The Impact of Building and Green Space Combination on Urban Thermal Environment Based on Three-Dimensional Landscape Index

Abstract

Urbanization transforms landscapes from natural ecosystems to configurations of impervious surfaces and green spaces, leading to urban heat island effects that impact health and ecosystem sustainability. This study in Xiamen City, China, categorizes urban areas into functional zones, employs Random Forest and Stepwise Regression models to assess thermal differences, and proposes optimization measures for the building–green space landscape. The optimization involves altering the characterization of the building–green space landscape pattern. Results indicate: (1) due to the spatial heterogeneity of the building–green space landscape pattern in different functional zones, the surface temperature also shows strong spatial heterogeneity in different functional zones; (2) different optimization measures for the building–green space pattern are needed for different functional zones; taking the urban residential zone as an example, the Normalized Difference Vegetation Index (NDVI) in the hot spot area can be adjusted according to the value range of the cold spot area; (3) considering the solar radiation process, Sun View Factor (S_unVF) plays an important role in indicating the change in surface temperature in the commercial service area, and as S_unVF increases, the surface temperature of the functional zone tends to rise. This research offers insights into urban thermal environment improvement and landscape pattern optimization.

1. Introduction

Over the past few decades, urbanization has emerged as a defining characteristic of human societal development [1,2]. The relentless march of urbanization has led to swift transformations in urban morphology and landscapes, exacerbating a range of environmental concerns, including local climate shifts [3,4], air pollution [5,6,7], ecological degradation [8,9], and the urban heat island (UHI) effect [10,11]. The UHI effect is characterized by urban areas being warmer than their surrounding rural regions [12]. In recent years, the relentless expansion and concentration of populations, coupled with intense human activities, have amplified the UHI effect globally. This phenomenon can markedly alter microclimates and intensify global warming [13]. Moreover, the UHI effect escalates energy consumption in buildings [14,15,16] and carbon emissions [17,18], posing potential health hazards to urban dwellers [19]. Consequently, mitigating the urban heat island effect is pivotal for enhancing urban climatic conditions and thermal comfort.

To enhance urban habitability and ensure sustainable development, there is an imperative need to investigate strategies for reducing the urban heat island (UHI) effect. Urban green spaces are pivotal in alleviating the UHI effect. A plethora of research has delved into the correlation between surface temperatures and the spatial arrangement of urban green spaces, yielding compelling evidence for UHI mitigation strategies [20,21]. These green spaces, often encompassing vegetation and other natural elements [22], are recognized for their significant role in reducing the UHI effect through processes, such as evapotranspiration and shading [23,24,25]. The characteristics of park green spaces have been observed to display spatial gradients in relation to UHI effects [26]. Moon et al. demonstrated that greenery situated at the center of an urban complex resulted in the most pronounced reduction in UHI effects among three distinct layouts [27]. Additionally, Liu et al. discovered that the impact of urban green space (UGS) landscape composition and configuration on land surface temperature (LST) (explanation of abbreviations see

Table A1. List of abbreviations.

Abbreviation	Fullname	Unit
LST	land surface temperature	°C
UFZ	Urban Functional Zone	None
RF	Random Forest	None
URZ	Urban Residential Zone	None
UVZ	Urban Village Zone	None
COZ	Commercial Zone	None
MUZ	Municipal Utilities Zone	None
IWZ	Industrial Warehouse Zone	None
NDVI	Normalized Difference Vegetation Index	None
NDBI	Normalized Difference Building Index	None
BCR	Building Coverage Ratio	None
FAR	Floor Area Ratio	None
SkyVF	Sky View Factor	None
SunVF	Sun View Factor	None

) varies with seasons and across urban green space ratio (UGSR) gradients [28]. Based on Morphological Spatial Pattern Analysis (MSPA), the UHI effect showed an inverse relationship with the core areas, perforations, and loops of green spaces, while it exhibited a positive correlation with islets [24].

The urban microclimate is significantly influenced by the three-dimensional structure and morphology of cities [29,30]. Urban buildings and transportation infrastructures can intensify the urban heat island (UHI) effect through vertical expansion, impacting surface energy balance and airflow, thus altering the thermal environment. Building height (BH) and street aspect ratio (SAR) show spatial variations in cooling effects, while the number of buildings (BN) and the standard deviation of building height (BH_SD) minimally affect the warming impact [31]. The proportion of built-up land (PCL), the normalized difference built-up index (NDBI), and building density (BD) have a notably variable influence on land surface temperature (LST) along the urban gradient [32]. Mean building height (MBH) and BD are the main determinants of LST, with 2D and 3D indices impacting LST differently across seasons [33]. Based on MSPA, the UHI was found to have a positive correlation with the proportions of the core, edge, and bridge areas within urban regions, while it was negatively correlated with the proportion of islet areas [34].

