Spatiotemporal Analysis of Surface Urban Heat Island Dynamics in Central Yunnan City Cluster

Abstract

The acceleration of urbanization has led to an increase in urban expansion and population density, exacerbating the urban heat island (UHI) effect. Moreover, the phenomenon has a significant impact on urban ecological environments and human health. Consequently, mitigating the UHI effect and enhancing the ecological environment is crucial. However previous research has primarily focused on individual cities or regional scales, with few studies analyzing all cities within urban agglomerations. This paper conducts a fine-grained spatiotemporal analysis of surface urban heat island (SUHI) effects in the Central Yunnan City Cluster from 2000 to 2021 using Landsat satellite data. We calculate the surface urban heat island intensity (SUHII) for 44 cities at the county or district level and discuss the quantitative estimation of overall SUHII changes and driving factors in the Central Yunnan City Cluster. Our findings are as follows: 1. Small cities also exhibit UHI effects, with a 75.4% probability of occurrence in the Central Yunnan City Cluster from 2000 to 2021, resulting in an overall decrease in SUHII of 1.21 °C. 2. The temperature increase rate in urban extension areas and suburban areas is faster than that in urban central areas, which is the main reason for the decreasing trend of SUHII. 3. Land use change inhibits the weakening of the SUHI effect, and population change contributes to the formation of this phenomenon. Additionally, the methods and results of this study can provide reasonable and effective insights for the future development and planning of the Central Yunnan City Cluster, thus promoting urban sustainable development.

1. Introduction

The urban heat island effect presents a global challenge, affecting the majority of urban areas worldwide [1]. This phenomenon occurs when urban regions experience higher temperatures compared to their surrounding rural areas. Its primary causes include urban infrastructure such as buildings, roads, transportation, and industrial activities, which intensify the absorption and retention of heat within cities, thereby generating the urban heat island effect [2]. It significantly impacts the exchange of energy between land and atmosphere as well as water circulation processes. Its adverse effects on urban residents include increased fatigue, disrupted sleep patterns, and heightened air pollution levels [3]. Furthermore, the urban heat island effect detrimentally affects urban ecosystems by accelerating land degradation and diminishing biodiversity [4].

Currently, research on urban heat islands predominantly employs two methodologies: analysis based on meteorological station data and analysis based on remote sensing data. The latter often utilizes datasets such as MODIS and Landsat satellite data. Conventional studies of urban heat islands primarily rely on temperature data collected from ground stations to characterize and analyze local urban heat environments, termed urban canopy layer heat island (CUHI) research [5]. For example, W. Liu et al. investigated the intensity of the heat island and its correlation with meteorological factors in Beijing from 1977 to 2000 using meteorological station data, revealing a significant relationship between urban heat island intensity and meteorological factors [6]. However, due to the sparse and uneven distribution of meteorological stations, these ground-based observations inadequately capture the spatial variations in urban heat environments [7]. Therefore, further investigations are warranted, utilizing a combination of technologies such as high-density sensor networks, remote sensing techniques, and high-resolution numerical models [8]. Advancements in remote sensing technology have enabled researchers to utilize satellite data to comprehensively monitor urban surface temperatures, facilitating the depiction of spatiotemporal variations in urban heat island effects. Consequently, remote sensing is widely employed to characterize and monitor spatiotemporal changes in land surface temperature (LST). To date, numerous scholars worldwide have conducted detailed studies employing remote sensing techniques, focusing on aspects such as algorithms for retrieving surface temperatures, spatial–temporal distribution characteristics, driving factors, and energy structures [9,10,11].

Urban agglomerations, comprising interconnected cities and their surrounding regions, face heightened probabilities of heat island effects amid accelerated urbanization [12]. In previous research, numerous scholars have investigated urban heat islands at the scale of metropolitan areas. For instance, Wang P et al. examined the long-term variations in both average and extreme temperatures perceived by humans across 20 major metropolitan areas in mainland China since the 1970s [13]. Hongchao Xu et al. utilized MODIS land surface temperature (LST) data to analyze the spatiotemporal patterns and characteristics of daytime regional urban heat islands (RHIs) in the Beijing–Tianjin–Hebei metropolitan region [14]. Meanwhile, Chen M et al. explored the spatiotemporal evolution characteristics of urban heat island effects in the Beijing–Tianjin–Hebei region using MODIS data. The findings indicate a continuous increase in the intensity of urban heat islands in this area from 2003 to 2013, with Beijing and Tianjin exhibiting the most pronounced heat island intensity [15]. However, these studies tend to focus on single cities or specific regions, and there are relatively few comprehensive analyses of the heat island effect of all cities within urban agglomerations. This may lead to the limitation of the understanding of the overall heat island effect of urban agglomerations, which affects the formulation and implementation of urban thermal environment improvement strategies. Therefore, in order to understand the heat island effect within urban agglomerations more comprehensively and develop effective mitigation measures, it is necessary to conduct comprehensive analysis and research on the heat island effect of all cities within urban agglomerations. This will be of great significance for improving the urban ecological environment, enhancing the quality of life of residents, and promoting sustainable development.

