Research on Thermal Environment Influencing Mechanism and Cooling Model Based on Local Climate Zones: A Case Study of the Changsha–Zhuzhou–Xiangtan Urban Agglomeration

Abstract

Urbanization has profoundly transformed land surface morphology and amplified thermal environmental modifications, culminating in intensified urban heat island (UHI) phenomena. Local climate zones (LCZs) provide a robust methodological framework for quantifying thermal heterogeneity and dynamics at local scales. Our study investigated the Changsha–Zhuzhou–Xiangtan urban agglomeration (CZXA) as a case study and systematically examined spatiotemporal patterns of LCZs and land surface temperature (LST) from 2002 to 2019, while elucidating mechanisms influencing urban thermal environments and proposing optimized cooling strategies. Key findings demonstrated that through multi-source remote sensing data integration, long-term LCZ classification was achieved with 1,592 training samples, maintaining an overall accuracy exceeding 70%. Landscape pattern analysis revealed that increased fragmentation, configurational complexity, and diversity indices coupled with diminished spatial connectivity significantly elevate LST. Rapid development of the city in the vertical direction also led to an increase in LST. Among seven urban morphological parameters, impervious surface fraction (ISF) and pervious surface fraction (PSF) demonstrated the strongest correlations with LST, showing Pearson coefficients of 0.82 and −0.82, respectively. Pearson coefficients of mean building height (BH), building surface fraction (BSF), and mean street width (SW) also reached 0.50, 0.55, and 0.66. Redundancy analysis (RDA) results revealed that the connectivity and fragmentation degree of LCZ_8 (COHESION8) was the most critical parameter affecting urban thermal environment, explaining 58.5% of LST. Based on these findings and materiality assessment, the regional cooling model of “cooling resistance surface–cooling source–cooling corridor–cooling node” of CZXA was constructed. In the future, particular attention should be paid to the shape and distribution of buildings, especially large, openly arranged buildings with one to three stories, as well as to controlling building height and density. Moreover, tailored protection strategies should be formulated and implemented for cooling sources, corridors, and nodes based on their hierarchical significance within urban thermal regulation systems. These research outcomes offer a robust scientific foundation for evidence-based decision-making in mitigating UHI effects and promoting sustainable urban ecosystem development across urban agglomerations.

1. Introduction

Since the onset of the 21st century, accelerating urbanization has emerged as a defining global phenomenon [1,2,3], with urbanization level now serving as a critical metric for evaluating socioeconomic advancement [4]. However this transformative process has concurrently triggered significant environmental challenges, including air pollution [5], loss of cultivated [6], water shortage and deterioration [7], noise pollution [8], etc., collectively undermining sustainable development objectives. Among these multifaceted ecological concerns, the UHI effect represents one of the most pressing environmental challenges [9], characterized by anomalous temperature elevation and consequent atmospheric system destabilization. Notably, it is worth noting that UHIs were observed in more than 1100 cities around the world during the 2015–2018 period [10]. In China, a report on 285 cities showed that UHIs had not occurred in only three cities [11]. Current evidence indicates that UHI impacts have progressively intensified, evolving into a persistent environmental stressor that critically compromises urban ecosystem stability and resilience.

Research on urban thermal environments can effectively elucidate the formation mechanisms and development trends of the UHI effect. Remote sensing (RS) offers the advantages of a large monitoring range, instantaneous imaging, and objectivity. It has been widely applied in areas such as the spatiotemporal distribution characteristics of UHIs [12], the relationship between UHIs and urban form [13], the relationship between UHIs and characteristics of the urban underlying surface pattern [14], the relationship between UHIs and vegetation distribution characteristics [15], and the relationship between UHIs and remote sensing parameters [16].

The primary objective of investigating UHI phenomena lies in developing targeted mitigation strategies to counteract thermal environmental deterioration. Notably, substantial scholarly attention had been concentrated on optimizing urban green infrastructure (GI), with particular emphasis on vegetated areas demonstrating substantial cooling capacities, thereby informing sustainable urban design principles. Some studies put the perspective on green spaces with significant cooling effects. For example, Hashim took Baghdad as a research area and found that LST was negatively correlated with the Normalized Difference Vegetation Index (NDVI) and positively correlated with the Normalized Difference Built-up Index (NDBI). Therefore, increasing vegetation cover was identified as a measure to improve the thermal environment [17]. Wu studied the relationship between vegetation connectivity and LST and suggested that adding larger green patches in urban areas may achieve better cooling effects [18]. Some studies also took the cooling effect of water bodies into account. Xiong found that the CONTAG of water bodies had a typical negative correlation with LST, and more open water should be planned in urban development [19]. Cai proposed the existence of a threshold for the cold island effect distance of water bodies, suggesting that within 500 m of a river bank, LST was predominantly influenced by the water body, while beyond this distance, building parameters became the primary influencing factor [20]. This conclusion was consistent with the views held by Wang [21]. Reasonable planning of urban patterns and urban morphology to reduce UHIs has also been widely investigated. Nevertheless, most of the improvement strategies proposed by these studies were mainly based on localization results, which were controversial. As it was generally believed that UHIs were negatively correlated with height–width ratio (H/W), some researchers suggested that deep street canyons should be planned [22]. However, more and more studies have shown that the relationship between H/W and LST is not linear, but rather there is a threshold for the value of H/W; that is to say, within the threshold, LST was positively correlated with H/W, while beyond the threshold, it was negatively correlated [23,24]. Another example indicated that building surface fraction (BSF) predominated as the key factor influencing UHIs at the local scale. A Beijing-based study further revealed more pronounced temperature biases in compact neighborhoods with higher BSF values [25]. However, some studies held the opposite view and believed that the increase in BSF would alleviate UHIs. Taking Hong Kong as an example, the BSF in high-rise and high-density building areas increased from 30% to 40%, but the LST decreased by 0.77 °C [26].

Overall, the research on UHIs and cooling strategies has achieved fruitful research results, but there were also shortcomings in the following three aspects: Firstly, the selection of land use types and parameters was highly subjective, and most studies selected several parameters of a certain land type for analysis. For example, only green space or water bodies were selected, only 2D parameters were selected, and the influence of 3D parameters on LST was rarely considered. Moreover, the generation mechanism of UHIs was extremely complex, and the cooling strategies proposed by the studies above may lack convincingness and integrity. Secondly, there were few studies on cooling strategies taking into account the spatial location information of low-temperature regions. Thirdly, adopting urban–rural dual structure zoning to study UHIs may not be rigorous. Some studies used prefecture-level municipal districts [27] or urban traffic loops [28] to distinguish the boundaries between “urban” and “suburban”, which are becoming more and more blurred nowadays, and due to advanced construction, the traffic loops of most cities are not absolute urban–rural boundaries.

In order to solve the deficiencies mentioned above, some researchers began to study the urban thermal environment from the perspective of climate characteristics, and surface classification systems such as urban terrain zones (UTZs) and urban climate zones (UCZs) appeared [29]. In 2012, Stewart and Oke introduced the local climate zones (LCZ) framework, a universally applicable and operational classification system [30]. This systematic scheme classified urban areas into 17 distinct types, comprising 10 built types and 7 natural land cover types, according to the relationships between key urban morphology parameters (building height, density, and surface roughness) and their associated thermal environmental effects. Without a doubt, LCZs were more concerned with the heterogeneity of the urban thermal environment at the local scale, which could provide decision makers with targeted design and management solutions.

At present, most LCZ mapping research employs a remote sensing image-based mapping method, which was integrated into WUDAPT workflow (The World Urban Database and Portal Tools: https://www.wudapt.org/) and follows a set of standard procedures described by Bechte [31]. Many achievements have been achieved in the study of urban thermal environments by applying WUDAPT. Nevertheless, two critical limitations persisted in current methodologies: First, the conventional WUDAPT framework primarily operates at the municipal scale with single-temporal data acquisition [32,33]. Although recent studies have extended LCZ investigations to urban agglomerations and multi-temporal analyses—as exemplified by Cai et al.’s work that expanded the study area to encompass the 16-city Yangtze River Delta urban agglomeration using aggregated biennial Landsat composites—such approaches still present technical constraints [34]. The authors themselves acknowledged that temporal discrepancies between Google Earth reference data and Landsat acquisitions may induce classification errors due to seasonal vegetation variations and ongoing construction changes. This underscores the necessity for systematic multi-temporal image utilization to enhance classification fidelity through phenological consistency. Second, training sample selection was time-consuming and labor-intensive. Summarizing the relevant research process of our predecessors, our study found that one LCZ class in a single city usually requires 30 samples, so it was estimated that at least 510 samples would be required for 17 LCZ classes. If the conventional WUDAPT framework was used to classify the urban agglomeration scale (assuming that the number of cities was three) for long-term series (assuming that there were four phases) LCZ classification, at least 6000 samples would be required, which may take a lot of time. Therefore, it was necessary to find a more concise and scientific strategy to achieve long-time series LCZ classification at the urban agglomeration scale.

In response to the above problems, this study met the requirements of refined exploration of urban thermal environments and took the Changsha–Zhuzhou–Xiangtan urban agglomeration (CZXA) with a significant UHI as the research area, explored the long-term LCZ classification at the urban agglomeration scale, investigated the impact mechanism of urban thermal environments in different spatial dimensions, and constructed a cooling model.