Cities are hybrid entities, comprising ‘gray’ infrastructure like residential and industrial buildings, and roads; ‘green’ infrastructure with green spaces; and ‘blue’ infrastructure with water bodies. Dominated by ‘gray’ and ‘green’ infrastructures, this hybrid interacts with local conditions to shape the microclimate and thermal comfort of urban residents, which constitute the urban thermal environment. This environment is primarily influenced by ‘gray’ infrastructure from buildings and ‘green’ from spaces. Solely analyzing green infrastructure may not encapsulate the full mechanisms of urban thermal change; it is essential to consider the combined effect of green spaces and buildings on the local thermal environment. The LST is significantly positively correlated with Impervious Surfaces (IS), and negatively correlated with Green Spaces (GS) and Blue Spaces (BS). The aggregation of IS has a sustained positive effect on LST [35]. The main factors influencing LST are population density, building density, and vegetation cover, listed in descending order of their impact [36].

Urban functional zones serve as critical spatial carriers for a city’s economic and social functions [37]. They are geographical areas designed to concentrate relevant social resources and to effectively perform specific urban functions. Functional zones of the same type share similar land use patterns, physical boundaries, organizational forms, human activities, and environmental characteristics [38]. The delineation of these zones simplifies complex urban and rural landscapes into a manageable number of landscape unit types, enhancing within-class homogeneity and between-class heterogeneity. These zones are a burgeoning classification in urban heat island research, representing regions of heterogeneity with diverse physical attributes and socio-economic activities. Gao et al. have evaluated the risk of high-temperature exposure across various urban functional zones [39]. Liu et al. have shown that the impact of urban form on land surface temperature (LST) is significantly dependent on the season and urban functional zone (UFZ) type [40]. Mo et al. noted a positive correlation between building density and surface temperature within each urban functional zone [41]. Chen et al. (2024) investigated the response of diurnal LSTs in different UFZs to 2D/3D urban morphology and socio-economic variables [42]. Despite this, current research remains sparse on the relationship between architectural spatial form and surface temperature within the context of different urban functional zones. Therefore, adopting urban functional zones as spatial analysis units and focusing on the landscape pattern characteristics of buildings and green spaces can enhance our understanding of the urban heat island effect. This approach can lead to the development of more practical and targeted landscape pattern optimization strategies.

Most current studies tend to consider the roles of green spaces or built environments in isolation. This research breaks through the limitations of traditional analytical methods and discusses the complex layout of urban landscapes, where green spaces and buildings are not just coexisting but are interwoven into a whole that affects the thermal environment. This study adopts a synergistic perspective, recognizing the collective impact of these landscape elements on urban heat patterns. Therefore, this study takes urban functional zones as the basic unit of analysis, focusing primarily on the landscape characteristics of building–green space combinations. It employs stepwise multiple regression and random forest regression to explore the linear and nonlinear relationships between Land Surface Temperature (LST) and various influencing factors, respectively. The aim is to analyze the main climate-sensitive parameters that impact the local thermal environment in different urban functional zones. The primary objectives of this research are to (1) contrast the spatial configurations of surface temperatures in different urban regions; (2) investigate the correlations between diverse landscape indicators and surface temperature within distinct urban functional zones; and (3) assess the significance of various landscape indicators in elucidating the distribution of surface temperatures, thereby identifying the predominant factors affecting each urban functional area.

2. Materials and Methods

2.1. Study Area

Xiamen, situated on the southeast coast of Fujian Province with coordinates ranging from 117°53′ to 118°26′ East longitude and 24°23′ to 24°54′ North latitude, serves as a significant hub in the economic zone of the west coast of the Taiwan Strait and the Maritime Silk Road. The city’s topography is characterized by coastal plains, tablelands, and hills, with a gradient that descends from the northwest mountains to the southeast coast. It falls within the subtropical monsoon climate zone globally, moderated by maritime influences, resulting in mild and humid winters and hot, humid summers. The region experiences an average annual temperature of approximately 21 °C, with the coldest month in January and the warmest in August. The average annual precipitation is about 1200 mm, predominantly from May to August. The total land area spans 1699.39 km², within which 132.5 km² constitutes the land area of Xiamen Island. As the urban heart of Xiamen City, the island boasts an early urbanization start, rapid development, a high urbanization rate, and a dense population (Figure 1). This study aims to devise a regulatory framework for the thermal environment of Xiamen City, informed by landscape pattern indices. Considering the actual research needs and in combination with the DEM (Digital Elevation Model) topographic map, areas that cannot undergo landscape pattern regulation, such as mountains, water bodies, airports, and docks, have been excluded.

2.2. Data Sources

2.2.1. Functional Urban Zoning

The study utilized high-resolution Google Earth imagery captured from 10 March 2018 to 14 November 2018, with a spatial resolution of 0.14 m. Based on the administrative boundaries of Xiamen Island, the imagery was mosaicked and clipped accordingly. Leveraging the ultra-high resolution of the remote sensing imagery, in conjunction with urban planning documents and large-scale base geographic maps, a visual interpretation was conducted to classify the land use and cover of Xiamen Island, resulting in vector maps. Post-classification, 200 random points were selected for accuracy assessment, achieving an overall accuracy (OA) of 96%.

Building upon the concept proposed by Sun et al. (2013) [38], which suggests that urban road networks can impede local thermal dissipation, the identification of Urban Functional Zones (UFZs) has predominantly focused on linear urban elements, including roads and waterways. In light of practical conditions and the study’s objectives, the interpretation results of the remote sensing imagery were partitioned according to socio-economic functions, and adjusted in conjunction with Digital Elevation Model (DEM) topographic maps, excluding mountainous areas and water bodies, as well as airports and docks, which do not contribute to landscape pattern regulation. The land use cover of Xiamen Island was thus categorized into five distinct urban functional zones: Urban Residential Zones (URZ), Urban Village Zones (UVZ), Commercial Zones (COZ), Municipal Utility Zones (MUZ), and Industrial and Warehouse Zones (IWZ).

