooling benefits; (4) to classify urban parks based on cooling characteristics through Ward’s hierarchical clustering analysis to inform functional park typologies and planning strategies (Figure 1).

2. Materials and Methods

2.1. Study Area

Qingdao City is situated on the southern coast of the Shandong Peninsula (35°35′–37°09′ N, 119°30′–121°00′ E), adjacent to the Yellow Sea. As a key economic center and port city in eastern China, Qingdao has an urban permanent population of 8.121 million, with an urbanization rate of 77.17%. Qingdao City covers an area of 11,282 km², and it consists of seven urban districts and three suburban county-level cities. In recent years, the urban heat island effect (UHI) has intensified with the rapid urban sprawl.

This study focused on the downtown Qingdao, encompassing the Shinan District, Shibei District, Licang District, and built-up area of Laoshan District. The total study area covers approximately 251.92 km², characterized by dense urban infrastructure, mixed land-use, and varying green space distribution. This region illustrates the relationship between rapid urbanization and thermal environmental challenges, making it a notable case for analyzing park cooling effects and their implications for sustainable planning.

2.2. Data Acquisition

At regional and global scales, Landsat 8 satellite remote sensing equipped with the operational land imager sensor (OLI) and the thermal infrared sensor (TIRS) is the optimal method for retrieving land surface temperature (LST) with high accuracy [27]. The data were georeferenced in the WGS84 coordinate system, at a spatial resolution of 30 m and a temporal resolution of 16 days. This study employed the Level-2 product dataset of Landsat 8. The dataset has undergone radiometric calibration and atmospheric correction and includes the thermal infrared band (ST_B10). Consequently, preprocessing is conducted to derive the land surface temperature data.

Landsat 8 OLI/TIRS images with zero cloud cover from June, July, and August, fully encompassing all urban parks in the study area, were selected from the USGS official website to provide representative summer LST values. As shown in the table, the selected image for this study was acquired on 23 August, with Path: 120 and Row: 035. After obtaining the specified image via Google Earth Engine, the red and near-infrared reflectance bands were used to calculate the Normalized Difference Vegetation Index (NDVI) for the study area. Additionally, Sentinel-2 Level-2A time series data spanning summer 2023 (1 June to 31 August 2023) were employed, with original band resolutions of 10 m and 20 m. The resolution was resampled to 10 m to align with the primary bands (B2, B3, B4, B8), enabling the calculation of five remote sensing indices:

(1)

NDVI (Normalized Difference Vegetation Index): Measures vegetation vitality [28].

(2)

NDWI (Normalized Difference Water Index): Identifies water bodies [29].

(3)

EVI (Enhanced Vegetation Index): Improves recognition of high-density vegetation [30].

(4)

BSI (Bare Soil Index): Distinguishes bare soil from other features [31].

(5)

IBI (Index-Based Built-up Index): Highlights built-up areas [32].

NASA DEM elevation data (NASA/NASADEM_HGT/001) were integrated for analysis. A random forest method was used to classify land cover into four categories (impervious surfaces, urban green spaces, water bodies, and others). Approximately 100 random points per land-use type were selected for accuracy assessment. The overall classification accuracy reached 91%, and the Kappa coefficient was above 0.85. To avoid confusion between building/mountain shadows and water bodies, partial classifications were manually adjusted through visual interpretation. The road network and building footprint data were obtained from OpenStreetMap in December 2023.

2.3. Park Boundary Extraction

As shown in Figure 2, the study selected parks located in downtown Qingdao with minimal or no adjacency to large contiguous green spaces (e.g., Laoshan Mountain) and water bodies (e.g., the Yellow Sea) within a 300 m buffer zone. Given the 30 m resolution of Landsat 8 C2L2 data, urban parks larger than 900 m² were included. In this study, a total of 64 urban parks were incorporated. The largest among them is Fushan Forest Park (581.85 hectares), while the smallest is Changchun Road Pocket Park (0.22 hectares), with an average park size of 26.94 hectares. The urban park boundaries were manually interpreted and extracted using Tianditu maps and ArcGIS (version 10.6, Esri Inc., Redlands, CA, USA) software. Meanwhile, we referred to the Qingdao Park City Construction Plan (2021–2035). Based on the social service function attributes defined in the Urban Green Space Classification Standard (CJJ/T85-2017) [33], the 64 urban parks in downtown Qingdao were categorized into five types: comprehensive parks, specialized parks, community parks, pocket parks, and scenic recreational green spaces outside built-up areas.

2.4. LST Retrieval and Cold/Hot Spot Analysis

Traditional studies rely on ENVI (version 5.6, 64-bit; Harris Geospatial Solutions, Boulder, CO, USA) for image processing, which involves complex and tedious procedures. In contrast, this study utilized Google Earth Engine (GEE) to process image data. This approach not only significantly reduces image processing time and improves efficiency, but also effectively addresses issues such as data gaps, cloud cover, color discrepancies, and temporal inconsistencies in remote sensing data, thereby accurately reflecting the surface thermal environment.