The Central Yunnan City Cluster is one of the eight proposed urban agglomerations in China, as outlined in the National New Urbanization Plan (2014–2020) and the 13th Five-Year Plan, and is experiencing rapid urbanization, resulting in ecological challenges [16]. To comprehensively analyze the urban heat island effect within the Central Yunnan City Cluster, this study focuses on small cities within the agglomeration. We utilized Landsat remote sensing images to determine surface urban heat island intensity, and its temporal changes from 2000 to 2021 were calculated for all cities within the Central Yunnan City Cluster. Additionally, quantitative estimation and discussion of the spatiotemporal variation of the surface urban heat island intensity index (SUHII) across the entire urban agglomeration were conducted to elucidate driving factors. The findings of this study offer valuable insights for the future development and planning of the Central Yunnan City Cluster, facilitating urban sustainable development initiatives.

2. Materials and Methods

2.1. Study Area

The Central Yunnan City Cluster encompasses 49 counties, cities, and districts in Kunming, Qujing, Yuxi, Chuxiong, and Honghe Prefectures, positioned roughly between 101.5° and 103.5° east longitude and between 24° and 27.5° north latitude. Renowned as the most economically vibrant region in Yunnan Province and a significant hub for economy, culture, and transportation in Southwest China, it benefits from abundant natural resources and a strategically advantageous geographical location [17]. With a predominantly subtropical climate marked by distinct seasons, the region experiences warm summers and relatively cold winters, fostering diverse ecosystems comprising subtropical evergreen forests, coniferous forests, deciduous forests, grasslands, and wetlands [18].

The rapid urbanization of the Central Yunnan City Cluster has brought about an expansion in urban areas, population influx, and rapid economic growth. However, this urban transformation has also engendered challenges such as urban heat island effects, declining air quality, water resource pressures, and ecological degradation. To foster sustainable development, efforts are needed to enhance urban planning, environmental conservation, and resource management efficiency, thereby alleviating the adverse impacts of heat-related environmental issues on cities and inhabitants [19].

In this study, considering location factors, we combine the six districts of Wuhua, Panlong, Guandu, Xishan, Chenggong, and Jinning District into the main urban area of Kunming City. Consequently, 44 cities within the Central Yunnan City Cluster are considered for analysis. Notably, in 2019, apart from Kunming’s core urban area, impervious surface areas of the remaining 43 cities did not exceed 50 square kilometers. Hence, excluding Kunming’s core urban area, these cities are termed “small cities” within the study, as illustrated in Figure 1.

2.2. Data Sources

This study employed four datasets (

Table 1. Dataset introduction.

Dataset	Type	Time
LANDSAT/LT05/C01/T1_SR	Landsat 5	1984–2013
LANDSAT/LC08/C01/T1_SR	Landsat 8	2013–2021
GISA 2.0	Impervious surface	1972–2019
ASTER GDEM V3	DEM	2019

). Given the extensive scope and temporal considerations, the Google Earth Engine (GEE) platform was utilized for land surface temperature calculations. For the period spanning 2000 to 2011, land surface temperatures were derived from the LANDSAT/LT05/C01/T1_SR dataset, sourced from the Landsat 5 ETM sensor’s atmospherically corrected surface reflectance data. These images encompassed four visible and near-infrared (VNIR) bands and two shortwave infrared (SWIR) bands, processed into orthorectified surface reflectance. Additionally, a thermal infrared (TIR) band was transformed into orthorectified brightness temperature data. While the VNIR and SWIR bands maintain a 30 m resolution, the TIR band, initially at 120 m/pixel (60 m/pixel for Landsat 7), was resampled to 30 m using cubic convolution. Subsequently, for the years 2013 to 2021, land surface temperature computations relied on the LANDSAT/LC08/C01/T1_SR dataset, comprising atmospherically corrected surface reflectance data from Landsat 8 OLI/TIRS sensor. These images consisted of five VNIR bands, two SWIR bands, and two TIR bands, processed similarly to the previous dataset. All data underwent atmospheric correction and included masks generated using CFMASK, accounting for cloud, shadow, water, and snow, as well as saturation masks for individual pixels. Normalized Difference Vegetation Index (NDVI) and Normalized Difference Built-up Index (NDBI) were calculated via GEE using Landsat 5 images from June to August, employing the maximum value method for annual synthesis. Impervious surface data were derived from the Global Impervious Surface Analysis (GISA) 2.0 dataset (1972–2019), released by the Remote Sensing Information Engineering Institute at Wuhan University, led by Professor Huang Xin. With an F1-score of 0.935, GISA 2.0 exhibits superior accuracy compared to GISA 1.0 (0.893), GAIA (0.721), and GAUD (0.809) [20]. Digital Elevation Model (DEM) data were sourced from the ASTER GDEM V3 dataset available on the NASA website. Additionally, to assess the impact of human activities on surface urban heat island intensity (SUHII) variations, socioeconomic data from various state statistical bureaus were incorporated.