2.2.2. Selection and Calculation of Landscape Pattern Indices

Based on previous studies, a multitude of indicators have been employed to describe the spatial morphology and distributional properties of urban structures. These indices include Building Coverage Ratio (BCR), Floor Area Ratio (FAR) [43], Sky View Factor (S_kyVF) [44,45], and Sun View Factor (S_unVF) [46]. Open Street Map (OSM; http://www.openstreetmap.org/) is an online collaborative mapping platform that enables users to freely contribute to map creation and editing. The data from OSM have been effectively utilized in urban thermal environment studies. In this research, we collected three-dimensional spatial information of buildings within the study area using OSM. The data underwent geometric correction on the ArcGIS platform, and its accuracy was verified through on-site inspections. These three-dimensional building data facilitated the calculation of three-dimensional landscape indices.

Landsat satellites capture electromagnetic radiation at different wavelengths to obtain surface information, and these band data can be used to calculate indices, such as the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Built-up Index (NDBI) [47]. NDVI is one of the most widely used land use intensity indices for assessing plant density [48]. NDVI primarily reflects the growth condition of vegetation; healthy vegetation tends to strongly absorb blue and red spectra while reflecting green spectra due to its chlorophyll content. NDVI is calculated using the high reflectance of NIR and the high absorption of red spectra, whereas NDBI is used to differentiate urban built-up and non-built-up areas, as buildings and bare lands tend to reflect shortwave infrared radiation (SWIR) more than near-infrared radiation (NIR) [49]. Both indices are calculated by analyzing the differences in reflectance across different bands and have been widely applied to investigate the relationship between urban land use cover and thermal environment [47,50,51].

Drawing from prior research, we selected six landscape pattern indices related to the thermal environment (

Table 1. Landscape indices used to characterize urban building and vegetation patterns.

Type	Name	Description
1	Normalized Difference Vegetation Index (NDVI)	NDVI = ${\displaystyle \frac{\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{d}5-\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{d}4}{\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{d}5+\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{d}4}}$ Band 4 and band 5: red and near infra-red bands of Landsat 8 OLI image, respectively. The Normalized Difference Vegetation Index (NDVI) quantifies vegetation by measuring the difference between near-infrared (strong vegetation reflection) and red light (vegetation absorption).
2	Normalized Difference Building Index (NDBI)	NDVI = ${\displaystyle \frac{\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{d}6-\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{d}5}{\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{d}6+\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{d}5}}$ Band 5 and band 6: near infra-red bands and middle infra-red of Landsat 8 OLI image, respectively. The Normalized Difference Building Index (NDBI) is a remote sensing feature index that describes urbanization intensity information.
3	Building Coverage Ratio (BCR)	BCR = ${\displaystyle \frac{\sum \mathrm{F}}{\mathrm{A}}}$ F: the land area of the building taken; A: the study area. Building coverage ratio (BCR) is used to reflect the ratio of building base area to plot area.
4	Floor Area Ratio (FAR)	FAR = ${\displaystyle \frac{{\sum }_{i-1}^{n}CF}{\mathrm{A}}}$ C: number of floors; F: the land area of the building taken; A: the study area. Floor Area Ratio (FAR) is used to reflect the ratio of the total building area above ground to the land area.
5	Sky View Factor (SkyVF)	SkyVF = $\sum _{i=1}^{n}B·{\displaystyle \frac{1}{\pi }}·\sin {\displaystyle \frac{\pi }{180}}·\sin {\displaystyle \frac{\pi (2{\alpha }_i-1)}{2n}}·{\displaystyle \frac{360}{{\beta }_i}}$ B: the binary image of shadow patterns; n: the total number of shadow images generated; αi: the altitude angle at the ith annulus level; βi: the azimuth angle used at the ith annulus level. Sky View Factor (SkyVF) is commonly used to measure the degree to which radiation transmission at a specific location is blocked. This project uses the UMEP plugin of QGIS Desktop 3.16.11 software to calculate the sky viewing angle.
6	Sun View Factor (SunVF)	S = $\sum _{i=1}^n{I}_0·P^{{\displaystyle \frac{1}{\sin {\alpha }_i}}}·\cos {\beta }_i$ SunVF = ${\displaystyle \frac{S-\mathrm{m}\mathrm{i}\mathrm{n}\left(S\right)}{\max \left(S\right)-\mathrm{m}\mathrm{i}\mathrm{n}\left(S\right)}}$ P: transmittance of atmosphere; n: the total number of time steps; I0: the solar constant with a value of 1367 W/m2; αi: the altitude angle at the ith annulus level; βi: the azimuth angle used at the ith annulus level; S: the total direct solar radiation incident on the target point. Sun View Factor (SunVF) considered that the aforementioned sky view angle did not take into account the orientation of the sun when describing incident solar radiation, and Wu et al. [46] developed a sun view angle.