Specifically, the objectives of this study were as follows:

(a)

To develop an enhanced WUDAPT methodology: Improve the traditional long-term LCZ classification framework of WUDAPT by extending it beyond single-city, single-time analysis. This generates unified classifications across interconnected urban areas to achieve long-term LCZ classification for the multi-centered Changsha–Zhuzhou–Xiangtan urban agglomeration. Employ appropriate methods for LST retrieval and analyze the spatial–temporal dynamic characteristics of LCZs and LST across different periods.

(b)

To comprehensively analyze thermal environment drivers using LCZs: Integrate a quantitative LCZ framework with 2D landscape metrics and 3D urban form parameters. Utilize techniques such as image processing, statistical analysis, and mathematical modeling to explore the driving mechanisms of the urban thermal environment in multidimensional spaces. Extract key parameters and provide targeted recommendations for improving the urban thermal environment.

(c)

To construct a regional cooling model: Leverage LCZ-based insights on cooling source distribution and resistance factors. Incorporate relevant socioeconomic parameters to construct a comprehensive resistance surface for the cooling model. Identify cooling sources, cooling corridors, and cooling nodes, and evaluate their significance in building a regional cooling model. This provides decision-making support for improving the urban thermal environment and enhancing the quality and stability of ecosystems.

2. Data and Methods

2.1. Study Area

The Changsha–Zhuzhou–Xiangtan urban agglomeration (CZXA), strategically positioned in the central–eastern region of Hunan Province, encompasses three major metropolitan centers: Changsha, Zhuzhou, and Xiangtan. CZXA is the core growth pole of economic development in Hunan Province. However, the rapid development of urbanization and industrialization has also changed the land surface form of CZXA, triggering a series of environmental and ecological problems [35,36]. It is necessary to study the formation mechanism and governance countermeasures of UHIs in CZXA, which has been one of the most important economies in the central south region of China. The study area was about 8627 km², including 13 municipal districts and two counties, and was dominated by plains and hills with gentle terrain and had a subtropical monsoon climate, with high temperatures and less precipitation in summer, accounting for only 18% of the annual precipitation. Figure 1 shows the location and the scope of the study area.

The overall workflow is shown in Figure 2.

2.2. LST Retrieval and LCZ Classification

2.2.1. Preprocessing of the Landsat Data

Our study chose Landsat images in 2002/2008/2013/2019 obtained from USGS, which had been radiometrically calibrated and geo-rectified for LST retrieval and LCZ classification. According to the quality and availability of the data (cloud coverage rate less than 1% and period interval less than one month), we selected the late summer and early autumn dates for the research. The Landsat data used in this study are shown in

Table 1. Landsat data acquired in this study.

Data	Resolution	Path/Row	Date
Landsat5	Multispectral 30 m, Thermal Infrared 120 m.	123/40 123/41	3 September 2002 at 10:31 a.m.
			19 September 2008 at 10:41 a.m.
Landsat8	Multispectral 30 m, Thermal Infrared 100 m.		17 September 2013 at 10:59 a.m.
			17 August 2019 at 10:57 a.m.

.

With the support of ENVI 5.3, our study adopted radiometric calibration and FLAASH atmospheric correction, downsampled the Landsat data to 100 m, and obtained local-scale urban structures instead of the spectral signals of smaller objects to meet the requirements of LCZ classification.

2.2.2. LST Retrieval and Accuracy Assessment

Landsat5 TM6 and Landsat8 TIRS10 were selected to retrieve LST based on the radiative transfer equation (RTE) method, chosen because of its relatively high accuracy [37]. The principle of RTE is to estimate atmospheric effects on the surface heat radiation, and then subtract the effects of the atmosphere on land surface emissivity from the total radiation from the satellite sensor to obtain the heat radiation intensity [38].

The thermal infrared radiance $\mathrm{L}\mathsf{\lambda }$ received by satellite sensors consists of three components: (1) surface thermal radiation attenuated by atmospheric transmission to the sensor, (2) upward atmospheric radiation, and (3) downward atmospheric radiation reflected by the surface and transmitted through the atmosphere to the sensor. Assuming both the surface and atmosphere exhibit Lambertian properties for thermal radiation, the radiative transfer equation can be expressed as Equation (1):

$$L_{\lambda }=B_{\lambda }\left(T_s\right){\epsilon }_{\lambda }{\tau }_{\lambda }+L_{\lambda }^{\uparrow }+(1-{\epsilon }_{\lambda })L_{\lambda }^{\downarrow }{\tau }_{\lambda }$$

(1)

where ${\mathrm{L}}_{\mathsf{\lambda }}$ is the thermal infrared radiance at wavelength λ received by the sensor, ${\mathrm{B}}_{\mathsf{\lambda }}\left({\mathrm{T}}_{\mathrm{s}}\right)$ is the blackbody radiance at temperature ${\mathrm{T}}_{\mathrm{s}}$, ${\mathsf{\epsilon }}_{\mathsf{\lambda }}$ is the land surface emissivity, ${\mathsf{\tau }}_{\mathsf{\lambda }}$ is the atmospheric transmittance in the thermal infrared region, and ${\mathrm{L}}_{\mathsf{\lambda }}^{\uparrow }$ and ${\mathrm{L}}_{\mathsf{\lambda }}^{\downarrow }$ are the atmospheric upwelling radiance and downwelling radiance, respectively.

According to Planck’s law, LST can be retrieved from the thermal infrared radiation intensity emitted by a blackbody, as described by Equation (2):

$$T_s={\displaystyle \frac{K_2}{\mathit{ln}\left[\frac{K_1}{B\left(T_s\right)}+1\right]}}$$

(2)

where ${\mathrm{K}}_1$ and ${\mathrm{K}}_2$ are the radiance constants corresponding to the thermal infrared band. For Landsat 5 TM Band 6, the radiance constants are ${\mathrm{K}}_1=607.76 \mathrm{W}·{\mathrm{m}}^{-2}·{\mathsf{\mu }\mathrm{m}}^{-1}·{\mathrm{s}\mathrm{r}}^{-1}$ and ${\mathrm{K}}_2=1260.56 \mathrm{K}$. For Landsat 8 TIRS Band 10, the radiance constants are ${\mathrm{K}}_1=774.89 \mathrm{W}·{\mathrm{m}}^{-2}·{\mathsf{\mu }\mathrm{m}}^{-1}·{\mathrm{s}\mathrm{r}}^{-1}$ and ${\mathrm{K}}_2=1321.08 \mathrm{K}$. $\mathrm{B}\left({\mathrm{T}}_{\mathrm{s}}\right)$ can be derived from Equation (3):

$$B\left(T_s\right)=\left[L_{\lambda }-\left(L^{\uparrow }\right)-\tau \left(1-\epsilon \right)\left(L^{\downarrow }\right)\right]/\left(\tau \epsilon \right)$$

(3)

As shown in Equations (2) and (3), the parameters required by RTF are land surface emissivity $\epsilon$, atmospheric transmittance in the thermal infrared band $\tau$, atmospheric upwelling radiance $L^{\uparrow }$, and atmospheric downwelling radiance $L^{\downarrow }$. By entering the image acquisition time and center latitude/longitude into the NASA website (http://atmcorr.gsfc.nasa.gov) (accessed on 2 December 2021), the atmospheric profile parameters could be retrieved. The land surface emissivity could be calculated using Equation (4) [38]:

$$\epsilon =0.004F_v+0.986$$

(4)

${\mathrm{F}}_{\mathrm{v}}$ denotes the fractional vegetation coverage, which is formulated as Equation (5):

$$F_v={\displaystyle \frac{NDVI-N_s}{N_v-N_s}}$$

(5)

$\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}$ is the normalized difference vegetation index, ${\mathrm{N}}_{\mathrm{s}}$ is the value of the completely bare soil or non-vegetation area, and ${\mathrm{N}}_{\mathrm{v}}$ is the value of an area that is fully covered by vegetation. Empirically, ${\mathrm{N}}_{\mathrm{s}}$ is 0.05, and ${\mathrm{N}}_{\mathrm{v}}$ is 0.7. When $\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}$ > 0.7, the value of ${\mathrm{F}}_{\mathrm{v}}$ is 1, but when $\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}$ < 0.05, ${\mathrm{F}}_{\mathrm{v}}$ is 0. $\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}$ can be calculated using Equation (6):

$$NDVI={\displaystyle \frac{{\rho }_{nir}-{\rho }_{red}}{{\rho }_{nir}+{\rho }_{red}}}$$

(6)

${\mathsf{\rho }}_{\mathrm{n}\mathrm{i}\mathrm{r}}$ is the reflectance in the near-infrared band, and ${\mathsf{\rho }}_{\mathrm{r}\mathrm{e}\mathrm{d}}$ is the reflectance in the infrared band.

Since 2013, the China Meteorological Observatory has provided temperature data at three-hour intervals (e.g., 06:00, 09:00, 12:00, etc.) for three stations: Changsha, Zhuzhou, and Xiangtan. Beginning in 2019, it has provided hourly temperature data (e.g., 09:00, 10:00, 11:00, etc.) for four stations: Changsha, Zhuzhou, Xiangtan, and Mapoling. The transit time of Landsat 8 on 17 September 2013 was 10:59 a.m., which can be approximated as noon (12:00). The transit time on 17 August 2019 was 10:57 a.m., approximating 11:00 a.m. Considering potential locational deviations in the geographic coordinates of the meteorological stations and the use of 100 m resolution Landsat 8 thermal infrared data for LST retrieval in both 2013 and 2019, the LST retrieval accuracy in this study was evaluated as follows: a 100 m radius buffer was established centered on each meteorological station, and the mean LST within each buffer was calculated and compared with the station’s actual temperature. LST retrieval and accuracy assessment were processed in ENVI 5.3.