The workflow begins with cloud removal and band correction. The QA band (QA_PIXEL) and saturated pixel flags (QA_RADSAT) are applied to eliminate clouds and invalid data. Next, the thermal infrared band is processed. Since the raw values of Landsat data are quantized and calibrated digital numbers (DN), they require scaling and offset adjustments to convert them into physical units, as shown in the following formula [34]:

LST = DN × 0.00341802 + 149 − 273.15

(1)

where LST is the land surface temperature, DN is calibrated digital numbers, 0.00341802 is scaling factor, and 149.0 is offset value. In order to convert the radiance value to Kelvin temperature (K), the radiance value was then converted to Celsius temperature (°C) by subtracting 273.15.

2.5. Definition and Quantification of Urban Park Cooling Effects

Given that the LST derived from Landsat 8 imagery has a spatial resolution of 30 m, ArcGIS10.6 software was used to construct 20 concentric multi-level buffers extending outward from park boundaries, and each of them spaced 30 m. Then, the zonal statistics tool in the ArcGIS10.6 software was applied to calculate the average land surface temperature (LST) within each buffer zone. Thus, the park cooling effects were quantified [35].

The distance from the park boundary to each buffer was defined as the independent variable l, and the average LST of each buffer was set as the dependent variable T. Then, an LST–distance relationship was established. Curve-fitting analysis revealed that a cubic polynomial equation was the optimal model to describe this relationship. The cubic polynomial function T(l) are as follows [15]:

$$T(l)={al}^3+{bl}^2+cl+d$$

(2)

where l is the independent variable representing the distance from the buffer zone to the park boundary, T is the dependent variable reflecting the average LST within each buffer ring, and the parameters of a, b, c, and d are coefficients obtained through the curve-fitting process via SPSS software (Version 25.0, IBM Corp., Armonk, NY, USA).

As illustrated in Figure 3, the distance between the buffer zone and the park boundary is defined as the independent variable l, and the average LST of each buffer zone is set as the dependent variable T(l). The x-axis represents the distance from the park boundary, and the y-axis represents the corresponding LST. As the distance from the park boundary increases, the land surface temperature (LST) rises continuously. However, when the distance reaches a certain point, the LST stabilizes until the slope of the curve approaches zero [36]. This inflection point is termed the urban park cooling inflection point, and the distance from the park boundary to this inflection point is defined as the maximum cooling distance (L_max). The corresponding LST at this point represents the maximum cooling temperature difference (ΔLST) [22,37]. Beyond this distance, the cooling effect of the park diminishes.

If the cubic polynomial does not exhibit a point where the derivative equals zero, the inflection point is determined by the minimum value of the first derivative of the T(l) function [21,38]. The first inflection point of T(l), representing the maximum cooling distance, can be calculated as

$$T\left(l\right)={\displaystyle \frac{-2\mathrm{b}-\sqrt{∆}}{6\mathrm{a}}}$$

(3)

where a and b are coefficients to be determined.

The cooling effect of urban parks is evaluated using PCA, PCE, PCI, and PCG as evaluation indicators as follows:

(1)

Park Cooling Area (PCA): The buffer area corresponding to the inflection point represents the maximum cooling influence range, i.e., the buffer area from the park boundary to the maximum cooling distance [39].

(2)

Park Cooling Efficiency (PCE): The ratio of the park cooling area to the park area, characterizing the cooling area per unit area of the urban park, reflects the economic efficiency of the park in generating cooling effects [39].

$$PCE=PCA/Spark$$

(4)

where Spark is the area of the park.

(3)

Previous studies have simplistically defined the quantification of park cooling intensity (PCI) as the difference between the temperature at the first inflection point and the average park temperature [1]. However, as shown in the graphs, the cooling effect process of parks is spatially nonlinear. Even if the “LST-Distance” curves differ, the value of L (e.g., maximum cooling distance) might remain the same. Therefore, evaluating PCI solely from the perspective of spatial maximum values is insufficient to fully reveal the distinct characteristics present in different parks, and spatial cumulative metrics must be incorporated [1]. Following the methodology of citation [21], this study defines PCI as the ratio of the land surface temperature (LST) reduction caused by park construction to the original LST within a specified maximum cooling range. PCI is used to gauge residents’ perception of the cooling effect: a higher PCI value indicates a more pronounced perception of cooling by residents. Specifically, the intensity is defined as the difference between the cumulative temperature without the park and the actual cumulative temperature with the park in place.

$$PCI={\displaystyle \frac{L_{max}\times T_D-{\int }_0^{L_{max}}T\left(l\right)dl}{L_{max\times T_D}}}$$

(5)

where T_D is the LST value of the first turning point, and $L_{max}$ × T_D reflects the accumulated temperature at each continuous point without urban parks. ${\int }_0^{L_{max}}T\left(l\right)dl$ reflects the actual accumulated temperature from the surrounding area of the park to various continuous points at a certain distance, and $L_{max}\times T_D-{\int }_0^{L_{max}}T\left(l\right)dl$ reflects the accumulated temperature difference caused by the park.