). The study includes three two-dimensional parameters (Normalized Difference Vegetation Index, NDVI; Normalized Difference Building Index, NDBI and Building Coverage Ratio, BCR) and three three-dimensional measures (Building Area Ratio, FAR; Sky View Factor, S_kyVF and Sun View Factor, S_unVF). Figure 2 displays the spatial distribution of the selected landscape pattern indices for the thermal environment in this study.

2.2.3. Surface Temperature Retrieval

In this study, we selected Landsat 8 remote sensing imagery collected on the morning of 22 September 2019 (Beijing time), with cloud cover below 5% and no cloud obstruction in the study area. An atmospheric correction algorithm was applied for retrieving surface temperatures. The imagery underwent pre-processing that included radiometric and geometric corrections to achieve an accuracy of better than 0.5 pixels. The fundamental concept of the atmospheric correction method involves estimating the atmospheric impact on surface thermal radiation. This estimated atmospheric effect is then subtracted from the total thermal radiation recorded by the satellite sensors to derive the surface thermal radiation intensity, which is subsequently converted to the corresponding surface temperature.

Landsat 8 employs the TIR10 band for surface temperature inversion. The thermal infrared radiation brightness value L_λ captured by the satellite sensor is composed of three components: the upward atmospheric radiation brightness L↑, which is the actual radiation brightness from the ground that reaches the satellite sensor after atmospheric transmission; and the energy reflected by the downward atmospheric radiation after it strikes the ground. The expression for the thermal infrared radiation brightness value L_λ received by the satellite sensor can be formulated using the radiative transfer equation (Equations (1) and (2)):

L_λ = [εB(T_S) + (1 − ε)L↓]τ + L↑

(1)

T_S = K₂/ln(K₁/B(T_S) + 1)

(2)

where B(T_S) is the brightness of blackbody thermal radiation; T_S is the surface temperature; L↑ is the brightness of atmospheric upwelling radiation; L↓ is the brightness of atmospheric downwelling radiation, and K₁ and K₂ are constants. The Landsat 8 thermal infrared band has an upwelling radiance (L↑) of 1.25 W/(m²-sr-μm), and a downwelling radiance (L↓) of 2.37 W/(m²-sr-μm). The constants K₁ and K₂ are 774.89 and 1321.08, respectively.

2.2.4. Identification of Hot/Cold Spot

Spatial autocorrelation analysis evaluates the statistical distribution of spatial data, with results that reflect the data’s spatial arrangement rather than relying solely on traditional econometric statistics. This type of analysis encompasses both global and local autocorrelation [52]. Local spatial autocorrelation, particularly relevant for hotspot analysis, identifies areas where values of a variable are unusually high (hotspots) or low (coldspots).

In spatial statistics, the Gi* coefficient, introduced by Getis and Ord [53], is a widely recognized measure of local spatial autocorrelation. It is calculated based on the full matrix of distances to detect clusters of high or low values in space and to quantify the density of these clusters. The Gi* statistic is accompanied by a Z value, which provides a standardized measure of spatial clustering. A higher Z value signifies a significant hotspot, indicating a concentration of high-value spatial data, while a lower Z value indicates a significant coldspot, reflecting a concentration of low-value spatial data.

Utilizing ArcGIS’s hotspot analysis tool, which specializes in spatial clustering, the study conducted a spatial analysis of high and low surface temperature values on Xiamen Island. The Getis_OrdGi* index was employed to identify hotspots (areas with a concentration of high values) and cold spots (areas with a concentration of low values). The distribution of these cold hotspots indicates the extent and impact of regional contributions to surface temperature. Hotspots, characterized by clustered high values, are regions where the surrounding land surface temperature (LST) is also elevated.

2.3. Statistical Analysis

Stepwise regression analysis is a method for constructing a regression model by sequentially introducing or removing predictor variables based on their significance. The process begins by screening potential predictor variables and selecting those that contribute significantly to the model. Each new variable is subjected to a significance test, and if it passes, it is included in the regression equation. Conversely, if it does not contribute significantly, it is excluded [54]. Additionally, with each iteration, the model re-evaluates the existing variables to ensure that all remain significant, maintaining a parsimonious and relevant set of predictors.

In this study, data from urban functional zones, landscape pattern indices, and the results of the cold and hotspot analysis of surface temperatures were integrated within ArcGIS 10.4.1 software. The software was used to extract the average surface temperature and landscape pattern indices for each identified hot spot and cold spot segment. A stepwise regression model was then applied to explore the correlation between surface temperature and landscape pattern indices. The analysis aimed to identify the primary landscape pattern indices that contribute to the spatial heterogeneity of surface temperatures. The findings from this stepwise regression analysis will be instrumental in informing and guiding future urban landscape planning strategies.

2.4. Random Forest Algorithm

Random Forest (RF) contains multiple decision tree classifiers and its output categories are determined by the plurality of individual tree outputs, and training and test samples are continuously generated by the bootstrap resampling technique. Random Forest improves prediction accuracy and is insensitive to multicollinearity without significantly increasing the amount of computation, and the results are robust to missing data and unbalanced data. Random Forest can predict the role of up to several thousand explanatory variables well and is regarded as one of the best algorithms currently available [55]. The principle of the RF algorithm is to enhance predictive accuracy by reducing variance. Random Forest offers high accuracy in both classification and regression problems by combining the predictions of multiple decision trees, thereby reducing the bias of individual models. Unlike least-squares linear regression, RF can capture nonlinear relationships between the predicted and feature variables with high precision [56]. As a result, the RF algorithm has gained popularity across various research domains as a component of machine learning analysis. In the context of our study, the RF model is leveraged to evaluate the relative influence of various landscape indices on land surface temperature (LST), assigning an importance score to each predictor variable that reflects its contribution to the model’s predictive accuracy [57].