2.3. Defining Urban and Rural Regions

Different methods for delineating urban and suburban areas significantly influence the results of calculations of the SUHI effect. Some commonly used methods in urban heat island research include the buffer zone method, which defines suburban areas within a certain range surrounding a built-up area [21], and the land use classification method, which uses supervised classification based on factors such as farmland and NDVI to distinguish urban and suburban areas [22]. Recent studies have used a physics-based clustering approach, such as “Using Clustering to Understand Intra-city Warming in Heatwaves: Insights into Paris, Montreal, and Zurich” [23]. Variability in these methods introduces considerable uncertainty in SUHII estimation [24]. Given the emphasis of this study on temporal trends, the impact of the urban–suburban division method on the results is relatively minor. The increase in impervious surfaces alters surface emissivity, roughness, and albedo, amplifying surface sensible heat flux and disrupting the regional energy balance, thereby exacerbating the SUHI effect [25]. This paper delineates urban and suburban areas for each year based on annual impervious surface data. Figure 2 illustrates the delineation of urban and suburban areas for Kunming City in 2019 using equal-area, double-area, and triple-area buffer zones as suburban areas. Considering the significant influence of water bodies on temperature [26], particularly in cities like Kunming surrounded by lakes, the calculation process excludes the area of Dian Lake (in Yunnan Province) to mitigate potential deviations in SUHI estimation.

2.4. Retrieval LST

To mitigate the influence of atmospheric factors, this study employed the GEE platform to compute LST from 2000 to 2012 using Landsat 5 and from 2013 to 2021 using Landsat 8. Cloud cover was constrained to below 50%. Given the pronounced urban heat island effect during summer [27], remote sensing images for LST calculation were specifically selected from June to August. Within the Dian Lake region, which encompasses 44 cities, remote sensing images were retrieved based on the urban–rural boundaries defined by each city. For each city, all images captured within the defined urban–rural boundary during the summer of a particular year were considered. The image or images with the highest average LST within this period were chosen to synthesize the LST for that city. LST is represented by temperature corrected for emissivity. Using the thermal infrared band of Landsat data, brightness temperature (Lλ) is calculated and corrected for emissivity ε. The calculation process proceeds as follows:

Conversion of Landsat satellite digital numbers to land surface radiance:

$${\mathrm{L}}_6=\mathrm{o}\mathrm{f}\mathrm{f}\mathrm{s}\mathrm{e}\mathrm{t}\times \mathrm{D}\mathrm{N}+\mathrm{g}\mathrm{a}\mathrm{i}\mathrm{n}$$

(1)

“Offset” and “gain” refer to the gain and bias values of the thermal infrared band, respectively.

Compute the radiance corresponding to the land surface brightness temperature:

$$\mathrm{T}=\frac{K_2}{\ln \left(\frac{K_1}{L_6}+1\right)}$$

(2)

where $K_1$ and $K_2$ are preset constants for emissivity correction.

Calculate the surface vegetation coverage (NDVI is calculated from contemporaneous remote sensing images):

$$F_V={\left[\frac{NDVI-NDV{I}_{\min }}{NDV{I}_{\max }-NDV{I}_{\min }}\right]}^2$$

(3)

Compute the surface emissivity:

$$\epsilon =0.004\times F_V+0.986$$

(4)

Calculate the land surface temperature after emissivity correction:

$$LST=\frac{T}{1+\left(\lambda \times T/\rho \right) \mathrm{l}\mathrm{n}\epsilon }$$

(5)

where $\lambda$ is the central wavelength of the thermal infrared band and $\rho$ is a constant with a value of 1.438 × 10⁻².

2.5. Calculation of SUHII and Fitting of Its Variation Trend

Due to the instantaneous nature of LST derived from Landsat satellite data and the presence of data gaps, the UHI intensity is defined as the temperature contrast between urban and suburban areas. Hence, as long as the delineated urban and suburban areas are encompassed within the same scene image, the SUHII of the city can be computed. However, due to practical constraints, certain cities may not meet this criterion, leading to two scenarios: first, when the division line lies at the periphery of the studied city, only the portion of the city exhibiting markedly smaller differences needs to be excluded; second, when the division line bisects the city, the city is to be clipped according to this line, with SUHII calculated separately for each side and then averaged to determine the annual SUHI. Given the pronounced seasonal variability of urban heat island (UHI) effects, we solely employ LST data for a single season (June, July, and August) to estimate the SUHII annually. The formula for SUHII calculation is as follows:

$$SUHII={LST}_{urban}-{LST}_{rural}$$

(6)

where SUHII stands for surface urban heat island intensity. ${LST}_{urban}$ and ${LST}_{rural}$ represent the average land surface temperatures of the urban and rural areas, respectively. As the urban limits vary from year to year, the urban and suburban limits are defined for each year based on impervious water data. The average temperature within the built-up area is taken to represent urban temperature, while the average temperature within the suburban area is considered rural temperature. The SUHII is calculated by subtracting the rural temperature from the urban temperature.