(4)

The park cooling gradient (PCG) is defined as the ratio of the cumulative cooling amount to the park cooling distance. A higher PCG value indicates stronger heat absorption capacity during the cooling process. PCG essentially represents the cooling magnitude of the park, reflecting the pattern of its cooling process. A large PCG corresponds to an intensive high-heat-absorption process during cooling, while a low PCG involves an extensive low-heat-absorption process [21].

$$PCG={\displaystyle \frac{L_{max}\times T_D-{\int }_0^{L_{max}}T\left(l\right)dl}{L_{max}}}$$

(6)

2.6. Identification of Influencing Factors

The potential factors influencing the cooling effect of parks can be primarily categorized into internal factors and external factors [1,40]. Nine indicators were selected based on widely applied influencing factors identified in previous studies (

Table 1. Factors influencing the cooling effect of parks.

Category	Influencing Factors	Definition
Urban Park Landscape Composition Characteristics	Park shape area (PA)	Park area (ha)
	Park shape length (PP)	Park perimeter (m)
	Park LSI (LSI)	Park Shape Index (dimensionless)
	Park water rate (PW)	Proportion of water bodies in urban parks (%)
	Park green rate (PG)	Proportion of urban parks and green spaces (%)
	Park Mean NDVI (NDVI)	The average value of normalized vegetation index (dimensionless)
External Environmental Factors of Urban Parks	Buffer Road density (BR)	Road network density within a 300 m buffer zone of urban parks (km/km2)
	Buffer Building density (BB)	Building density within the 300 m buffer zone of the urban park (%)
	Buffer green rate (BG)	Green space ratio within a 300 m buffer zone of urban parks (%)

).

The internal landscape characteristics include: (1) park area; (2) park perimeter; (3) Landscape Shape Index (LSI) of the park (Equation (7)); (4) Normalized Difference Vegetation Index (NDVI), indicating the vegetation growth status within the park; (5) proportion of water body area within the specific park.

The surrounding environmental characteristics include: (1) road network density within the park buffer zone; (2) building density (aoi) within the park buffer zone; (3) proportion of green space in the buffer zone. The LSI is calculated as follows:

$$LSI={\displaystyle \frac{P}{2\sqrt{\mathsf{\pi }\times {\mathrm{S}}_{\mathrm{p}\mathrm{a}\mathrm{r}\mathrm{k}}}}}$$

(7)

where P represents the perimeter of the park, and Spark represents the size of the park. A higher LSI value indicates a more complex landscape shape [22,37].

To further investigate the relationship between park cooling effect indices and potential factors, Pearson correlation coefficient analysis was employed to examine variable relationships. Subsequently, independent variables were assessed using the covariance diagnostic function in SPSS25.0 to calculate Variance Inflation Factors (VIFs). In the output results, the VIF value for “Park shape length” was found to exceed 5. This case indicates multicollinearity issues, leading to the exclusion of this variable from both the model and feature significance analysis. Within SPSS25.0 software, each cooling indicator was visualized, and outliers were manually identified and excluded based on interquartile range (IQR). Then, a random forest model was utilized to evaluate and quantify the relative importance of normalized potential driving factors within the model [22].

2.7. Quantification of Efficiency Threshold (TVoE)

The efficiency threshold (TVoE) is originally an economic concept that reflects how to achieve maximum benefits through optimized resource allocation under limited resources, representing the most cost-effective choice [41]. Regression analysis between park area, perimeter, and PCA (park cooling ability) indicators within the study area was conducted using SPSS25.0 software. The cooling intensity of parks initially showed a rapid increase with the growth of park area. However, once a certain threshold was exceeded, the rate of cooling intensity growth slowed down and stabilized. The TVoE value corresponds to the park area at which the tangent slope of the logarithmic fitting curve between park area and cooling intensity equals 1. This TVoE-associated park area represents the optimal design point for urban parks. Notably, the threshold may vary depending on the geographic location of the study area and local climatic conditions.

2.8. Typical Urban Park Clusters Classified Based on Park Cooling Indices

A series of ecosystem services repeated in space and time is generally referred to as ecosystem service bundles [42]. This study drew lessons from this method to explore the cooling benefits associated with different types of urban parks. Hierarchical clustering (Ward’s method) was utilized to classify urban parks in downtown Qingdao. Prior to conducting cluster analysis, the four park cooling indices were normalized to a range of 0–1. Levene’s test for homogeneity of variance was applied to compare variances across different clustering schemes, aiding in the selection of the most appropriate number of clusters. Based on previous research, the number of clusters typically ranges from 2 to 10 [21].