Initially, a Pearson correlation coefficient test was conducted to determine the impact of the feature variables on thermal environment parameters. Variables that demonstrated significant correlation (p-value < 0.05) were selected for further analysis. Following the filtering of variables and data preprocessing, the dataset was randomly partitioned into training (70%) and testing (30%) sets to facilitate model development. The Scikit-Learn library in Python was utilized for RF estimation To optimize the model’s predictive performance, a five-fold cross-validation approach was implemented for hyperparameter optimization prior to model training [58]. We utilize the Python platform to implement the Random Forest model, with parameters set to ‘oob_score = True’ and ‘random_state = 10’. It seeks to optimize the following two hyperparameters: ‘n_estimators’ and ‘max_features’. This method’s strength lies in its ability to address multicollinearity and account for the interactive effects among different variables.

3. Results

3.1. Spatial Pattern of Land Surface Temperature (LST)

Figure 3a illustrates the spatial distribution of LST on Xiamen Island, tailored to the study’s requirements, with temperatures ranging from 28.73 °C to 49.06 °C, reflecting a significant temperature gradient of 20.33 °C. For the global Moran’s I analysis and local hotspot analysis using the Getis-Ord Gi* statistic, the data were vectorized, and a grid system was implemented with 100 m cells. The mean LST within each cell was utilized for both the global Moran’s I analysis and the hotspot analysis. The global Moran’s I index of 0.696 indicates that surface temperatures exhibit positive spatial autocorrelation. The z-score of 107.464 is quite high, and the p-values are very low (less than 0.01), suggesting that there is significant spatial clustering of surface temperatures (statistically significant hot spots or cold spots). Figure 3b presents the hotspot analysis of land surface temperatures, indicating that the cold spot area has a minimum temperature of 28.00 °C, a maximum temperature of 38.04 °C, and an average temperature of 34.51 °C. In contrast, the hot spot area records a minimum temperature of 34.30 °C, a maximum temperature of 48.14 °C, and an average temperature of 38.57 °C, which is 4.06 °C higher than the average temperature of the cold spot area.

The hotspot analysis results for surface temperature were overlaid with urban functional zone statistics (Figure 4), revealing distinct thermal characteristics within these zones. Notably, the area of hot spots within the UVZ, COZ, and IWZ is substantially larger than that of the cold spots, with respective areas of 33.91 km² versus 11.39 km², 23.53 km² versus 19.80 km², and 68.18 km² versus 3.47 km². The hot spots in the IWZ constitute a significant proportion, representing 68.18 percent of the total zone area. Conversely, the cold spots within the URZ cover a considerable area as well, making up 34.02 percent of the total zone area. For the MUZ, the areas of cold and hot spots are nearly identical.

3.2. Relationship Between LST and Landscape Pattern Indices

From the results of the Pearson correlation analysis in Figure 5, it can be observed that the cold and hot spot areas within the same urban functional zone share significant landscape indices at the 0.01 level (bilateral). These common significant landscape indices are then employed to analyze the impact of landscape patterns on the spatial heterogeneity of the urban thermal environment.

Figure 6 delineates the fluctuations in the aforementioned common indices across various urban functional zones. The urban residential, urban village residential, industrial and mining, and public facility land uses exhibit 1, 4, 2, and 2 shared landscape pattern indices, respectively. A prominent commonality is the NDVI, which significantly correlates with the urban thermal environment, with the UVZ showing a higher NDVI value compared to other zones. In the cold spot regions, the average NDVI and S_unVF values surpass those of the hot spot areas. Conversely, the FAR and BCR in the cold spots are marginally lower than in the hot spots. These disparities underscore the distinct landscape patterns associated with cold and hot spots. The characteristics of the cold spot areas can be leveraged as a benchmark for modifying the landscape patterns in hot spot areas. Such adjustments aim to ameliorate the local thermal environment by introducing features that foster colder conditions.

3.3. Spatial Interaction Between LST and Landscape Pattern Indices by RF Algorithm

The Random Forest (RF) model was utilized to further explore the combined impact of six urban landscape indicators on surface temperature variations across different UFZs. In the cold spot areas, UVZ exhibited better predictability, accounting for 74.64% of the variance in surface temperature. In the hot spot areas, IWZ was more effectively predicted, explaining 65.45% of the surface temperature. There are notable differences in the dominant factors influencing surface temperature among various UFZ types.

In the cold spot zones, for URZ, the NDVI has a more pronounced effect on surface temperature, followed by the FAR, BCR, and S_unVF. In COZ, NDBI and BCR are more significantly associated with temperature impacts. S_unVF has the highest importance score in IWZ, constituting about 37.93% of the total. In UVZ and MUZ, the impact of each landscape indicator does not show a significant difference.