2.2.3. LCZ Mapping Method and Accuracy Assessment

In this paper, the conventional WUDAPT framework was improved to realize the long-term LCZ classification of CZXA. With the support of Google Earth historical images, field research data, Baidu Street View, and Anjuke, training samples of CZXA were produced using the improved methodology which contained the following three steps:

1. Selecting training samples in Google Earth: Representative areas of each LCZ class were selected by polygons as training samples (

Table A1. Snapshots of different LCZ classes in CZXA from Google Earth.

LCZ_1: Compact high-rise	LCZ_2: Compact mid-rise	LCZ_3: Compact low-rise
Tightly packed buildings with more than 10 stories Little or no green space Built from concrete, steel, stone, and glass	Tightly packed buildings with three to nine stories Little or no green space Built from stone, brick, tiles, and concrete	Tightly packed buildings with one to three stories Little or no green space Built from concrete, steel, stone, and glass
LCZ_4: Open high-rise	LCZ_5: Open mid-rise	LCZ_6: Open low-rise
Openly arranged buildings with more than 10 stories Abundance of green space Built from concrete, steel, stone, and glass	Openly arranged buildings with three to nine stories Abundance of green space Built from concrete, steel, and glass	Openly arranged buildings with one to three stories Abundance of green space Built from wood, brick, tiles, and concrete
LCZ_8: Large low-rise	LCZ_9: Sparsely built	LCZ_10: Heavy industry
Large, openly arranged buildings with one to three stories Little green space Land cover is mostly paved	Sparse arrangement of small or medium-sized buildings in natural setting Abundance of pervious cover	Low-rise and mid-rise industrial structures (towers, tanks, and stacks) Few or no trees
LCZ_A: Dense trees	LCZ_B: Scattered trees	LCZ_D: Low plants
Heavily wooded landscape of deciduous and/or evergreen trees	Lightly wooded landscape of deciduous and/or evergreen trees	Grass or herbaceous plants/crops
LCZ_E: Bare rock or paved	LCZ_F: Bare soil or sand	LCZ_G: Water
Rock or paved cover	Soil or sand cover	Large, open water bodies

). On average, each LCZ class contained about 30 polygons, and samples for each class were found from three cities and were distributed as evenly as possible on the imagery.

It should be emphasized that this study used Baidu Street View map to determine the number of floors, and compared the requirements for building spacing in the residential area design code to distinguish LCZ_1-3 and LCZ_4-6; we also employed Anjuke, a real estate information website, to judge the completion time, so as to determine the year to which the sample belongs. In addition, this study firstly identified samples that had not changed during the study period, such as dense vegetation, low-rise dense buildings, etc., and then identified other samples in each year by means of Google Earth historical images. A total of 15 LCZ classes (LCZ_7 and LCZ_C not found) were found and a total of 1592 samples of CZXA were selected in 4 years.

2. Supervised classification: The preprocessed Landsat images and 1592 samples were input into SAGA GIS (v2.0) to perform LCZ classification through the random forest (RF) algorithm at a spatial resolution of 100 m. The LCZ maps of CZXA in 4 years were therefore generated.

3. Accuracy evaluation and assessing LST intra-zonal variation among different LCZs.

The accuracy of the LCZ classification was crucial and underlies all later studies. In this study, two datasets were selected to verify the classification accuracy of LCZs. One was randomly selected from new LCZ training samples. Another was GLC_FCS30 (Global Land-cover Product with Fine Classification System at 30 m) data, which was produced by Professor Liu Liangyun of the Chinese Academy of Sciences [39,40,41]. After transform projection and downsampling, this study obtained GLC_FCS30 in 2000, 2005, 2010, 2015, and 2020. However, since the classification system of GLC_FCS30 was different from LCZs, we need to match the GLC_FCS30 data with the LCZ classification system [42]. In addition, GLC_FCS30 data were produced every five years; although there was a time offset of approximately two years between two datasets, it was generally considered acceptable.

Different LCZs had different thermal characteristics, so it was also essential to ensure that the LST of each LCZ class was significantly different so as to reflect the applicability of LCZs in UHI analysis [30]. One-way analysis of variance (ANOVA) was adopted to test the overall validity of the LCZs for the discriminative ability of LST, and Tukey’s Honest Significant Difference (Tukey’s HSD) was used to perform pairwise comparisons of all LCZ classes [43].

2.3. Calculation of Parameters Affecting Thermal Environment

2.3.1. Landscape Pattern Calculation

Landscape metrics highly condense landscape pattern information and are quantitative indicators that can reflect the composition of landscape structure and spatial configuration characteristics. Our study selected eight commonly used landscape indicators from the four aspects of fragmentation, complexity, aggregation, and diversity, including Percentage of Landscape (PLAND), Patch Density (PD), Largest Patch Index (LPI), Landscape Shape Index (LSI), Aggregation Index (AI), Contagion Index (CONTAG), Patch Cohesion Index (COHESION), and Shannon’s Diversity Index (SHDI). For a detailed explanation of these eight indicators, please refer to [44]. For the class-level landscape metrics, our study used suffixes according to the number in Table A1 to distinguish different LCZ classes. For example, the proportion of LCZ_1 coverage was represented as “PLAND1”, the patch density of LCZ_G was represented as “PDG”, and so on.

Moving window analysis can be used to quantify the landscape pattern in a specific range and clearly show the detail of the dynamic change in landscape pattern, which can achieve an in-depth understanding of the changes in various LCZ classes and the resulting changes in urban thermal environment at the local scale.

Landscape heterogeneity is highly scale-dependent. If the moving window size is too large, the details of landscape change will be lost, while if the window size is too small, the curve fluctuates too frequently, which is not conducive to analysis [45,46]. In order to select the optimal scale, given the actual resolution, our study chose 200 m, 400 m, 600 m, 800 m, 1000 m, 1200 m, and 1500 m as the size of the moving window to achieve the spatial distribution map of the landscape-level metrics along the east–west direction of the center of Changsha (Wuyi square), where the land use changes were the most frequent and dramatic in CZXA. As was shown in Figure A1, a smaller window size (less than 800 m) resulted in great fluctuations in the values of landscape metrics, while the curves were generally stable and the fluctuation ranges were small at larger window sizes (more than 800 m), so 800 m was the optimal scale of the moving window. In addition, 800 m was also close to the window size selected in previous studies on Changsha and Wuhan [47,48]. Landscape pattern calculations were performed using Fragstats 4.2.

2.3.2. Urban Volume Calculation

Digital surface model (DSM) data include the elevation of other features above the ground surface [49], such as buildings and trees. The DSM data adopted in our study had a resolution of 5 m, which was obtained after the dense matching of the stereo pair data of the ZY-3 satellite in 2019. The data of the bare-earth height information (‘JZM’ by default) used in this study was a digital elevation model (DEM) with a resolution of 12.5 m. These data were collected by the ALOS (Advanced Land Observing Satellite, launched in 2006) satellite phased array L-band synthetic aperture radar (PALSAR), which can be downloaded from Tuxinyun Platform (http://www.tuxingis.com/index.html) (accessed on 5 February 2022). Even in urban areas, the data did not include the elevation information of buildings and vegetation [50].

The elevation datums of DSM and JZM were both geodetic height, corresponding to the reference ellipsoid. There were no data anomalies such as data holes and morphological abnormalities in the two datums in the study area after inspection. JZM was resampled to a resolution of 5 m corresponding to the DSM by cubic convolution.

The volume is equal to the area multiplied by the elevation, so on the premise of the same area, the elevation of built types is introduced to calculate the urban volume in CZXA. Firstly, we subtracted JZM from DSM to obtain the elevations of all features above the surface. Secondly, we used the built type ROI to intercept the elevation data, while the elevations of natural land cover types were defaulted to 0. Finally, the urban volume was regarded as a spatial polyhedron composed of cuboids with the same base area and different heights, which was equal to the sum of cuboid accumulations, and the volume of each cuboid was the volume corresponding to each built type grid. It was worth noting that the urban volume used in this study not only included the volume of physical buildings but also included other building categories (such as viaducts, etc.) and vegetation located in the built types (such as greening in high-rise communities), so it could comprehensively reflect the changing trend of the volume of man-made objects. Our study chose volume mean (VM) and volume standard deviation (VSD), which were representative to measure the changes in urban volume. The calculation formulas of VM and VSD are shown in reference [44].

In order to intuitively reflect the overall urban volume changing trend of CZXA and combine volume with the landscape metrics for analysis, the volume distribution maps were also calculated using an 800 m moving window. Urban volume computation was carried out in Matlab 2012a.

2.3.3. Urban Morphology Calculation

Urban morphology refers to the mutual relationship and organizational characteristics of various urban elements in the horizontal and vertical directions [10]. Our study selected sky view factor (SVF), mean building height (BH), building surface fraction (BSF), mean street width (SW), height width ratio (H/W), pervious surface fraction (PSF), and impervious surface fraction (ISF) to analyze the impact of morphologies on the urban thermal environment. The definition and calculation formula of each parameter are shown in

Table 2. Urban morphology definition.