For the overall estimation of a large area’s time series, a common approach involves characterizing it by calculating the average value at each time point. However, in this study, due to the use of instantaneous temperature to compute SUHII, the likelihood of data variation caused by random effects is high. The simple averaging method cannot effectively represent the SUHII variations across the entire urban cluster. Therefore, this study characterizes the SUHII trend of the entire central Yunnan urban cluster by fitting the variations in SUHII for all cities. Initially, using the first city as a starting point, the SUHII variation ranges of the remaining 43 cities are normalized to the SUHII variation range of the first city. All cities’ SUHII trend lines are connected, resulting in an extended dataset with X and Y values expanded 44 times. By optimizing a linear model for this extended dataset, a linear expression is obtained. Finally, by further optimizing the linear function, the optimal model is derived to calculate the overall trend of SUHII variation for the central Yunnan urban cluster. The conceptual diagram of this transformation process is shown in Figure 3. Although this transformation alters the specific values of the curve, it preserves the curve’s trend. Additionally, by cumulatively summing the curve’s trend changes, an overall trend for all curves can be obtained.

2.6. Calculation of the Variation Trend of SUHII

Sen + Mann–Kendall trend analysis is a widely employed method in time series analysis for detecting significant trend changes in a dataset. Firstly, the Sen slope estimation method is applied to compute the slope values of SUHII for each city. Subsequently, the Mann–Kendall non-parametric test is conducted to assess the statistical significance of trend changes in the slope sequence. The Sen slope estimation method calculates the median slope between any two data points to derive the overall slope of the dataset. Meanwhile, the Mann–Kendall test evaluates the monotonicity of the slope sequence, indicating whether it exhibits an increasing or decreasing trend. Significant monotonic changes in the slope sequence suggest notable trend variations in the dataset. Moreover, the Mann–Kendall test can determine the significance level of the trend changes, providing insight into their statistical significance. This analysis assesses the trend in SUHII for each city from 2000 to 2021.

2.7. Driver Analysis

The spatial autoregressive (SAR) model is a statistical tool used to analyze spatial data by considering the interdependence among neighboring geographical locations. This model is particularly useful in handling datasets with spatial dependence, where observations at one location are influenced by nearby observations. At its core, the SAR model incorporates a spatial weight matrix to capture these relationships, acknowledging that each location’s characteristics are not only influenced by its attributes but also by those of neighboring locations.

The SAR model is expressed as follows:

$$\mathrm{y}=\rho W_1\mathrm{y}+\beta \mathrm{x}+u,u=\lambda W_2\mu +\epsilon$$

(7)

In this equation, y represents the dependent variable (observation at each location), x is the explanatory variable, β denotes the spatial regression coefficients, u is the error term, μ represents white noise, W₁ captures the spatial trend of the dependent variable, and W₂ reflects the spatial trend of the residuals. The parameters ρ and λ signify the spatial lag and spatial error coefficients, respectively.

The SAR model is interpreted as follows: When ρ and λ are both zero, the model reduces to a simple linear regression without spatial autocorrelation. Nonzero values of ρ and β indicate a spatial lag model, while nonzero ρ and λ suggest a spatial error model.

In the study, 12 indicators were utilized to investigate SUHII variations (

Table 2. Driver selection factors.

	Factor	Variable Calculation	Factor Calculation	Source	Denotative Meaning
variable	ΔNDBI	Sen slope estimation	NDBIurban-NDBIrural	Landsat dataset	Urbanization trend
	ΔNDVI	Sen slope estimation	NDVIurban-NDVIrural	Landsat dataset	The difference between vegetation cover change in urban areas and rural areas
	Built-up area	Sen slope estimation	-	GISA 2.0 (1972~2019)	Urbanization process and land use change
	population	Sen slope estimation	counting by county	statistical yearbook	Demographic trends and demographic changes
	GDP	Sen slope estimation	counting by county	statistical yearbook	Changes in the level of economic growth and economic activity
	Primary_industry	Sen slope estimation	counting by county	statistical yearbook	Changes in agriculture and related industries
	Secondary_industry	Sen slope estimation	counting by county	statistical yearbook	The development and change of manufacturing and related industries
	Tertiary_industry	Sen slope estimation	counting by county	statistical yearbook	The development and change of the service industry and related industries
quantification	ΔDEM	-	DEMurban-DEMrural	ASTER GDEM V3	The elevation difference between urban and suburban areas
	Δslope	-	slopeurban-sloperural	ASTER GDEM V3	Urban and suburban gradient differences
	Longitude	-	-	-	Longitude
	Latitude	-	-	-	Latitude

). The Theil–Sen slope estimation method was applied to compute changing slopes for various factors, including ΔNDBI, ΔNDVI, built-up area, population, GDP, and industry sectors. Socioeconomic data were sourced from statistical yearbooks, while topographic factors like ΔDEM, Δslope, longitude, and latitude were considered.

3. Results

3.1. Trends in SUHII Variation

During the period from 2000 to 2021, the majority of cities within the Central Yunnan City Cluster experienced the urban heat island effect, with SUHII values ranging from −4.34 °C to 4.9 °C. Notably, there was a high probability, reaching 75.4%, of cities encountering the urban heat island effect (Figure 4). The distribution of SUHII reveals that the most frequent occurrences were within the range of 0.4 °C to 1.2 °C, with a peak between 0.8 °C and 1.2 °C. This observation underscores the widespread nature of the urban heat island effect during the specified timeframe.