3. Results

3.1. Cooling Effect of Urban Parks

As shown in Figure 4, the average land surface temperature in the study area was 37.72 °C in summer, with a range of 27.23–52.26 °C. The spatial pattern of land surface temperature revealed distinguished urban heat island effects in downtown Qingdao. Several distinct heat accumulation zones were observed in densely populated commercial districts, industrial zones, residential areas, and large public facilities. Low-temperature areas were predominantly distributed across large-scale and dispersed urban parks. A comparison of the internal land surface temperatures of 64 urban parks with their surrounding areas (first inflection point surface temperatures) showed that the average temperature within parks was significantly lower than that in adjacent regions, with a decrease of 2.74 °C. Hotspot and coldspot analyses were employed to examine spatial clustering characteristics of high and low surface temperatures. Hotspots or coldspots represented areas where surface temperatures were significantly higher or lower than surrounding regions [43]. Results indicated that 81% of park areas overlap with coldspots (confidence level ≥ 90%), demonstrating the notable cooling effects of urban parks. A multiple regression equation was fitted to quantify the cooling performance of urban parks in the study area through four cooling indicators, as illustrated in Figure 5. The indicators have the following properties:

(1)

PCA (park cooling area): It ranged from 5.31 to 303.68 ha (mean: 63.71 ha). Twenty parks exhibited PCA values above the mean, notably the Shimei’an–Laohushan Park Cluster and Fushan Forest Park, which recorded the highest values.

(2)

PCE (park cooling efficiency): It varied from 0.51 ha to 36.50 ha (mean: 18.50). Twenty-three urban parks exhibited above-average PCE values, including Zhanqiao Square and Qingxiaoyuan Pocket Park, both of which demonstrated relatively high performance.

(3)

PCI (park cooling intensity): It varied from 0.002 to 0.05 (mean: 0.026). Thirty parks recorded PCI values above the mean, with Loushan Park and Licun Park exhibiting the highest intensities.

(4)

PCG (park cooling gradient): It ranged from 0.09 to 1.99 ha (mean: 1.04 ha). Thirty parks exhibited PCG values above the mean, with Licun Park and Fushan Forest Park showing particularly strong gradients.

3.2. Analysis of Potential Drivers of Park Cooling Effects

3.2.1. Correlation Between Cooling Indicators and Potential Influencing Factors

Through Pearson correlation analysis, the impacts of internal and external factors on the four cooling effect indicators were explored (Figure 6).

The park area and perimeter demonstrated a strong positive correlation with PCI increase, consistent with findings from most previous studies [44]. PCI showed a significant positive correlation with the Normalized Difference Vegetation Index (NDVI) within parks, suggesting that greater vegetation coverage is associated with enhanced cooling intensity. In addition, PCG demonstrated a highly significant positive correlation with park area or perimeter, as well as a significant positive correlation with NDVI. The correlation patterns between PCG and park perimeter, area, and NDVI were congruent with those observed for PCI.

With regard to the maximum value, PCA showed a strong positive correlation with the most influencing factors. PCA exhibited an extremely significant positive correlation with park area, perimeter, and LSI, suggesting that larger parks generally possess more vegetation and water bodies, thereby bringing about greater cooling coverage. Parks with longer perimeters and more complex shapes, implying that there were broader interfaces with their surroundings, resulted in larger cooling areas. PCA also showed significant positive correlations with NDVI and the proportion of water area within parks, indicating that blue-green spaces in parks could effectively mitigate the urban heat island effect. In contrast, PCE displayed significant negative correlations with most influencing factors. Notably, PCE was negatively correlated with park area and perimeter, implying that larger parks do not necessarily exhibit higher cooling efficiency, thus highlighting the reference value of the threshold derived from PCA fitting. PCE also exhibited a negative correlation with NDVI.

Furthermore, relationships between factors across different buffer zones and cooling effects were examined for the 64 urban parks. When treating buffer zones as environmental features, building density exhibited a significant positive correlation with cooling efficiency. Road network density exhibited a significant negative correlation with the cooling area. The proportion of green space in the buffer zone demonstrated a significant negative correlation with cooling efficiency but a significant positive correlation with cooling area.

This suggests that park location selection also plays a role in determining cooling performance. Urban parks, situated within dense clusters of buildings, may not provide extensive additional cooling areas but could deliver more cost-effective cooling per unit area. Such spatial distribution characteristics enabled compact urban parks to achieve higher economic efficiency in thermal environment regulation. A higher proportion of green space in the buffer zone implied greater surrounding vegetation coverage, which synergized with the park to enhance cooling effects and expand the area where temperature reductions occur.