Morphologies	Definition	Data Source	Calculation Formula	Parameter Definition
SVF	The ratio of sky hemisphere visible from the ground	DSM	$\mathrm{S}\mathrm{V}\mathrm{F}=1-{\displaystyle \frac{\sum _{\mathrm{i}=1}^{\mathrm{n}}\sin {\mathsf{\gamma }}_{\mathrm{i}}}{\mathrm{n}}}$	λi is the height angle between observation point i and surrounding buildings. n is the number of horizon search directions.
BH	Mean building height of an LCZ grid	Building data	$\mathrm{B}\mathrm{H}={\displaystyle \frac{\sum _{\mathrm{i}=1}^{\mathrm{n}}\mathrm{B}\mathrm{S}\mathrm{i}\times \mathrm{B}\mathrm{H}\mathrm{i}}{\sum _{\mathrm{i}=1}^{\mathrm{n}}\mathrm{B}\mathrm{S}\mathrm{i}}}$	n is the number of buildings of the LCZ sample site. BSi is the ground area of a building. BHi is the height of a building.
BSF	The fraction of land surface covered by buildings	Building data	$\mathrm{B}\mathrm{S}\mathrm{F}={\displaystyle \frac{\sum _{\mathrm{i}=1}^{\mathrm{n}}\mathrm{B}\mathrm{S}\mathrm{i}}{{\mathrm{S}}_{\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{e}}}}$	BSi is the ground area of a building. Ssite is the area of the LCZ sample site.
SW	Mean street width of an LCZ grid	Street data	$\mathrm{S}\mathrm{W}={\displaystyle \frac{\sum \mathrm{S}\mathrm{s}\mathrm{t}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{t}}{\sum _{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{S}\mathrm{L}}_{\mathrm{i}}}}$	Sstreet is the total area of the LCZ sample site. SLi is the total length of the LCZ sample site.
H/W	The ratio of height to width of a street canyon	Building and street data	$\mathrm{H}/\mathrm{W}={\displaystyle \frac{\mathrm{B}\mathrm{H}}{\mathrm{S}\mathrm{W}}}$	$\mathrm{B}\mathrm{H}$ is the mean building height of an LCZ grid. SW is the mean street width of an LCZ grid.
PSF	Pervious surface fraction of the LCZ sample site	Remote sensing data	$\mathrm{P}\mathrm{S}\mathrm{F}={\displaystyle \frac{\sum \mathrm{S}\mathrm{p}\mathrm{e}\mathrm{r}}{\mathrm{S}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{e}}}$	Sper is the area of pervious surface of the LCZ sample site. Ssite is the area of the LCZ sample site.
ISF	Impervious surface fraction of the LCZ sample site	Remote sensing data	$\mathrm{I}\mathrm{S}\mathrm{F}=1-\mathrm{P}\mathrm{S}\mathrm{F}$

.

Scarano’s research showed that the number of n (search directions) had little effect on SVF, but R (the search radius), which mainly affected λi, had a great effect [51]. Another study by Downey and Zakšek showed that n = 8 was the appropriate minimum number of search directions for calculating SVF. If n > 32, SVF would not change significantly, and the appropriate number of search directions should be 8/16/32 [52]. In addition, n and R should also satisfy the following relationship [53]:

$${\displaystyle \frac{\mathrm{R}}{2}}<\mathrm{n}<\mathrm{R}$$

(7)

$R$ depended on the size of the smaller objects to be visualized. For this study, the DSM with a resolution of 5 m was used to calculate the SVF. Considering that the final result needed to be downsampled to a resolution of 100 m, this study chose 20 as the search radius. Combining the above formula, we can see that the number of suitable search directions is 16.

The street data were downloaded from OSM (Open Street Map), which contained 27 street types, but the width of each type was not provided. Combined with field research, SW calculation was performed by sampling each type of street width in Google Earth and conducting a buffer analysis in ArcGIS 10.8. Other parameters were calculated according to the formula given in Table 2.

2.4. Statistical Analysis

Similar to [44], this study applied Pearson correlation analysis, traditional regression, stepwise linear regression (SLR), and redundancy analysis (RDA) to explore the relationship between related parameters and urban thermal environment.

This study adopted Pearson correlation analysis to show the significant relationship between landscape metrics and LST. Both traditional regression and Pearson correlation analysis were applied to analyze the relationship between urban volume, urban morphologies, and LST. This study applied frequently used regression models such as linear, logarithm, quadratic polynomial, exponential, power function, etc., and then selected the model with the highest determination coefficient R². The positive and negative correlations were determined through the Pearson correlation coefficient. This step was processed on SPSS 25.

There were a total of 106 parameters of landscape pattern, urban volume, and morphology. From the perspective of the operability of improving the urban thermal environment, it was difficult to cover all the parameters. Therefore, for improving the pertinence, SLR was used to select the factors that were sensitive to thermal environment, and then RDA analysis was used to rank these factors [54]. Firstly, we standardized the landscape metrics, urban volume, urban morphologies (explanatory variable), and LST (dependent variable) to eliminate the differences in dimensions in the SLR analysis. The explanatory variables with high collinearity were tested and deleted through the VIF (variance inflation factor, VIF < 10) and tolerance (tolerance > 0.1) to make sure that explanatory variables entered in the RDA analysis were independent. Finally, we set the explanatory variables selected by SLR analysis as environment variables and LST as species variables for RDA analysis [55]. Based on the results of RDA, this study discussed the inner relationship between the three most explanatory variables and LST and proposed targeted solutions to alleviate the urban thermal environment. SLR was processed on SPSS 25, and RDA was run on the CANOCO 5.0 program (Microcomputer Power Company, Palo Alto, CA, USA).

2.5. Cooling Model Construction

Low-temperature patches serve as critical cooling sources for reducing LST. Enhancing the transmission and circulation of cold air, while improving the ecological quality between these patches, benefits the overall amelioration of the urban thermal environment. Previous studies have constructed few regional cooling models explicitly aimed at improving the urban thermal environment [56,57]. Furthermore, the selection of the resistance factor for constructing the ecological resistance surface has primarily relied on conclusions from prior studies rather than objective empirical evidence derived from dedicated experiments. Therefore, for the purpose of paying full attention to the role of low temperature patches as a cooling source, and making full use of the spatial information of the cooling sources to optimize the cooling pattern, based on the optimization principle of “combination of concentration and dispersion” [58], our study constructed a regional cooling model with the help of security pattern theory [59]. Specific steps were as follows:

1. Calculating the comprehensive resistance surface of the cooling model. The resistance parameters were selected from the four aspects of socioeconomic development, physical geography, urban construction, and land use barrier. According to previous experience and the actual situation of the study area, socioeconomic development included population density, GDP density [60,61], and PM2.5 concentration [62]; physical geography included elevation, slope [63], and vegetation coverage [17]; urban construction included key factors affecting the thermal environment obtained from the RDA analysis; land use barrier was based on a standard deviation interval; and the average LST of each LCZ class was grouped by the standard deviation level method. In addition, this study applied Jenks to divide the resistance value into 8 levels and used integers from 1 to 8 to represent the resistance coefficient. The larger the coefficient, the greater the resistance value. Finally, with the support of ArcGIS 10.8, spatial principal component analysis (SPCA) was used to determine the weight of each parameter [64]. Combined with the principal component loading matrix, the variance contribution rate of the first N principal components was calculated with the cumulative contribution rate of variance greater than 85% as the threshold, and the weight of each indicator was obtained.

2. Selection and identification of cooling sources. The cooling source is the core component of the regional cooling model. According to the ecological environment characteristics of the study area, several criteria for the selection of cooling sources were determined: (1) GI with cold island effect was the main source, including water bodies (LCZ_G) and vegetation (LCZ_A) [65]. Among them, dense vegetation is very important for urban cooling. The greater the vegetation coverage, the stronger the cooling effect [15], so LCZ_A was selected instead of LCZ_B and LCZ_D. (2) The planned ecological functional areas of the CZXA, such as national forest parks, important reservoirs, etc. (3) LCZ_A with an area larger than 0.5 km² and LCZ_G patch larger than 0.3 km² [66]. Considering the integrity of the cooling source as much as possible, the small patches that were close to and within the cooling source were also included.

3. Extraction of cooling corridors. Our study applied the minimum cumulative resistance mode (MCR) to calculate the cumulative resistance value that the cooling air needs to overcome to circulate between cooling sources and to determine the minimum cost path, thereby identifying the trend of cooling air flow and cooling corridors [67]. The calculation formula of MCR is as follows:

$$\mathrm{M}\mathrm{C}\mathrm{R}={\mathrm{f}}_{\mathrm{m}\mathrm{i}\mathrm{n}}\sum _{\mathrm{j}=\mathrm{n}}^{\mathrm{i}=\mathrm{m}}{\mathrm{D}}_{\mathrm{i}\mathrm{j}}\times {\mathrm{R}}_{\mathrm{i}}$$

(8)

$\mathrm{m}$ and $\mathrm{n}$ represent the number of the cooling source; $\mathrm{M}\mathrm{C}\mathrm{R}$ represents the minimum cumulative resistance from any cooling source to a certain point in space; $\mathrm{f}$ represents the positive correlation function between $\mathrm{M}\mathrm{C}\mathrm{R}$ and other variables; ${\mathrm{D}}_{\mathrm{i}\mathrm{j}}$ represents the spatial distance between the cooling sources $\mathrm{i}$ and $\mathrm{j}$; and ${\mathrm{R}}_{\mathrm{i}}$ represents the diffusion resistance coefficient of cooling source $\mathrm{i}$ in a certain direction in space. This study used Linkage Mapper 3.0.0 developed by Mc Rae and Kavanagh to extract the cooling corridors [68,69].