As depicted in Figure 5, among the 44 cities comprising the Central Yunnan City Cluster, the SUHII trend is declining in 39 cities. This suggests a deceleration in the urban heat island effect across these cities. While some cities experienced notable temperature variations in certain years, the overall downward trend in SUHII remains unaffected. This observation underscores the resilience of the trend toward mitigating the urban heat island effect in the region.

3.2. Spatial Distribution of SUHII Change Rates

As depicted in Figure 6, the calculation of SUHII change rates for all cities within the Central Yunnan City Cluster reveals a notable trend: the majority of cities exhibit a significant decrease in SUHII change rates, indicative of a slowdown in the urban heat island effect. Importantly, this trend shows a relatively uniform spatial distribution, without evidence of spatial clustering. However, a handful of cities, including Kunming, Huize, Hekou, Fuyuan, and Malong, defy this trend by demonstrating an increasing SUHII change trend. Among these outliers, the Malong District of Qujing stands out with a significant upward trend in SUHII, potentially driven by factors such as accelerated urbanization, increased building density, and decreased green coverage. It is imperative for policymakers and urban planners to address the urban heat island effect in these cities with targeted measures aimed at mitigating its impacts and fostering sustainable urban development.

3.3. SUHII Change Rates in the Central Yunnan Urban Cluster

As depicted in Figure 7a, the direct linear fitting of SUHII trends for all cities from 2000 to 2021 yields unsatisfactory results, with an R² of only 0.1. The trend line fails to adequately explain the variation observed in all data points. Conversely, Figure 7b demonstrates that fitting the trend changes of SUHII for all cities leads to significant improvement, with an R² of 0.97 and a root mean square error of only 5.3 (

Table 3. Model accuracy evaluation.

Model	Direct Fitting	Transformation Fitting
Equation	Y = A + B × X	Y = A + B × X
A	144.25228 ± 13.05655	4.12511 ± 0.14864
B	−0.07128 ± 0.00649	−0.05781 ± 2.65766 × 10−4
Reduced Chi-Sqr	1587.31481	5.33879
R-squared	0.11089	0.94999
Adjusted R-squared	0.10997	0.94997

). This indicates that the obtained linear equation effectively captures the overall variation in SUHII for the Central Yunnan City Cluster. By inputting time into this equation, it is calculated that the SUHII for the Central Yunnan City Cluster decreased by 1.21 °C from 2000 to 2021. This finding underscores the importance of considering trend changes in SUHII analysis, which can provide more accurate insights for urban planning and policy formulation.

3.4. Results of Spatial Autoregressive Analysis

Spatial autocorrelation analysis was performed on the change rates of SUHII, area, NDBI, and NDVI variables. The results indicate that the Moran’s Index for all four variables is greater than 0, suggesting spatial autocorrelation. Notably, both SUHII and NDVI exhibit a p-value less than 0.05, indicating significant spatial autocorrelation (

Table 4. Moran’s I values for the slope of SUHII, area, DBVI, and NDVI.

Model	Direct Fitting	Transformation Fitting
	Moran’s I Index	Variance
SUHII	0.146	0.0039
Built-up area	0.032	0.0037
ΔNDBI	0.089	0.004
ΔNDVI	0.214	0.004

). This suggests that the spatial patterns of SUHII and NDVI are not randomly distributed but exhibit clustering or spatial dependency. The presence of significant spatial autocorrelation has implications for spatial modeling and analysis, indicating that neighboring areas tend to have similar values for these variables. Understanding spatial autocorrelation is crucial for effective spatial analysis and decision-making in urban planning and environmental management.

Due to the presence of spatial autocorrelation in the study variables, traditional OLS linear models are inadequate in explaining the relationship between independent and dependent variables. However, after the incorporation of spatial weights, it was observed that models fitted with spatial error or spatial lag terms outperform traditional OLS models in terms of accuracy. Upon comparison, the SEM emerged as the most accurate, with independent variables explaining nearly 60% of the variance in SUHII.

Table 5. Table of regression coefficients of independent variables.