3.2.2. Relative Contribution of Influencing Factors

The random forest method was employed to determine the relative importance of influencing factors on park cooling indicators. The dominant factors varied across different cooling indices, but all of them were intrinsic landscape characteristics of the parks themselves rather than environmental elements (Figure 7). The random forest model demonstrated excellent fitting performance for all four cooling effect indicators, with coefficient of determination (R²) values exceeding 0.85. Hence, it indicated strong robustness of the model to explain variations in cooling indices.

For all four cooling indicators, the most influential factors were park area (PA) and NDVI. To be precise, the indicators presented the following features:

(1)

PCI: PA and NDVI accounted for 52.65% and 21.63% of the total relative importance, respectively. The most significant environmental factor was building density (BB), contributing 6.15%.

(2)

PCG: The most important internal landscape factor still accounted for 33.21% of the area. However, the dominant environmental factor was BG, accounting for 4.83%. The distribution of influencing factors for PCI closely resembles that of PCG.

(3)

PCA: PA overwhelmingly dominated, accounting for 34.05% of relative importance. NDVI and water area proportion (PW) followed, with contribution rates of 16.49% and 16.21%, respectively. BG was the most influential environmental factor for PCA, contributing 5.86%.

(4)

PCE: PA and NDVI held a substantial relative importance at 52.65% and 21.63%, respectively. Contributions from PW, park perimeter (PG), LSI, BB, BG, and buffer road density (BR) were all below 10%.

These results underscored intrinsic characteristics of urban parks, particularly area and vegetation coverage (NDVI), which were the primary drivers of cooling performance, while environmental factors are of secondary importance.

3.3. TVoE of Urban Parks

TVoE (threshold of efficiency) refers to the specific park area at which further increases in size yield negligible improvements in the cooling effect (PCA, park cooling area). To determine the TVoE for urban parks in Qingdao, logarithmic curve fitting was applied to reveal the relationship between park area and perimeter (Figure 8). When the slope of the fitted curve equaled 1, the corresponding park area was 30.24 ha, i.e., TVoE. This suggested that once a park reached 30.24 ha, further area expansion had minimal impact on enhancing PCA.

For the park perimeter, the TVoE was 54.44 km. The trend of the fitted curve between PCA and perimeter suggested that expanding parks beyond this scale did not significantly increase cooling coverage. To be precise, excessively large parks may be less efficient in terms of PCA in contrast with the smaller ones that approached TVoE.

It should be noted that these TVoE results differed from previous studies. For example, the area and perimeter of TVoE were, respectively, 103.61 ha and 206.31 km for five urban parks in temperate monsoon climate zones [41]. However, these values for urban parks were, respectively, 429 ha and 13,283 m in Shenzhen City [36]. These discrepancies highlighted that the applicability of TVoE thresholds varied with climatic and regional conditions [15,36].

3.4. Cluster Analysis of Different Park Types

Based on the analysis results shown in Figure 9, the urban parks in the main study area were classified into four types:

Cooling Bundle 1: They made up 43.75% of the total 64 parks, which consisted primarily of specialized parks (e.g., botanical gardens) and comprehensive parks, with some scenic recreational green spaces, e.g., Beiling Mountain Forest Park, Guanxiang Mountain Park, Niumao Mountain Park, Jinshui River Park, and Licun Riverside Park. With regard to key features, they were dominated by PCI (average 0.02) and PCG (average 0.94). These medium-sized parks exhibited relatively higher water body ratios (average 0.18). More precisely, incorporating water bodies could effectively mitigate the urban heat island effect. The shape was the simplest (lowest LSI: 1.58).

Cooling Bundle 2: They made up 6.25% of the total 64 parks, which consisted primarily of specialized parks and scenic recreational green spaces, including Shimei’an Park, Laohushan–Qinglongshan Park Cluster, Zhongshan Park, Zoo–Botanical Garden Green Cluster, and Fushan Forest Park. With regard to key features, they were dominated by PCI (average 0.03), PCG (average 1.14), and PCA (average 275.43 ha), delivering the largest cooling area (PCA) among all four bundles. These parks were characterized by the largest park area (274.89 ha), NDVI (0.32), park green space ratio (0.87), LSI (2.89), and buffer green space ratio (0.31), but the smallest water area ratio (0.11).

Cooling Bundle 3: They made up 25% of the total 64 parks, and the majority of these parks were pocket parks, such as Songling Road Pocket Park and Dazaoyuan Pocket Park. In terms of key features, all four cooling indices were relatively small, with PCE being dominant. Their irregular boundaries enhanced thermal connectivity with the surrounding environment, enabling rapid localized cooling. However, due to the limited size (the smallest among the four types), their cooling magnitude and gradient were inferior to Bundles 1 and 2.