4. Importance evaluation and cooling model construction. Evaluating the importance of cooling sources, corridors, and nodes is the basis for building a regional cooling model, and it is also an important support for determining the priority of protection strategies. This study used the Circuitscape program of the Centrality Mapper toolbox to evaluate the current centrality value (CF_Central, CFC) of each cooling source [70] and then used the gravity model (GM) to calculate the interaction force between the sources to judge the importance of each cooling corridor. The calculation formula of the gravity model was as follows:

$${\mathrm{G}}_{\mathrm{i}\mathrm{j}}={\displaystyle \frac{{\mathrm{N}}_{\mathrm{i}}\times {\mathrm{N}}_{\mathrm{j}}}{{\mathrm{D}}_{\mathrm{i}\mathrm{j}}^2}}={\displaystyle \frac{\left(\frac{1}{{\mathrm{P}}_{\mathrm{i}}}\times \ln {\mathrm{S}}_{\mathrm{i}}\right)\times \left(\frac{1}{{\mathrm{P}}_{\mathrm{j}}}\times \ln {\mathrm{S}}_{\mathrm{j}}\right)}{{\left(\frac{{\mathrm{L}}_{\mathrm{i}\mathrm{j}}}{{\mathrm{L}}_{\mathrm{m}\mathrm{a}\mathrm{x}}}\right)}^2}}={\displaystyle \frac{{\mathrm{L}}_{\mathrm{m}\mathrm{a}\mathrm{x}}^2\times \ln {\mathrm{S}}_{\mathrm{i}}\times \ln {\mathrm{S}}_{\mathrm{j}}}{{\mathrm{L}}_{\mathrm{i}\mathrm{j}}^2\times {\mathrm{P}}_{\mathrm{i}}\times {\mathrm{P}}_{\mathrm{j}}}}$$

(9)

${\mathrm{G}}_{\mathrm{i}\mathrm{j}}$ represents the interaction force between cooling sources $\mathrm{i}$ and j; ${\mathrm{N}}_{\mathrm{i}}$ and ${\mathrm{N}}_{\mathrm{j}}$ represent the weight values of cooling sources $\mathrm{i}$ and $\mathrm{j}$, respectively; ${\mathrm{D}}_{\mathrm{i}\mathrm{j}}$ represents the normalized resistance value of the potential corridor between cooling sources $\mathrm{i}$ and $\mathrm{j}$; ${\mathrm{P}}_{\mathrm{i}}$ and ${\mathrm{P}}_{\mathrm{j}}$ represent the resistance values of cooling sources $\mathrm{i}$ and $\mathrm{j}$, respectively; ${\mathrm{S}}_{\mathrm{i}}$ and ${\mathrm{S}}_{\mathrm{j}}$ represent the areas of cooling sources $\mathrm{i}$ and $\mathrm{j}$, respectively; ${\mathrm{L}}_{\mathrm{i}\mathrm{j}}$ represents the cumulative resistance value of the potential corridor between cooling sources $\mathrm{i}$ and $\mathrm{j}$; and ${\mathrm{L}}_{\mathrm{m}\mathrm{a}\mathrm{x}}$ represented the maximum cumulative resistance value between cooling corridors.

Based on the results of CFC and ${\mathrm{G}}_{\mathrm{i}\mathrm{j}}$, our study used the quantile grading method to rank the cooling sources and cooling corridors, which were divided into three levels: extremely important, important, and generally important. In addition, the intersections between the cooling corridors were identified as cooling nodes [71]. The intersections between extremely important corridors and the intersections between extremely important and important corridors were defined as primary cooling nodes, while the intersections between extremely important and generally important corridors and the intersections between important corridors were defined as secondary cooling nodes. The intersection of important and general important corridors and the intersection between general important corridors were defined as tertiary cooling nodes. Finally, a regional cooling model of “cooling comprehensive resistance surface–cooling source–cooling corridor–cooling node” of CZXA was constructed.

5. Development of cooling strategies. Based on the established cooling model, planning and optimization recommendations were proposed for cooling sources, corridors, and nodes with varying levels of importance.

3. Results

3.1. Spatial–Temporal Distributions of LCZs and LST

It can be seen from

Table A2. The errors of LST retrieval in 2013 and 2019.

Years	Site Name	Longitude (Degree Minutes)	Latitude (Degree Minutes)	Measured Temperature/°C	Retrieved LST/℃	Errors/°C
17 August 2019 at 10:57 a.m.	Changsha	11,255	2813	33.60	33.26	−0.34
	Zhuzhou	11,310	2752	33.90	35.03	1.13
	Xiangtan	11,249	2752	34.30	35.04	0.74
	Mapoling	11,247	2807	34.20	34.66	0.46
17 September 2013 at 10:59 a.m.	Changsha	11,255	2813	33.20	32.03	−1.17
	Zhuzhou	11,310	2752	32.20	33.48	1.28
	Xiangtan	11,249	2752	32.20	32.29	0.09

that the errors of the retrieved LST were about 1 °C, which could meet the accuracy requirements.

The results of the LCZ accuracy assessment are shown in

Table A3. The results of the LCZ accuracy assessment.

The Accuracy of the New Samples			The Accuracy of GLC_FCS30
Years of New Samples	Overall Accuracy	Kappa Coefficient	Years of LCZs	Years of GLC_FCS30	Overall Accuracy	Kappa Coefficient
2002	77.30%	0.70	2002	2000	71.19%	0.70
2008	81.83%	0.74	2002	2005	71.58%	0.70
2013	77.27%	0.72	2008	2010	75.30%	0.62
2019	83.10%	0.77	2013	2015	73.33%	0.72
			2019	2020	74.78%	0.74

; the overall accuracy was greater than 70%, and the Kappa coefficient was greater than 0.7.

The results of Tukey’s HSD (

Table A4. The results of Tukey’s HSD.

Years	DF	Sig = 0	LCZ Pairs with Insignificant Differences
2002	14	LCZ_2 and LCZ_1, LCZ_3 and LCZ_1, LCZ_4 and LCZ_1, LCZ_5 and LCZ_1, LCZ_6 and LCZ_1, LCZ_8 and LCZ_1, LCZ_8 and LCZ_4, LCZ_10 and LCZ_1, LCZ_10 and LCZ_3, LCZ_E and LCZ_1, LCZ_F and LCZ_1	11/105
2008	14	LCZ_5 and LCZ_1, LCZ_E and LCZ_8, LCZ_F and LCZ_1	3/105
2013	14	LCZ_6 and LCZ_4, LCZ_10 and LCZ_2, LCZ_B and LCZ_9, LCZ_E and LCZ_1	4/105
2019	14	LCZ_10 and LCZ_1, LCZ_E and LCZ_1, LCZ_E and LCZ_10	3/105

) showed that in the comparison of 105 pairs in 15 LCZ classes, at the 0.05 confidence level, only in 2002 were there 11 pairs of groups that were not significantly different, which may be due to the very small number of LCZ_1 in 2002 (only 4 pixels), resulting in insignificant differences between LCZ_1 and nine other classes. In other years, only the differences between groups 3 and 4 were not significant; that is to say, more than 96% of the LCZ pairs had significant differences in temperature. Therefore, the improved WUDAPT classification method proposed in this study had a good degree of recognition for distinguishing and identifying relatively homogeneous temperature regions.

The LST retrieval results are shown in Figure 3. The LCZ maps are shown in Figure 4.

From Figure 3 and Figure 4, it can be seen that the characteristics of cities and suburbs were relatively clear, and the spatial pattern of the potential UHI had been detected. Changsha, Zhuzhou, and Xiangtan all spread to both sides with the Xiangjiang River as the center, and the urban area rapidly expanded, while the development of the outer suburbs was slow, which may be due to unbalanced development level. Changsha, the capital of Hunan Province and one of the new first-tier cities, had the largest urban area. With the support of relevant policies proposed by the Chinese government and the Hunan provincial government, the urban areas of three cities gradually converged and merged into a large spatial gathering area. In the downtown area of Changsha, high-rise buildings were often mixed with low- and medium-rise buildings (LCZ_2, 3, 5, and 6), which led to a compact urban structure and obvious UHI. Zhuzhou and Xiangtan, as ordinary prefecture-level cities, were relatively small and mainly consisted of residential areas. In addition, since the beginning of the 21st century, urban construction has begun to accelerate, and bare land (LCZ_F) has appeared in the suburbs. In 2013 and 2019, LCZ_F appeared on a large scale, destroying the natural features originally dominated by villages and farmland, which were developed into industrial areas, forming LCZ_8, or developed into residential and business districts, forming LCZ_2-6. Urban sprawl led to the erosion of lower-temperature land cover types, which were replaced by built types with higher temperature, resulting in a deteriorating urban thermal environment.

3.2. Correlation Between Urban Thermal Environment and Different Aspects Factors

The relationship between thermal environment and landscape metrics was complex. In

Table 3. Pearson coefficients of landscape metrics and LST.