Factors Type	Variable	Coefficient	Std.Error	Z-Value	Probability
Geographical	Longitude	4.25 × 10−8	1.74 × 10−8	2.43726	0.0148
	Latitude	−2.28 × 10−8	1.40 × 10−8	−1.63171	0.10274
Topographical	ΔDEM	−7.90 × 10−5	6.96 × 10−5	−1.13563	0.25611
	ΔSLOPE	−0.00038	0.0012	−0.31809	0.75042
Land use change	Built-up area	−0.08692	0.15682	−0.55428	0.57939
	ΔNDBI	−0.0745	0.1107	−0.673	0.50095
	ΔNDVI	−0.22362	0.08164	−2.73921	0.00616
Socioeconomic	population	0.06709	0.0916	0.73239	0.46393
	GDP	−0.53477	0.36294	−1.47346	0.14063
Industrial Outcome	Primary_industry	−0.45736	0.45437	−1.00658	0.31414
	Secondary_industry	−0.04027	0.22382	−0.17993	0.8572
	Tertiary_industry	0.43398	0.21262	2.04112	0.04124
	CONSTANT	−0.40656	0.20548	−1.97859	0.04786
	λ	−0.41772	0.24928	−1.67572	0.09379
	R-squared	OLS		0.55238
		SLM		0.56401
		SEM		0.58683

presents the regression coefficients for the spatial error model, revealing significant impacts of changes in GDP, primary industry, and tertiary industry on SUHII. Specifically, the coefficients for GDP and primary industry are negative, indicating a decrease in SUHII with increasing values of these variables, while the coefficient for tertiary industry is positive, suggesting an increase in SUHII. Moreover, the impact of the tertiary industry on SUHII is statistically significant at the p < 0.05 level. Additionally, ΔNDVI demonstrates a significant negative impact on SUHII changes, while the regression coefficients for other factors are relatively small. Overall, the spatial error model provides valuable insights into the factors influencing SUHII dynamics, aiding in effective urban planning and management strategies.

Following the averaging of regression coefficients for factors categorized into geographical factors, topographical factors, socioeconomic factors, land use change factors, and industrial outcome factors, it was observed that only geographical factors exhibit a positive impact on SUHII changes, albeit with a small coefficient. Conversely, topographical factors, land use change factors, socioeconomic factors, and industrial outcome factors demonstrate negative impacts on SUHII changes (Figure 8). This suggests that these factors also contribute to the mitigation of SUHII.

3.5. Exploratory Analysis

During our investigation, an intriguing observation arose: a significant decline in SUHII was noted across most cities. To elucidate this phenomenon, we systematically selected four cities meeting specific criteria to examine their temperature trends (Figure 9). Our findings demonstrate a consistent temperature increase across nearly all regions of these cities from 2000 to 2021. Notably, the urban expansion zones, delineated between the built-up areas in 2000 and 2019, exhibited the most rapid temperature rise, followed by suburban areas in 2019, while the temperature rise was slowest in the built-up areas of 2000, located at the city center.

The magnitude of urban areas significantly influences SUHII variations. In Figure 10a, 44 cities are grouped into 10 categories based on the size of their built-up areas, with decreasing areas from left to right. The results indicate the presence of urban heat island effects across all built-up area categories. Figure 10b depicts SUHII variations in the Central Yunnan City Cluster under various suburban range divisions, with reconstructed SUHII values plotted over time. The findings reveal a more pronounced decline in SUHII with larger suburban range definitions during SUHII calculation.

4. Discussion

4.1. SUHII Change Characteristics

Throughout the developmental trajectory of the Central Yunnan City Cluster from 2000 to 2021, the pace of built-up area expansion in smaller cities, excluding Kunming City’s main urban area, was relatively modest, typically increasing by 3 to 5 times. Despite the prominence of UHI phenomena in larger cities, as evidenced by studies like Bin Zhou et al.’s examination of European cities, where UHI intensity escalates with urban scale [28], our findings align. Kyungil Lee et al.’s analysis of UHI intensity in eight major Asian cities further corroborates this trend, indicating a proportional increase in urban areas [29]. Similar patterns were observed in small cities, such as Aveiro in Portugal, where distinct UHI characteristics were documented [30]. Danijel Ivajnšič et al. studied UHI patterns in the small city of Ljutomer and found that built-up areas in Ljutomer were, on average, 1 °C warmer in winter than rural areas [31]. Our study affirms the presence of urban heat island effects in both Kunming City and its surrounding small cities.

Many studies have indicated an exacerbation of the urban heat island effect. For instance, Peng Tian et al. utilized Landsat thermal sensor imagery and established an urban–rural index (URI) to reveal the spatiotemporal characteristics of surface urban heat island effects, finding an intensification of urban heat island effects in city centers [32]. Shirao Liu et al. studied the spatiotemporal trends of SUHII in 201 Chinese prefecture-level cities, revealing overall annual and seasonal increases in SUHII trends across Chinese cities [33]. However, some studies have observed a weakening of urban heat island effects in certain cities. Xiaoxue Peng et al., based on MODIS data from 2001 to 2019, analyzed the spatiotemporal patterns of surface urban heat islands in 180 Chinese cities and identified a declining trend in SUHII for some cities [34]. In a comparison with other scholars’ studies, it was found that apart from differences in urban factors such as topography, climate, and urban morphology, different definitions of SUHI (or different methods for defining suburban areas) can also lead to differences in trend formation [35].