Cooling Bundle 4: They accounted for 26.56% of the total parks, which consisted primarily of community parks and small recreational parks, such as Yongxing Recreational Park, Nanshan Park, and Xishan Senior-themed Park. In terms of key features, they were characterized by the smallest park area (1.77 ha) and NDVI (0.22), but higher green space ratios. They exhibited the highest buffer road network density (0.47) and building density (0.64) among all types. PCA was less prominent due to its compact size, but it played an important role in cooling intensity (PCI) and gradient (PCG). Their regular shapes resulted in lower cooling efficiency, yet their cumulative cooling effects could significantly mitigate heat islands in highly urbanized areas with limited green space availability, thereby providing greater economic efficiency in such contexts.

This classification elucidated the intricate interrelation among park design, size, and contextual factors in shaping cooling performance, thereby offering practical insights for the optimization of urban green space planning.

4. Discussion

4.1. Cooling Effects of Urban Parks

With respect to the cooling effects of urban parks, some researchers are inclined to conduct field measurements [45], while a majority prefer to utilize remote sensing data [46]. When deriving land surface temperature (LST) from remote sensing data, the selection of buffer zone sizes significantly impacts the accuracy. Due to the resolution limitations of Landsat 8 imagery, previous studies have employed various buffer approaches: 30 m buffers [3], equal-area buffers [47], and equal-radius buffers [39,48]. Twenty buffers at a 30 m scale were adopted for each urban park in this study.

Among the 64 urban parks in downtown Qingdao, both internal and surrounding LSTs were lower than that in other urban areas, indicating that parks generate cooling effect through shade and evapotranspiration, with cold air diffusing, advection, and convection to the adjacent zones, thereby reducing LST and mitigating urban heat islands [49,50]. To investigate the cooling patterns, this study constructed LST–distance curves. The results obtained from 20 buffers at a 30 m scale for each urban park indicated that the LST generally exhibited an initial increase with the distance from the park, followed by stabilization or a decline. This implied a finite spatial extent for the cooling effect of the park, which is in accordance with the previous findings [17,51].

Comparing the internal LST of the 64 parks with their surrounding LSTs (first inflection point), the surrounding areas were significantly at a higher temperature, with a mean increase of 3.35 °C. Additionally, notable variations were observed across the four cooling indices:

(1)

PCI (park cooling intensity): The study area averaged 0.02, and was similar to that in Hangzhou (0.05) [3], Fuzhou (0.0181) [15], Shenzhen (0.029 °C) [21], and Zhengzhou (0.026) [38].

(2)

PCA (park cooling area): It averaged 63.71 ha, and it was higher than that in Fuzhou (44.19 ha) [15] and Zhengzhou (40.09 ha) [38], but lower than that in Shenzhen (81.3 ha) [21] and Hangzhou (26.97 ha) [3].

(3)

PCG (park cooling gradient): The mean value was 0.71, and it was lower than that in Fuzhou (0.8 °C) [15] and Zhengzhou (1.04) [38], but higher than that in Taiyuan (0.23) [52].

(4)

PCE (park cooling efficiency): The mean value was 10.07, and it was lower than that in Zhengzhou (18.50 ha) [38] but higher than that in Fuzhou (2.77 ha) [15].

These disparities primarily stemmed from variations in climatic conditions, urban morphology, and park characteristics. This study, conducted in Qingdao with a temperate monsoon climate, found an average temperature increase of 3.35 °C in areas surrounding urban parks. Compared to Fuzhou city in a subtropical humid climate zone, the lower park cooling intensity (PCI) value (0.0181) of Qingdao reflects the saturation effect of evapotranspiration cooling under high humidity conditions. When contrasted with Shenzhen city in a tropical monsoon climate, Shenzhen’s higher park cooling gradient (PCG) value (0.8 °C) demonstrates how high-temperature and -humidity environments enhance cold air diffusion. In comparison to Zhengzhou city in a temperate continental climate, Qingdao’s relatively lower park cooling efficiency (PCE) value (10.07) highlights the unique advantages of strong radiative cooling in arid regions. These systematic differences reveal the regulatory mechanisms of climatic background on urban park cooling performance: humid zones rely on vegetation–water synergy, arid regions emphasize scale and geometric configuration, while transitional zones need to balance both characteristics.

4.2. Factors Influencing Cooling Effects

This study revealed that the cooling effects of urban parks were primarily influenced by their internal characteristics (e.g., blue-green infrastructure) and the layout of the surrounding urban environment [53,54]. Previous studies argued that the park area and perimeter were key factors influencing the cooling performance [55]. This study not only validated the findings but also quantified the contributions of these factors. Park size and NDVI emerged as the dominant drivers of PCA (park cooling area). Our study on the relative importance of factors generally aligns with previous studies, and the discrepancies in the significance of surrounding environmental factors probably arise from regional climatic variations and differences in analytical methodologies [56].