In hot spot areas, for URZ, the importance of NDBI, S_unVF, and FAR on surface temperature gradually decreases. In the UVZ, S_unVF has a particularly significant impact on surface temperature, accounting for approximately 24.97% of the total influence. In COZ, NDVI and S_unVF show a more pronounced importance on temperature. In IWZ and MUZ, BCR is the primary indicator affecting surface temperature.

The RF model outcomes do not entirely align with the Pearson correlation analysis results, suggesting that assessing the influence of urban spatial patterns on surface temperature should not be based solely on bivariate relationships. Figure 7 illustrates the comparison of importance scores for various variables within the RF model. When examining the impact of UGS spatial patterns on surface temperature in each urban functional area, S_unVF and NDBI are more significant in hot/cold spots.

3.4. Relationship Between LST and Parameters at the Scale of Temperature Classes

To pinpoint the critical parameters influencing temperature variations within urban functional areas, the temperature data for each zone were stratified into four quartile-based ranges using the 10th, 30th, 50th, 70th, and 90th percentiles. These categories are designated as class 1 (10th-30th percentile), class 2 (30th–50th percentile), class 3 (50th–70th percentile), and class 4 (70th–90th percentile). Employing a stepwise regression model, we examined the spatial correlations among surface temperatures, varying temperature tiers, and a spectrum of landscape parameters across different urban functional zones (as detailed in

Table 2. Results of a stepwise regression model for the temperature classes in UFZs.

LST	Regression Model	R2	SE
CLASS 1	TURZ = −5.353NDVI + 35.535	0.190	0.893
CLASS 2	TURZ = 4.400NDVI + 8.599NDBI + 36.591	0.126	0.696
CLASS 3	TURZ = −1.370SkyVF + 8.262NDBI + 37.493	0.322	0.704
CLASS 4	TURZ = −0.053FAR + 8.670NDBI + 37.943	0.395	0.789
CLASS 1	TUVZ = −3.642NDVI + 1.732FAR − 14.212SunVF + 48.871	0.791	0.912
CLASS 2	TUVZ = −3.432SkyVF + 9.079NDBI + 37.797	0.718	0.75
CLASS 3	TUVZ = 6.711NDBI + 37.812	0.393	0.519
CLASS 4	TUVZ = −3.903NDVI + 37.479	0.349	0.592
CLASS 1	TCOZ = 3.722NDBI + 0.842BCR − 1.766NDVI − 0.049FAR + 35.431	0.383	0.793
CLASS 2	TCOZ = 6.065NDBI − 0.024FAR + 36.685	0.201	0.835
CLASS 3	TCOZ = 9.603NDBI + 7.168SunVF + 31.556	0.347	0.878
CLASS 4	TCOZ = 4.540NDBI − 2.077SunVF + 38.253	0.229	1.012
CLASS 1	TMUZ = −1.192NDVI − 0.650SkyVF + 35.546	0.132	0.331
CLASS 2	TMUZ = −0.556BCR + 36.257	0.072	0.256
CLASS 3	TMUZ = −12.346SunVF + 48.543	0.153	1.187
CLASS 4	TMUZ = −1.278NDVI + 2.508NDBI + 0.613BCR + 38.297	0.211	0.284
CLASS 1	TIWZ = −1.546NDVI + 36.924	0.093	0.429
CLASS 2	TIWZ = −1.126NDVI + 0.031FAR + 37.383	0.22	0.265
CLASS 3	TIWZ = 3.101SkyVF + 37.683	0.212	0.91
CLASS 4	TIWZ = 2.362NDBI + 39.244	0.114	1.28

). This analysis is instrumental in discerning the urban landscape attributes that distinguish low-temperature areas from high-temperature zones. The landscape configurations characteristic of cooler regions can be harnessed to inform and direct the thermal management strategies for warmer areas.

The NDBI primarily affects the surface temperature in Urban Residential Zones (URZ), with the impact of the NDVI being more pronounced in cooler areas. In warmer areas, NDBI, S_kyVF, and FAR have a combined influence. For Urban Village Zones (UVZ), NDVI, FAR, S_unVF, S_kyVF, and NDBI significantly influence surface temperature, with the respective regression equations having R-squared (R²) values of 0.791 and 0.718. In Commercial Zones (COZ), NDBI and FAR are the main parameters affecting the cooler zones, while NDBI and S_unVF are key for the warmer zones. At the lowest temperature tier, the class 1 level, the combined effects of NDBI, BCR, NDVI, and FAR are observed.

In the cooler zones of Municipal Utilities Zones (MUZ), NDVI and BCR impact both low and high temperature areas, whereas S_kyVF and S_unVF negatively affect surface temperature, and NDBI positively affects it. In the Industrial Warehouse Zones (IWZ), the low-temperature areas show a significant negative effect of NDVI on surface temperature and a positive effect of FAR. In the high-temperature areas, the positive effects of S_kyVF and NDBI are predominant.