Urban expansion exhibits diversity influenced by various factors such as topography, climate, altitude, and urban morphology, resulting in different spatiotemporal patterns of urban heat island effects [36,37,38]. In this study, after calculating annual temperatures for the cities of Mile, Qilin, Shizong, and Fuyuan and conducting statistical analysis, it was found that the city center was not the fastest-growing temperature area. Min Min’s study on the spatiotemporal changes of urban heat islands in Zhengzhou also found a shift in heat island distribution from the city center to the suburbs [39]. Xu Wang et al., after exploring the dynamic characteristics of physiological equivalent temperature (PET) gradients in different urban–rural gradients (urban core area, urban expansion area, and suburbs) in 101 Chinese cities, found that the severity of PET aggravation in suburbs was more severe than that in urban centers [40]. Yanfei Wu et al., through a study on the evolution mechanism of SUHII in Hangzhou from 2000 to 2020, found that changes in SUHII were mainly due to urban expansion in the suburbs, while the SUHI intensity in the city center remained stable [41].

This paper defined urban and suburban areas using the buffer zone method, calculated SUHII for 44 case studies in the Central Yunnan City Cluster over 22 years, and found a declining trend in SUHII for most cities through trend analysis. Furthermore, through temperature calculations in four randomly selected cities, it was found that the temperature growth rate in urban core areas was relatively slow, followed by the urban expansion areas, with suburbs having the fastest temperature growth rate. Urbanization has led to an increase in LST [42], coupled with the definition of SUHI methods in this paper, resulting in the anomalous phenomenon of declining SUHI in multiple cities. The urbanization process can lead to changes in land use and land cover, such as increased building density and reduced green spaces in urbanized areas, which can affect the thermal characteristics of cities. Additionally, the implementation of urban planning and greening projects may reduce the urban heat island effect. Furthermore, the impact of climate change and atmospheric conditions may also affect SUHII. For instance, warming climates may lead to a reduction in temperature differences between urban and suburban areas, thereby decreasing SUHII.

4.2. Comparison of Estimation Methods of Different SUHII Trends

This research utilizes annual summer remote sensing imagery to derive the current year’s instantaneous temperature, resulting in a SUHII that lacks seasonal and diurnal variations. The interannual SUHII variation is assumed to be linear. In analyzing the trend of SUHII time series data, the least squares method is initially employed for linear fitting, revealing substantial variability in calculated SUHII (Figure 11b). However, the least squares method, which employs the sum of squared residuals as the loss function, is highly influenced by outliers during fitting. Huber regression, distinguishing between squared and absolute loss based on residual thresholds, attempts to mitigate this issue but proves unsuitable for large-sample time series data fitting, as the chosen thresholds fail to accommodate SUHII trend fitting for all cities (Figure 11a). Contrastingly, Theil–Sen regression, estimating trends by computing medians, demonstrates consistent fitting of SUHII change trends across all cities (Figure 11c,d). Theil–Sen slope estimation is widely adopted in urban heat island studies, including those by Qiquan Yang [43], Asfa Siddiqui [44], and Zongyang Wang [45], for quantifying SUHII trend changes.

4.3. Factors Influencing SUHI Variations

This study employs spatial autoregressive methods to assess the impact of five distinct factors—geographical, topographical, socioeconomic, land use change, and industrial outcome—on SUHI trends. Notably, only geographical factors exhibit a positive influence on SUHI trends, underscoring the significant role of geographical location in SUHI variations. Research by Minjun Kim et al. in Ulsan City reveals varying UHI intensities across different geographical locations, with urban center roads experiencing reduced UHI intensity in winter and summer [46]. Population dynamics and the tertiary industry also contribute to the formation of a downward SUHI trend. Specifically, 16 out of 44 cities in the study witnessed a population decline, except for Kunming’s main urban area and cities with decreasing populations, suggesting minimal population changes in other urban areas. Similarly, an analysis by Xiaoxue Peng et al. demonstrates a downward trend in SUHI changes across 180 shrinking cities in China [34]. The impact of primary and secondary industrial structures on SUHI trends appears negligible, indicative of the intricate interplay between urbanization, industrial structural changes, and the urban heat island effect [47]. In our study, the observed decline in the overall urban heat island effect is driven by suburban temperature growth outpacing that of urban centers. Negative driving factors, notably ΔNDVI and the primary industry, significantly suppress temperature variations, aligning with existing research findings [48,49]. Vegetation, through transpiration and shading effects, mitigates surface temperatures, thereby alleviating the urban heat island effect [50]. Notably, the influence of ΔDEM and ΔSLOPE on SUHI change rates is found to be insignificant.

Moreover, this study extends its investigation to analyze temperature variations in four cities exhibiting SUHI decreases, categorized into three regions: the built-up area in 2000, the urban expansion area from 2000 to 2019, and the suburban area in 2019. The findings reveal a faster temperature growth rate in the urban expansion area compared to other regions, while the built-up area in 2000, serving as the city center, exhibits the slowest temperature growth rate. This observation highlights the significant influence of urban development on the urban heat island effect, indicating that the observed decreasing trend in SUHII is driven by accelerated temperature growth rates in suburban areas compared to urban regions. Urban expansion exacerbates the urban heat island effect, yet certain factors such as increased urban greenery and water bodies can mitigate this effect [51]. It is important to note that while the majority of cities exhibit a negative trend in SUHI changes, this does not imply a reduction in the adverse effects of urban-development-induced surface temperature increase on the ecological environment and human life. Factors such as urban center saturation, prompting expansion into suburban areas, contribute to increased suburban temperatures and reduced temperature differences between urban and suburban areas. Additionally, the development of heavy industries in suburbs may further diminish temperature differences between urban and suburban areas, consequently influencing the observed SUHII change trend. However, further scientific validation and explanation are required to substantiate these hypotheses.