NDVI showed significant positive correlations with PCI, PCA, and PCG. This consistently indicated that enhancing vegetation diversity and coverage within parks would strengthen their cooling functions and boost cooling efficiency. Water bodies within parks also correlate significantly with PCA, where larger water areas correspond to better cooling performance. However, in contrast to previous research [57], our results showed no significant correlation between PCG/PCI and water bodies. This discrepancy probably stems from the limited influence of water bodies when their proportion is below 10% (only 9 parks in this study exceeded this threshold). Thus, urban park planning should prioritize integrating blue-green infrastructure to alleviate heat island effects.

Park shape complexity, measured by LSI (Landscape Shape Index), exhibited a significant positive correlation with cooling area, which is consistent with previous findings [57,58]. Generally, complex shapes would enhance interactions between parks and their surroundings, to be precise, facilitate heat and energy exchange. However, LSI shows no significant relationship with other cooling indices, and it is likely due to the merging of adjacent green patches during analysis.

Additionally, surrounding environmental features impact cooling effects. Exactly, roads and buildings increase local heat loads, weaken cooling performance, hinder cold air diffusion, and then restrict the spatial reach of park cooling. Under such circumstances, urban parks are smaller, but achieve higher cooling efficiency per unit area. This finding is consistent with the results from temperate monsoon regions [22]. Another study on how dense buildings and busy roads constrain park cooling also supports this finding [18]. These insights underscore the need for context-sensitive park design, balancing internal blue-green infrastructure with adaptive strategies for varying urban morphologies.

4.3. Recommendations for Park Planning and Design

The increase in park area enhances park cooling coverage (PCA), park cooling intensity (PCI), and park cooling gradient (PCG). However, indiscriminate expansion in urban parks is impractical due to economic and land-use constraints. This study proposes the threshold of efficiency (TVoE) to guide effective resource allocation. It contributes to promoting optimal park shape and land-use configuration within limited spaces as long as the cooling effects are met. For example:

(1)

In dense urban areas, prioritizing compact, regularly shaped parks or interconnected green networks (e.g., green corridors, green roofs) would maximize cooling efficiency per unit area [59,60,61].

(2)

When planning internal configurations, keep a balance of surrounding vegetation and building distributions, as these features significantly influence park cooling. Prioritizing factors with higher relative importance (e.g., NDVI, water bodies) is recommended when trade-offs are unavoidable, to ensure maximal cooling benefits for urban thermal environment improvement.

Based on the four cooling bundles identified:

(3)

High-Density Residential Zones: It is recommended to deploy community parks and pocket parks (Bundle 4) to rapidly mitigate localized heat islands.

(4)

Urban Periphery/Suburbs: It is recommended to develop large comprehensive parks and scenic green spaces (Bundles 1 and 2) with high vegetation coverage. These parks act as ecological barriers, enhancing regional microclimates.

(5)

Highly Urbanized Cores: Small, flexible pocket parks (Bundle 3) are proposed for integration due to their demonstrated cost-effectiveness and adaptability in constrained spaces.

In general, urban park planning should align with land-use characteristics, population density, and residents’ cooling demands [62]. This stratified approach ensures targeted cooling solutions while balancing ecological and socioeconomic priorities.

The increase in park area enhances park cooling coverage (PCA), intensity (PCI), and gradient (PCG). However, due to economic and land-use constraints, indiscriminate expansion of urban parks is often impractical. To address this, our study introduces the threshold value of efficiency (TVoE) as a strategic tool to guide optimal resource allocation. This concept supports more efficient design of park shape and land-use configurations, ensuring sufficient cooling performance within limited urban space.

For practical implementation, the following suggestion was given.

(1)

In dense urban areas, compact and regularly shaped parks or interconnected green networks (e.g., green corridors, vertical greening, and green roofs) are recommended to maximize cooling efficiency per unit area [59,60,61].

(2)

When designing internal park configurations, planners should balance surrounding vegetation and building patterns. Priority should be given to high-impact factors (e.g., NDVI, water bodies) when trade-offs are necessary, to enhance overall cooling effectiveness.

Based on the four identified cooling bundles, tailored planning strategies are proposed:

(1)

High-density residential zones: Community parks and pocket parks (Bundle 4) are recommended to quickly mitigate localized heat islands.

(2)

Urban periphery and suburban areas: Large-scale parks with high vegetation coverage (Bundles 1 and 2) are suggested to function as ecological buffers and enhance regional microclimates.

(3)

Highly urbanized cores: Small, adaptable pocket parks (Bundle 3) offer a cost-effective and flexible cooling solution where space is limited.

Overall, urban park planning should be aligned with land-use intensity, population distribution, and residents’ cooling needs [62]. This stratified approach enables climate-responsive urban greening strategies while balancing ecological functions and socioeconomic constraints.

4.4. Limitations and Future Research Directions

This study focused on quantifying the cooling effects of urban parks in downtown Qingdao from cumulative and maximum perspectives, which provides comprehensive insights into park cooling services and their influencing factors to guide urban park planning. However, some limitations remain.