The NDBI and the NDVI influence the surface temperature in each functional area. The S_unVF primarily impacts the surface temperature in UVZ, COZ, and MUZ, whereas the S_kyVF affects the URZ, UVZ, MUZ, and IWZ. In URZ, the optimal range for NDVI does not necessarily indicate better conditions, and it may be beneficial to also consider reducing NDBI values. In UVZ, the NDVI value is significantly higher in the low-temperature areas compared to high-temperature areas, suggesting a need to increase S_unVF values concurrently. In COZ, the low-temperature zones as a reference, NDBI values should be lowered, and NDVI should be raised in the high-temperature areas. In MUZ, NDVI is observed to decrease as temperatures rise, while NDBI shows an opposite trend, increasing with temperature. In IWZ, the low-temperature areas primarily depend on an increase in NDVI, whereas adjustments in the high-temperature areas should be made by taking into account the lower NDBI values found in the cooler zones (Figure 8). For UFZ with higher temperatures, such as COZ and IWZ, NDBI and SunVF are the primary landscape parameters affecting the high-temperature areas in COZ. In contrast, NDBI and FAR are the main landscape parameters influencing the low-temperature areas in COZ. Within COZ, referring to the low-temperature areas, NDBI values should be reduced in high-temperature areas, NDVI should be increased in these zones. In IWZ, the low-temperature areas are primarily influenced by the negative effect of NDVI on land surface temperature and the positive effect of FAR; in the high-temperature areas, the effects are mainly manifested by the positive influences of SkyVF and NDBI. In IWZ, the low-temperature areas largely depend on the increase in NDVI, while the high-temperature areas should be adjusted based on the reduction in NDBI values observed in the low-temperature areas. NDVI, NDBI, and FAR are landscape parameters that commonly affect both functional zones, which means that in functional zones with higher average temperatures, we should focus on NDVI, NDBI, and FAR to achieve the goal of cooling down the area.

4. Discussion

4.1. Spatial Pattern of Surface Temperature

Surface temperature serves as a key variable for characterizing the urban heat island effect. On Xiamen Island, hot spots are predominantly concentrated in the industrial, mining, and storage areas in the northwest, while cold spots are mainly found in residential areas of town buildings and public facilities. The extensive impermeable surfaces in these areas reduce subsurface heat capacity and surface runoff infiltration. Our findings align with previous studies that have reported regional differences in factors influencing the urban heat island effect [2]. Urban development is strongly and positively spatially correlated with the regional thermal environment [59]. Along the urban–rural gradient approach, land surface temperature (LST) decreases with increasing distance from the urban core. For every one-kilometer increase from the urban core to the periphery, LST drops by approximately 0.45 °C [35]. This study confirms that different urban functional zones exhibit distinct spatial characteristics of the thermal environment [38,60].

4.2. The Impact of Landscape Pattern Indices on Surface Temperature

Both two-dimensional and three-dimensional landscape pattern indices were utilized to delineate the spatial heterogeneity of surface temperature. Six climate-sensitive parameters were selected to gauge changes in surface temperature. Each urban functional zone (UFZ) can be identified as a hotspot or coldspot based on Xiamen’s land use planning, which segments the city into various UFZs. The landscape traits of coldspots—encompassing the composition, configuration, and structure of urban landscapes—offer a definitive benchmark for ameliorative thermal environment strategies in China’s hotspots. A comparative analysis of the shared parameters between hotspots and coldspots within the same UFZ regression model indicates that the thermal environment can be mitigated by adjusting key parameters. Surface temperature is influenced by numerous factors associated with 2D/3D landscape metrics [61]. The impact of 2D/3D urban form indicators on the urban heat island index significantly varies across different urban locales [29]. Both two-dimensional and three-dimensional building attributes are equally influential in driving changes in surface temperature, with two-dimensional vegetation features outstripping three-dimensional counterparts [30]. The effects of three-dimensional building forms on surface temperature also differ across seasons and spatial scales [62]. During the urban development process, different functional zones are formed according to planning, and within the same type of functional zones, similar architectural designs and green space layouts are created. This leads to a tendency for similar surface temperatures in the same type of functional zones, which is precisely why we first use functional zones to distinguish urban landscapes. Meng et al. explore the relationship between land surface temperature (LST) in the center of Beijing and various environmental factors, particularly considering the impact of the surrounding environment and urban vertical expansion. The seasonal effects of vegetation index (NDVI), water body index (MNDWI), building index (NDBI), building density (BD), and building height (BH) on the correlation with LST [63]. Lin et al., based on the MSPA method, analyzed the relationship between the morphological characteristics of urban green space and built-up area and the Surface Urban Heat Island (SUHI). They found that the core areas of green spaces, perforations, and loops were negatively correlated with the intensity of the UHI [24], while the proportions of the core, edge, and bridge parts of the built-up areas were positively correlated [34]. Pearson correlation is only suitable for linear analysis, but the random forest algorithm can successfully calculate the nonlinear importance of various influencing factors on the SUHI effect. Another important feature of the random forest algorithm is that it is not severely affected by issues of multicollinearity or overfitting [34].

UFZs, which are regions designated for similar economic and social activities, serve as fundamental units in urban planning practices. To address the elevated thermal conditions within a specific UFZ, it is logical to adopt the cooler temperature stratum of that zone as a benchmark for comparison. For instance, low-temperature zones characteristic of COZ could be employed as a reference temperature goal for indicators like NDBI in COZ experiencing elevated LST. Essentially, urban planners can utilize this cooler range as a planning objective to alleviate the urban heat island impact.