Fragmentation
Years	PD	PD1	PD2	PD3	PD4	PD5	PD6	PD8
2002	0.47 **	0.04 **	0.39 **	0.33 **	0.31 **	0.47 **	0.46 **	0.42 **
2008	0.42 **	0.20 **	0.54 **	0.46 **	0.47 **	0.55	0.51 **	0.52 **
2013	0.59 **	0.28 **	0.64 **	0.54 **	0.48 **	0.33 **	0.52 **	0.66 **
2019	0.68 **	0.35 **	0.74 **	0.61 **	0.66 **	0.64 **	0.55 **	0.74 **
Years	PD9	PD10	PDA	PDB	PDD	PDE	PDF	PDG
2002	−0.20 **	0.33 **	−0.19 **	0.23 **	0.09 **	0.39 **	0.42 **	0.24 **
2008	−0.21 **	0.26 **	−0.22 **	−0.15 **	−0.04 **	0.43 **	0.40 **	0.06 **
2013	0.03 **	0.29 **	−0.14 **	−0.08 **	−0.31 **	0.55 **	0.46 **	−0.07 **
2019	0.08 **	0.32 **	−0.23 **	−0.20 **	−0.16	0.60 **	0.38 **	0.00 **
Years	PL1	PL2	PL3	PL4	PL5	PL6	PL8	PL9
2002	0.04 **	0.27 **	0.25 **	0.30 **	0.45 **	0.40 **	0.35 **	−0.08 **
2008	0.16 **	0.48 **	0.39 **	0.40 **	0.52 **	0.54 **	0.52 **	−0.01 **
2013	0.24 **	0.56 **	0.49 **	0.48 **	0.38 **	0.56 **	0.68 **	−0.09 **
2019	0.28 **	0.65 **	0.56 **	0.64 **	0.60 **	0.53 **	0.79 **	−0.15 **
Years	PL10	PLA	PLB	PLD	PLE	PLF	PLG
2002	0.28 **	−0.38 **	0.15 **	0.05 **	0.33 **	0.33 **	0.07 **
2008	0.26 **	−0.18 **	0.00 **	0.12 **	0.43 **	0.42 **	0.19 **
2013	0.20 **	−0.29 **	−0.06 **	−0.42 **	0.51 **	0.39 **	−0.21 **
2019	0.23 **	−0.54 **	−0.16 **	−0.23 **	0.53 **	0.33 **	−0.09 **
Years	LPI	LPI1	LPI2	LPI3	LPI4	LPI5	LPI6	LPI8
2002	−0.24 **	0.09 **	0.38 **	0.33 **	0.39 **	0.52 **	0.44 **	0.42 **
2008	−0.24 **	0.05 **	0.29 **	0.26 **	0.32 **	0.39 **	0.38 **	0.38 **
2013	−0.35 **	0.27 **	0.48 **	0.46 **	0.41 **	0.35 **	0.47 **	0.60 **
2019	−0.52 **	0.32 **	0.59 **	0.57 **	0.65 **	0.53 **	0.45 **	0.70 **
Years	LPI9	LPI10	LPIA	LPIB	LPID	LPIE	LPIF	LPIG
2002	0.02 **	0.33 **	−0.19 **	−0.08 **	−0.24 **	0.36 **	0.32 **	−0.15 **
2008	0.06 **	0.26 **	−0.20 **	0.00 **	−0.22 **	0.36 **	0.35 **	−0.16 **
2013	−0.06 **	0.18 **	−0.26 **	−0.05 **	−0.39 **	0.49 **	0.34 **	−0.20 **
2019	−0.14 **	0.22 **	−0.50 **	−0.15 **	−0.22 **	0.48 **	0.31 **	−0.09 **
Complexity
Years	LSI	LSI1	LSI2	LSI3	LSI4	LSI5	LSI6	LSI8
2002	0.36 **	0.08 **	0.58 **	0.50 **	0.40 **	0.41 **	0.52 **	0.56 **
2008	0.30 **	0.21 **	0.55 **	0.48 **	0.49 **	0.54 **	0.52 **	0.50 **
2013	0.45 **	0.31 **	0.64 **	0.58 **	0.45 **	0.29 **	0.49 **	0.66 **
2019	0.56 **	0.39 **	0.76 **	0.64 **	0.58 **	0.63 **	0.56 **	0.78 **
Years	LSI9	LSI10	LSIA	LSIB	LSID	LSIE	LSIF	LSIG
2002	−0.29 **	0.47 **	−0.20 **	0.00 **	−0.07 **	0.50 **	0.44 **	0.02 **
2008	−0.22 **	0.29 **	−0.33 **	−0.14 **	−0.17 **	0.45 **	0.42 **	0.06 **
2013	−0.15 **	0.30 **	−0.27 **	−0.09 **	−0.43 **	0.56 **	0.46 **	−0.08 **
2019	−0.07 **	0.34 **	−0.43 **	−0.22 **	−0.22 **	0.62 **	0.40 **	0.00 **
Aggregation
Years	AI	AI1	AI2	AI3	AI4	AI5	AI6	AI8
2002	−0.35 **	0.00 **	0.43 **	0.37 **	0.16 **	0.38 **	0.32 **	0.37 **
2008	−0.29 **	0.11 **	0.41 **	0.40 **	0.35 **	0.43 **	0.43 **	0.39 **
2013	−0.43 **	0.13 **	0.45 **	0.38 **	0.31 **	0.24 **	0.34 **	0.54 **
2019	−0.54 **	0.16 **	0.51 **	0.30 **	0.46 **	0.41 **	0.30 **	0.65 **
Years	AI9	AI10	AIA	AIB	AID	AIE	AIF	AIG
2002	−0.06 **	0.30 **	−0.19 **	−0.03 **	−0.22 **	0.26 **	0.32 **	−0.05 **
2008	0.03 **	0.13 **	−0.31 **	−0.12 **	−0.31 **	0.23 **	0.35 **	−0.07 **
2013	−0.05 **	0.14 **	−0.28 **	−0.03 **	−0.40 **	0.30 **	0.37 **	−0.11 **
2019	−0.11 **	0.14 **	−0.49 **	−0.11 **	−0.21 **	0.28 **	0.34 **	−0.03 **
Years	CO	CO1	CO2	CO3	CO4	CO5	CO6	CO8
2002	−0.40 **	0.00 **	0.48 **	0.39 **	0.20 **	0.52 **	0.43 **	0.45 **
2008	−0.38 **	0.13 **	0.46 **	0.42 **	0.39 **	0.50 **	0.51 **	0.48 **
2013	−0.55 **	0.17 **	0.54 **	0.44 **	0.42 **	0.36 **	0.49 **	0.66 **
2019	−0.64 **	0.19 **	0.60 **	0.37 **	0.60 **	0.55 **	0.44 **	0.77 **
Years	CO9	CO10	COA	COB	COD	COE	COF	COG
2002	−0.13 **	0.33 **	−0.19 **	−0.06 **	−0.22 **	0.28 **	0.35 **	−0.07 **
2008	−0.05 **	0.15 **	−0.31 **	−0.16 **	−0.31 **	0.27 **	0.38 **	−0.09 **
2013	−0.14 **	0.17 **	−0.31 **	−0.05 **	−0.42 **	0.36 **	0.40 **	−0.14 **
2019	−0.16 **	0.16 **	−0.53 **	−0.13 **	−0.24 **	0.37 **	0.35 **	−0.04 **
Aggregation		Diversity
Years	CONTAG	Years	SHDI
2002	−0.13 **	2002	0.41 **
2008	−0.09 **	2008	0.37 **
2013	−0.39 **	2013	0.54 **
2019	−0.50 **	2019	0.66 **

Note: ** indicates significance at the 0.01 level. Darker red shades indicate higher positive correlations, while darker blue shades represent stronger negative correlations.

, darker red shades indicate higher positive correlations, while darker blue shades represent stronger negative correlations. It can be seen from Table 3 that in the built types, strong correlations between LCZ_2 (PD2, PLAND2, and LSI2), LCZ_4 (PD4, PLAND4, LPI4, and LSI4), LCZ_5 (PD5, PLAND5, and LSI5), LCZ_8 (PD8, PLAND8, LPI8, LSI8, and AI8), and LST were shown in multiple years, which resulted in the increase in fragmentation degree, the increase in area ratio, and the deterioration of connectivity leading to an increase in LST. In terms of positive and negative correlation, LST usually had positive correlations with the landscape metrics of built types (except LCZ_9) and had negative correlations with the land cover types (except LCZ_E and LCZ_F).

During the research period, the overall correlation between landscape-level metrics and LST increased, indicating that landscape pattern has a more significant impact on the formation mechanism of regional thermal environment. The importance of landscape patterns in mitigating UHIs has significantly increased. Compared with the impact of a single land cover type, landscape configuration has become a key factor in regulating regional thermal environment pattern and a key lever for achieving climate adaptation planning.

The distribution of urban volume, VM, and VSD is shown in Figure 5. The three cities were all built along the Xiangjiang River, and the intensity of urbanization on both sides of the Xiangjiang River was the greatest. However, unlike Xiangtan and Zhuzhou, the peaks of VM and VSD in Changsha were also concentrated in some new urban areas far away from the Xiangjiang River, and Changsha presented a trend of multi-core development in the vertical direction. And it can be seen from

Table 4. Regression results of urban volume metrics and LST.

Urban Volume Metrics	Mean (m3)	Regression Equations with LST	R2	Pearson
VM	14031	$y=1.40x+32.99$	0.40 **	0.63
VSD	18659	$y=0.83x+32.90$	0.21 **	0.46

Note: ** indicates significance at the 0.01 level.

that LST had strong correlation with VM and a moderate correlation with VSD.

It can be seen from

Table 5. Regression results of urban morphologies and LST.