(1)

Temporal Scope: The analysis is restricted to daytime cooling effects during summer, while ignoring annual and diurnal variations. Given seasonal differences in urban cooling and heat island dynamics [63,64], future research should explore park cooling performance across seasons and day–night cycles. For instance, a study in Zhengzhou found that LST in urban parks reduced by 0.65 °C (spring), 1.41 °C (summer), and 2.84 °C (fall) but increased by 1.92 °C in winter [65].

(2)

Sample Diversity: The limited sample size (64 parks) may constrain the generalizability of findings, particularly regarding cooling bundle classifications and TVoE thresholds. Further studies should include more park types and climatic regions.

(3)

Previous studies have highlighted the significant cooling effect of water bodies in urban environments, often exceeding that of vegetation [66]. However, in this study, the water body coverage within parks was generally low (average proportion < 5%), and statistical analyses did not reveal a significant correlation between water body coverage and cooling indicators. Some studies have suggested that the cooling effect becomes statistically significant only when the water body proportion exceeds 10% [15]. It is important to clarify that this finding does not negate the theoretical cooling value of water bodies but rather reflects the statistical outcome within the specific sample scope of this study. Future research is encouraged to conduct more refined observations on water bodies of varying scales, particularly: comparing the cooling effects of large (>10% of park area) versus small water bodies; quantifying the impact of water body shape complexity on microclimates; and evaluating the synergistic cooling effects of water–vegetation configurations.

(4)

Methodological Refinements: Incorporating advanced remote sensing data (e.g., higher-resolution thermal imagery) and field measurements (e.g., microclimate sensors) would validate and complement satellite-derived results.

(5)

Socioeconomic Integration: Future work could consider the cooling efficiency of socioeconomic factors (e.g., park accessibility, visitors) to optimize equity in green space distribution.

(6)

Recent studies have made significant progress in pedestrian thermal comfort modeling. For instance, some research has combined the Universal Thermal Climate Index (UTCI) with computational fluid dynamics (CFD) through the “City Comfort+” model to dynamically simulate UTCI at the pedestrian scale and predict future climate scenarios [67]. Xia et al. (2025) further utilized DepthmapX and XGBoost to identify key design factors, such as visibility percentage (PV) and connectivity (CON), that contribute to campus “cool spots” in Hong Kong [68]. However, these innovative methods rely on high-precision 3D models and dense field data, making them difficult to implement in resource-constrained areas. In contrast, the large-scale urban heat island (UHI) assessment methods employed in this study, such as macro analysis based on meteorological stations and LST, although limited in resolution, offer the advantages of ease of access and broad coverage, making them suitable as preliminary screening tools. By incorporating spatial indicators such as PV and CON into the existing evaluation framework, it becomes possible to preliminarily identify micro-scale hotspot areas that require prioritized intervention. Future research could establish a three-tiered system: (1) macro-scale screening using satellite data to identify high-risk UHI zones; (2) meso-scale validation using UTCI simulations to quantify pedestrian heat stress; (3) micro-scale design integrating machine learning to generate cool spot solutions. This framework would not only reduce the cost of fine-scale comprehensive simulations but also provide decision-makers with full-chain support from strategic planning to specific design. Addressing these limitations will deepen our understanding of urban park cooling mechanisms and support climate-adaptive urban planning.

5. Conclusions

This study utilized Landsat 8 and Sentinel-2 satellite imagery to evaluate four cooling indicators (PCA, PCI, PCG, PCE) for 64 urban parks in downtown Qingdao from both maximum and cumulative spatial perspectives. The results demonstrated that parks significantly reduced surrounding temperatures. The key findings are as follows.

(1)

Urban parks exhibited notable cooling effects, and the mean PCI, PCG, PCA, and PCE were 0.02 °C, 0.71, 63.72 ha, and 10.71 °C, respectively. The average temperature reduction and cooling distance were, respectively, 3.35 °C and 211.53 m. Correlation analysis revealed that park area, park perimeter, and NDVI (Normalized Difference Vegetation Index) were significantly positively correlated with PCA, PCI, and PCG. Conversely, these factors presented a significant negative correlation with PCE. Additionally, water body ratio and green space ratio were positively correlated with PCA.

(2)

The building density, road network density, and green space ratio in the buffer zone also impact cooling.

(3)

The identified TVoE (threshold of efficiency) for urban park area was 30.24 ha in Qingdao. And, it represented the minimum area of urban park green space required to maximize cooling benefits.

(4)

Through Ward’s hierarchical clustering method based on cooling indicators, 64 urban parks were grouped into four bundles with distinct cooling characteristics. To enhance the cooling effects for these bundles, differentiated measures should be taken according to the characteristics of urban parks. Among them, bundle 1 accounted for 43.75% of the 64 park green spaces, and it was dominated by PCI and PCG.

These findings could deepen our understanding of the cooling mechanisms of urban parks and provide practical strategies for alleviating heat islands in the context of climate change and land scarcity.