We employed six landscape indices to characterize the structure of urban green spaces and building landscapes. Among these, the S_kyVF measures the openness or enclosure of a location relative to its surroundings, commonly used in outdoor thermal environment research to assess thermal comfort and correlated with near-surface air temperature. Research by Yang et al. [64] has shown S_kyVF to have linear or nonlinear relationships with near-surface air temperature within building clusters across seasons. Additionally, a study by Scarano & Mancini [44] has revealed a significant linear correlation between S_kyVF and surface temperature. In the majority of studies, S_kyVF is positively correlated with near-surface air temperature or surface temperature. However, some studies, such as that by Chen et al. [65], have shown a negative correlation between the two. The inconsistencies observed in the aforementioned studies may arise from the fact that the calculation of S_kyVF does not account for direct solar radiation, leading to two potential scenarios: First, two locations at different latitudes with identical surrounding building structures would have the same S_kyVF. Second, at the same location, different times of the day could have the same S_kyVF. In both cases, the same S_kyVF is used to represent two distinct thermal environments, which may result in the aforementioned contradictions or inconsistencies. In light of this, our study introduces the S_unVF, an index that incorporates the solar elevation angle and the duration of solar radiation reaching the surface, potentially providing a more accurate indication of the thermal environment changes at specific urban locations. As demonstrated in our study, S_unVF shows a higher contribution to the variation in surface temperature compared to S_kyVF.

While some of the indices utilized in this study are established in the extant literature, such as the NDVI and NDBI, we have also introduced the S_unVF, a less conventional index that is critical for capturing the dynamics of solar radiation reception at specific urban locales. The S_unVF takes into account the solar elevation and azimuth angles, which are pivotal for understanding urban surface temperature escalation. The Random Forest model’s outcomes highlight the significant contribution of S_unVF, with an importance score exceeding 0.15 across all urban functional zones, a distinction not shared by other landscape indices.

4.3. Limitations

This study still has some limitations that require further exploration and resolution. Firstly, urban functional zoning is influenced by various factors, including natural geography, socio-economic conditions, and human activities. Therefore, to achieve more accurate classification results, future research should incorporate additional open datasets that can represent high-resolution remote sensing images of urban space, as well as statistical data describing population mobility. Secondly, among the six landscape indices we selected, only one index, the Normalized Difference Vegetation Index (NDVI), is used to characterize the distribution of green spaces. In previous studies, at the local scale, the canopy structure of individual plants or plant communities could be measured using instruments such as handheld LiDAR. However, there is currently a lack of detailed data on green spaces at the urban scale, which prevents the differentiation of the spatial structure of urban greenery based on canopy structure or leaf area density. Consequently, in this study, we have chosen the commonly used NDVI index, which reflects the density and health of vegetation, to indicate the spatial pattern of green spaces in the study area. Lastly, due to the scarcity of data, the temporal series of urban thermal environmental spatial variations based on functional zoning has not been included in the analysis. Therefore, future research will incorporate time series analysis, such as the differences in the contribution of urban functional zones to the urban thermal environment across different seasons or the evolutionary patterns of the contribution of urban functional zones to the urban thermal environment.

Although our study focuses on a select set of landscape indices, we acknowledge the presence of other relevant factors that may influence the urban heat island phenomenon. The selection of indices in this study was strategic, aiming to balance indices commonly found in urban planning regulations with those frequently utilized in thermal environment assessments. We intend to conduct a systematic review of commonly used indices in future research to further expand our understanding of the urban heat island effect.

5. Conclusions

Surface temperature, a critical variable in the Earth’s surface energy and water balance, is significantly influenced by the interplay between urban landscape patterns and thermal environments. This relationship, once understood, can be harnessed to mitigate risks associated with thermal environments. Given the complexity and heterogeneity of urban landscapes, our study began by segmenting the area into urban functional zones (UFZs) to minimize intra-zone landscape variability. Subsequently, we evaluated the correlation between the building–green space landscape pattern and surface temperature within each UFZ. Using low-temperature areas as a benchmark, we proposed optimization strategies for the landscape pattern in high-temperature areas, aiming to improve the thermal environment across the urban scale. This study reveals several key findings: (1) Due to the spatial heterogeneity of the building–green space landscape pattern across different functional zones, surface temperatures also exhibit strong spatial heterogeneity. (2) From the perspective of cold and hot spot zoning, the same landscape parameters apply to the four functional zones in this study area. From the results of temperature grading, the UVZ, COZ, and MUZ have the most landscape indices entering the regression equation. Even for the same parameters in different functional zones, the optimal threshold range for low-temperature areas varies, indicating that there are different cooling adjustment schemes for different functional zones. (3) In light of the uncertainties associated with estimating surface and near-surface air temperatures using the S_kyVF, this research introduces a novel three-dimensional parameter, S_unVF. S_unVF, which incorporates solar radiation dynamics, is particularly effective in signaling changes in surface temperature within UVZ and IWZ by the results from the RF analysis. While our study provides specific insights into the relationship between urban landscape patterns and the heat island effect within the context of Xiamen Island, the principles and methodologies employed can be applied to urban environments globally. Our study serves as a foundation for further research that can expand the understanding of urban heat islands and inform global urban planning practices.