Urban Morphologies	Regression Equations with LST	R2	Pearson
SVF	$y=3E-05x+34.31$	0.00 **	0.20
BH	$y=0.25x+34.24$	0.25 **	0.50
BSF	$y=30.41x+34.20$	0.31 **	0.55
SW	$y=0.46x+33.45$	0.44 **	0.66
H/W	$y=3.64x+34.48$	0.02 **	0.15
PSF	$y=-9.45x+42.91$	0.68 **	−0.82
ISF	$y=9.45x+33.46$	0.68 **	0.82

Note: ** indicates significance at the 0.01 level.

that ISF, PSF, and LST showed strong correlation, and the Pearson coefficients reached 0.82 and −0.82, respectively, which indicated that impervious surface was an important factor leading to LST increasing, while pervious surface could effectively alleviate UHIs. Both BH and BSF showed a moderate positive correlation with LST, while SW showed a strong positive correlation with LST. This indicated that the larger the coverage area and height of the building, as well as the wider the width of the street, all led to an increase in LST. The goodness of fit between SVF, H/W, and LST was very low, and the specific explanation will be discussed in Section 4.1.

3.3. Redundancy Analysis Results

Table 6. SLR parameters of LST and feature parameters.

No.	Metrics	Standardized Coefficient	Tolerance	VIF
1	PLAND1	0.044 **	0.449	2.229
2	AI1	0.011 **	0.671	1.49
3	PLAND3	0.046 **	0.32	3.128
4	AI3	−0.012 **	0.699	1.431
5	AI5	−0.015 **	0.515	1.941
6	PD8	0.057 **	0.204	4.905
7	COHE8	−0.044 **	0.116	8.616
8	AI9	0.048 **	0.123	8.159
9	PLAND10	0.045 **	0.752	1.33
10	LSIB	0.069 **	0.361	2.767
11	LSID	−0.093 **	0.227	4.401
12	PLANDE	0.044 **	0.524	1.907
13	LPIF	0.035 **	0.393	2.544
14	LSIG	0.041 **	0.461	2.169
15	VM	0.104 **	0.126	7.966
16	VSD	−0.078 **	0.193	5.176
17	SVF	0.111 **	0.211	4.749
18	BH	−0.059 **	0.273	3.665
Goodness	R2	0.885

Note: ** indicates significance at the 0.01 level.

showed that there were 18 indicators with high correlation with LST, including 14 landscape metrics, 2 urban morphological parameters, and 2 urban volume indices. The landscape metrics of LCZ_1, LCZ_3, LCZ_5, LCZ_8, LCZ_9, LCZ_10, and LCZ_B-G had high correlation with LST.

The RDA results showed (Figure 6) that the above parameters accounted for 78.9% of LST, with COHESION8 having the highest degree of explanation (58.5%), followed by VM (8.1%) and PD8 (5.5%), which were all positively correlated. LCZ _8 was the most important factor affecting urban thermal environment. The increase in the connectivity (COHESION8) and fragmentation (PD8) of LCZ_8 and the increase in the mean urban volume (VM) would lead to the increase in LST.

The results shown in Figure 6 were also consistent with the correlations presented in Table 3 and Table 4. Specifically, Table 3 indicated strong correlations between all landscape metrics of LCZ_8 and LST. For instance, in 2019, the Pearson correlation coefficients between LST and PD8, PL8, LPI8, LSI8, AI8, and COHESION8 reached 0.74, 0.79, 0.70, 0.78, 0.65, and 0.77, respectively. Urban development had progressed rapidly in the vertical dimension. Similarly, Table 4 showed a Pearson correlation coefficient of 0.63 between VM and LST, which also corroborated that vertical urban development was a key factor influencing the urban thermal environment in CZXA.

3.4. Cooling Model

The data source and preprocessing of the resistance surface parameters are shown in

Table A5. The sources and preprocessing of the parameters of the regional cooling model.

Criterion Layer	Parameters	Data Sources	Preprocessing
Socioeconomic Development	Population Density	Resource and Environment Science and Data Center	Projection Transformation, Downsampling
	Gross Domestic Product (GDP) Density		Projection Transformation, Downsampling
	2.5-micrometer Particulate Matter (PM2.5) Concentration	China High Air Pollutants (CHAP)	Panoply, ArcGIS, Inverse Distance Weight (IDW), Projection Transformation, Resampling
Physical Geography	Elevation	USGS-SRTM	Projection Transformation, Downsampling
	Slope	USGS-SRTM	ArcGIS, Slope Calculation Tool
	Vegetation Coverage	Landsat imagery	Threshold segmentation with NDVI
Urban Construction	COHESION8	Conclusion of the previous article	RDA analysis
	VM
	PD8
Land Use Barrier	Distribution of each LCZ	Conclusion of the previous article	The standard deviation level method

. The comprehensive resistance surface of the cooling model is shown in Figure 7, which found that the resistance value in the urban area was generally higher than the southeastern, northeastern, and southwestern mountains, and the “Green Heart” area. The distribution characteristics of resistance value echoed the pattern of thermal environment and urban sprawl.

Our study selected a total of 50 cooling sources with a total area of 2136.10 km² and extracted 139 cooling corridors with a total length of 2576.04 km. The cooling source is the patch that plays a vital role in cooling pattern. Protecting extremely important corridors was irreplaceable for improving the connectivity of air-conditioning transmission. Among the 50 cooling sources, 17 were extremely important, accounting for 57.73% of the total source area, mainly included the main stream and tributaries of the Xiangjiang River and some national parks. The cooling corridor is the line segment for cooling air circulation between the cooling sources, which can judge the trend of the cooling air flow and is also the key component of the regional cooling model. The total length of the 139 corridors was 2576.04 km, and the extremely important corridors are 418.04 km long, mainly distributed in the northern mountainous areas of Changsha, the western suburbs of Xiangtan, and the southern mountainous areas of Zhuzhou. The cooling node is the point where each cooling corridor converges and intersects. A total of 68 cooling nodes were extracted, including 27 primary cooling nodes. Strengthening the construction of nodes to prevent erosion due to excessive external resistance can improve the smoothness and stability between the cooling corridors. Finally, the regional cooling model of CZXA was constructed (Figure 7). The detailed statistical results of cooling sources, corridors, and nodes of different degrees of importance are shown in

Table 7. The statistical results of cooling sources, corridors, and nodes of different degrees of importance.

Cooling Sources	Extremely Important	Important	Generally Important	Cooling Corridors	Extremely Important	Important	Generally Important	Cooling Nodes	Primary	Secondary	Tertiary
Number	16	17	17	Number	46	46	47	Number	27	26	15
Area	1233.30	407.08	495.76	Total Length	418.04	805.11	135.88
Area Ratio	57.73	19.06	23.21	Average Length	8.89	17.50	29.41
				Length Ratio	16.23	31.25	52.52

.

4. Discussion

4.1. Analysis of Influence Mechanism of Urban Thermal Environment

Within the study period, urban expansion in both the horizontal and vertical directions is becoming more and more obvious. This phenomenon causes the thermal environment to be affected by a variety of factors.

Compared with LCZ_2, LCZ_4, and LCZ_5, LCZ_8 had larger area, more uniform building materials and more heat absorption, although the building materials were all stone, brick, tile, and concrete. With the expansion of the city, a large number of large, openly arranged buildings belonging to LCZ_8 appeared in the suburbs of CZXA, which played an important role in stimulating the economy but also aggravated the accumulation of heat in the city and formed a larger-scale heat island. For the land cover types, our study focused on LCZ_A, because PD, PLAND, LPI, LSI, AI, and COHESION of LCZ_A were all negatively correlated with LST. But in detail, connectivity may have a greater impact on LST than fragmentation and complexity, which demonstrated that the aggregated distribution provided a stronger cooling effect than the fragmented and complex distribution when given a fixed amount of vegetation cover [72]. For CZXA, reducing the fragmentation degree and centralized layout of built types, especially LCZ_2, LCZ_4, LCZ_5, and LCZ_8, and interspersing cooling patches such as vegetation inside built types to increase the dominance and connectivity of cooling patches was of positive significance to improve the regional thermal environment.

Urban development became increasingly drastic in the three-dimensional direction. Judging from the urbanization process in China, the larger the VM, that is, the higher the urbanization level, the greater the land hardening rate, the stronger the anthropogenic heat emission, and the more possible it was to increase the LST. The large VSD meant that there was a big height difference between the buildings in the area, which brought about the improvement of urbanization and vertical space differentiation and was not conducive to the formation of regular street canyons. The asymmetrical and discontinuous buildings in height also had an impact on street canyon ventilation conditions and would eventually lead to a rise in LST [73]. Therefore, in terms of urban volume, under the premise of land intensification, reducing the building density and the height difference between buildings in adjacent areas, increasing the distance between buildings, and forming a relatively neat sky boundary were beneficial to improving the urban thermal environment.

The increase in ISF was one of the main characteristics of urbanization. The larger the impervious surface area per unit area, the easier it was to cause the temperature to rise. While the increase in PSF meant that there were more natural features such as vegetation and water bodies, and the cooling effect was more obvious, the rest of the urban morphological parameters were positively correlated with LST. In addition, regions with higher BSF, BH, and SW often had higher temperatures. The reason was that these areas were mainly located in the city center, which had sparse buildings, narrow roads, and abundant green plants, unlike the suburbs, causing an inevitable temperature increase. SVF calculation was based on DSM, which ensured the continuity and integrity of the results. In the suburbs, although the SVF differences between the plain and the mountain were large, they both showed low temperatures, while in the urban area, although the temperature difference within the city was small, the SVF difference was large in areas with different building densities and building heights. However, unlike highly developed cities, such as Melbourne [44] and Hong Kong [74], the urban construction of CZXA still existed in the form of increasing the urban surface area, so the correlation between LST and SVF, representing the proportion of visible sky and open canyon space in urban street valleys in the whole region, was not obvious. In addition, constrained by the lack of coverage of the data source, H/W in the suburbs was 0 by default, which may be the reason for the low correlation. Therefore, after removing the 0 value, the correlation between H/W and LST was found to be higher after refitting; the formula was y = 2.73x + 35.68, R² = 0.49, and the Pearson coefficient was 0.70. H/W was affected by the width of the street and the height of the buildings on both sides of the street. Under the premise of a certain street width, with a higher height of the canyon buildings (mostly concentrated in the city center), the increase in H/W would lead to temperature rise. However, in wider street canyons, wind velocities were more comfortable, and the cooling effect was better [75]. This finding was consistent with the high correlation between BH and LST. In fact, several previous studies demonstrated that the influence of urban morphological indicators on the thermal environment was more significant at the micro-scale and local scale. The correlation between urban morphological parameters and LST would fluctuate to a certain extent, but this fluctuation would be smoothed and become irrelevant as the scale increased, so on larger scales, urban morphological indicators had a weaker impact on climate, such as SVF, H/W, and street orientation. A study in the Netherlands found that SVF and street building heights were key contributors to the occurrence of high temperature in street canyons at the local scale but had little effect at the urban scale [76]. This also explained the lower correlation between SVF and H/W compared with other indicators in the study area.

4.2. Key Parameters Affecting Thermal Environment

To investigate the influence of key factors on the urban thermal environments in detail, our study conducted a concrete and quantitative analysis based on the results of RDA and the development status of the economy, society, and related policies. In CZXA, our study indicated that the urban thermal environment was highly correlated with the landscape metrics of LCZ_8 (large low-rise) and Volume Mean. LCZ_8 was the most critical factor changing the temperature, and the total explanation of COHESION8 and PD8 reached 64%. The rapid development of the city was inseparable from the industry. In 2005, Changsha, Zhuzhou, and Xiangtan successively issued policies to accelerate the construction and development of an advantageous industrial cluster. Since then, the amount of investment attracted has increased year by year, and various large-scale industrial zones have been built by the local government. Data from the China Prospective Industry Research Institute (https://f.qianzhan.com/) show that there are currently 1202 industrial parks in CZXA, most of which exist in the form of large low-rise buildings (LCZ_8). Such industrial parks were mainly composed of modern building materials such as steel, iron, and cement, which have extremely poor sensible heat storage capacity. In addition, as an industrial plant, LCZ_8 also emits a large amount of artificial and industrial heat, and the LCZ maps show that LCZ_8 is mostly located on the edge of the city and surrounds the urban area, making it difficult for the air inside and outside the city to circulate. Therefore, the occurrence of a large number of LCZ_8 in CZXA may be an important factor leading to the increase in temperature in the urban area, which was similar to the conclusion of a study by Hefei [77]. VM also had a strong explanation for LST (8.1%). With the increasing level of urbanization, a large number of buildings were built to meet the needs of urban expansion. Tall building landscapes with high building density were distributed in urban areas, which had a significant warming effect on LST. Therefore, the larger the number, height, and density of the buildings, the higher the LST was bound to be.

The results of key factor extraction showed that, from an overall perspective, although the cluster layout of industrial zones was conducive to the development of advantageous industries and economic development, it was undeniable that from the perspective of improving the urban thermal environment and optimizing the regional ecological environment, CZXA should create a more reasonable layout for the industrial park and try to avoid the existence of new industrial plants in the form of a circular distribution around the urban area; otherwise, it would greatly exacerbate the UHI. In addition, reducing the aggregation of new industrial plants and optimizing the internal structure of existing industrial parks should be considered, such as increasing vegetation coverage and increasing green space patches inside and outside industrial areas (such as street trees, etc.), which were of great significance for reducing the UHI and controlling microclimate effects [78,79]. Our study also suggested that urban expansion should be appropriately slowed. While urbanization was crucial for economic development, slowing new city construction, controlling building density, and preventing excessive “hot spots” were imperative to reduce UHI intensity and extent, thereby promoting sustainable development in CZXA.

4.3. Cooling Suggestions Based on Cooling Model

After evaluating the importance of the cooling sources, corridors, and nodes by CFC and the gravity model, our study constructed a regional cooling model to provide basis and decision support for CZXA to formulate ecological protection and restoration strategies:

1. The ecological quality of extremely important cooling sources, cooling corridors, and primary cooling nodes was the best, which will have a huge impact on the overall cooling pattern if they are missing [80]. Therefore, the following strategies are recommended to enhance urban thermal environment management: (1) Prioritize ecological protection to obtain better cooling benefits. For regions dominated by dense vegetation, strict adherence to the “Green Heart Area Protection Regulations” enacted by the Hunan Provincial People’s Congress Standing Committee should be enforced. We recommend strictly limiting the construction of non-essential urban expansion projects, excluding those involving ecological restoration, landscape protection, essential public facilities, and local rural residential development. Concurrently, efforts should focus on continuously improving regional ecological quality. This includes prohibiting the degradation and pollution of water bodies while appropriately increasing the area, distribution density, and dominance of aquatic ecosystems within the study area. (2) Increase the internal filling of the source. For cooling sources with large internal pores, the filling of the internal source range should be strengthened to increase the proportion of cooling patches inside the source. (3) Formulate effective laws and regulations to eliminate projects posing ecological risks. Governments should facilitate the withdrawal of construction and industrial ventures that may cause environmental pollution.

2. For important cooling sources, cooling corridors, and secondary cooling nodes, restoration should be the main focus, and related protection measures could refer to strategies for more important sources, corridors, and nodes, but their importance and urgency may decrease. In addition, limiting the construction scope and human disturbance, reducing the ecological pressure in the area, filling in the cooling patches inside the source, and taking measures to prevent ecological damage were all thoughtfully designed strategies [81].

3. For generally important cooling sources, cooling corridors, and tertiary cooling nodes, maintenance should be given priority. We propose that the original land use patterns can be maintained in harmony with the ecosystem, and the total number of cooling sources in urban areas should be increased, cooling sources should be distributed reasonably, and “stepping stone patches” dominated by GI should be added appropriately, such as multiple continuous parks, along cooling corridors with long distances to enhance the stability of the cooling corridor [80]. Another viable approach involves increasing the density of small gardens and groves capable of regulating the urban microclimate, an effect demonstrated in cooling studies conducted in Fuzhou [82].

4. The construction of multi-functional cooling corridors should be encouraged. Taking full advantage of the abundant water and forest resources of CZXA, the government can combine the regional cooling model serving the construction of ecological civilization with economic and cultural construction to build multi-functional cooling sources and cooling corridors. For example, Songya Lake Wetland Park (cooling source 13) not only played an important role in regulating climate and protecting biodiversity but also improved the urban living and investment environment and provided places for recreation and education. Another example was the typical waterfront corridor between cooling sources 15 and 44, and the typical road-type green corridor between cooling sources 40 and 47.

5. Conclusions

This study investigated urban thermal environment dynamics and their underlying mechanisms through a refined analytical lens, aiming to develop targeted mitigation strategies for UHI effects. Leveraging remote sensing (RS) and geographic information system (GIS) technologies, our study conducted a comprehensive analysis of local climate zone (LCZ) evolution and land surface temperature (LST) variations in CZXA since the start of the 21st century. Through multidimensional assessment of thermal environment drivers, this study identified critical factors exhibiting strong LST correlations and developed a predictive cooling model incorporating socio-economic parameters. Our findings provided evidence-based solutions for ecological optimization, thermal environment improvement, and sustainable urban development in rapidly urbanizing regions.

Based on multi-source data, our study only selected 1592 samples to achieve the LCZ maps of CZXA within 4 years; the overall accuracy was greater than 70%, and more than 96% of the LCZ pairs had significant differences in temperature. The increase in fragmentation, complexity, and diversity and the decrease in connectivity could easily lead to an increase in LST. The landscape metrics of built types were positively correlated with LST, of which land cover types were negatively correlated with LST. In addition, aggregation had a greater impact on the thermal environment than other landscape metrics. Larger, contiguous land cover types had a stronger cooling effect than discrete ones. Urban volume also had a great influence on LST, and the vertical development of the city can easily lead to an increase in temperature. Among the urban morphological parameters, ISF, PSF, and LST had strong positive correlations. LCZ_8 was the key land cover type that affected the urban thermal environment, and its reasonable distribution was important for alleviating UHIs. Based on this, our study constructed a regional cooling model of “cooling comprehensive resistance surface–cooling source–cooling corridor–cooling node” and proposed appropriate protection strategies. The research results are a big step forward in understanding the influence mechanisms of urban thermal environments and improving urban thermal environment and ecological quality for cities with similar natural endowments to CZXA. However, this study had several limitations. Firstly, the research did not take into account the economic costs associated with implementing cooling strategies. Secondly, although ENVIMET software has been extensively applied in both academic research and engineering practices—such as examining microscale climate formation mechanisms and optimizing urban built environments—the research failed to incorporate computational fluid dynamics (CFD)-based approaches through ENVIMET modeling for cooling strategy development [83,84]. Furthermore, this research utilized an extensive dataset comprising Landsat imagery, OSM datasets, Google Maps, ground station observations, DSM, and socioeconomic information. Regional disparities in data accessibility and resource distribution make it challenging to adopt uniform parameters or metrics for evaluating localized thermal conditions, thereby constraining the broader applicability of our methodology.

Within the construction of real-world 3D urban models in China, we will achieve fine classification of surface cover, extract building geometry and urban morphological parameters, construct a refined thermal environment model, diagnose the causes of UHIs, and build a visualization platform, based on a high-precision 3D database. This will form the focus of our future work.