Even in small cities, population activities and energy consumption contribute to heat release, affecting local temperatures. Activities such as population aggregation and traffic flow generate heat, particularly in commercial areas and transportation hubs [52]. While the urban heat island effect in small cities may not be as pronounced as in larger cities, it remains a significant concern. Rational urban planning, construction of green infrastructure, management of population activities, and implementation of climate adaptation measures are crucial for alleviating the heat island effect in small cities. As cities evolve, small cities can enhance their urban environments’ comfort and sustainability through scientific management and planning.

4.4. Verification of Urban Pixels

We utilized nighttime light data products (NPP/VIIRS) and overlaid them with remote sensing imagery from the TianDiTu platform to visually determine thresholds. This process enabled the identification of urban pixels. Subsequently, we created buffer zones twice the size of the urban areas to delineate suburban boundaries. Using these redefined urban and suburban boundaries, we calculated the SUHII for cities within the Dian Central Urban Cluster in 2019 (due to data precision issues, only 35 cities could have their urban boundaries delineated) and validated them against the previously computed SUHII (Figure 12). The root mean square error (RMSE) was 0.78 degrees Celsius, and R² was 0.86. The accuracy of the research results meets the requirements.

4.5. Limitations and Future Prospects

Conducting large-scale spatiotemporal studies on UHI using Landsat imagery presents challenges, primarily due to the need for multiple image acquisitions to cover the extent of a city cluster and the varying transit times of satellite images. These challenges make it difficult to assess surface temperatures for a city cluster in a given year. However, SUHII, representing the temperature difference between urban and rural areas, can be quantified if both urban and suburban areas of a study city are within the scope of a single image. Techniques can also be employed for cases where two image acquisitions are needed to cover a city. It is crucial to ensure that LST data from the same season are used for SUHII calculation due to its seasonal variations. Despite significant differences in calculated LST among cities for a particular year, the difference in urban–rural temperature (SUHI) is minimal. Instantaneous temperature is used to characterize the temperature of a given year, despite significant data gaps, with SUHII calculated for as many years as possible to mitigate this error. However, verifying the accuracy of computed SUHII results is challenging due to the sparse distribution of meteorological stations and the low spatial resolution of MODIS LST products, necessitating utmost rigor in the calculation process. Moreover, limitations exist in the analysis of driving factors, prompting exploration into whether fitting SUHII driving factor models using non-parametric models would yield higher accuracy.

Small cities also face the challenge of the urban heat island effect. Although their scale may be smaller, the impact is not negligible. Governments and communities should recognize this and take appropriate measures in response. Our research results indicate that the temperature increase rate is highest in the urban expansion areas, generating necessary heat during urban development. Priority should be given to implementing heat mitigation measures in these regions. This may involve increasing green infrastructure, utilizing cool-colored building materials, and improving the preservation and utilization of water bodies and wetlands [53]. In suburban or urban fringe areas, attention should be paid to land use conversion issues, preserving natural green spaces and water bodies, and encouraging low-carbon transportation modes. While the temperature increase rate in the city center is relatively slower, it still requires moderate management to prevent further exacerbation of the heat island effect. For instance, this can be achieved by improving urban planning and design, increasing green spaces, employing thermally efficient buildings and efficient indoor cooling, reducing building heat absorption, and promoting sustainable transportation and energy utilization. Solutions incorporating multiple mitigation measures are typically more effective in alleviating heat than individual measures [54].

5. Conclusions

The urban heat island effect, as a global issue, persists within the rapid urbanization process, with even small cities being unable to escape its impact. Landsat satellite data meet the research requirements for small cities. In this study, using GEE, we inverted the land surface temperature of each county-level city within the Central Yunnan City Cluster from 2000 to 2021. We calculated the SUHII and its variation rate for 44 cities in the study area, analyzed the spatiotemporal evolution patterns of SUHII in small cities, and quantitatively estimated the overall SUHI trend in the Central Yunnan City Cluster, while exploring its driving factors. The following conclusions were drawn: Small cities also exhibit the urban heat island effect. Previous studies have predominantly focused on the UHI effect in large cities, seemingly overlooking whether the UHI effect exists in small cities. We found that from 2000 to 2021, the probability of the UHI effect occurring in the Central Yunnan City Cluster was 75.4%. Interestingly, most cities showed a decreasing trend in SUHII, with the overall SUHII in the Central Yunnan City Cluster decreasing by 1.21 °C. The study found that as the defined suburban area increases, the intensity of the urban heat island decreases. Transforming and fitting the SUHII change curves of all cities within the urban agglomeration improved the fitting goodness of the overall SUHII change trend. This method not only preserves the transformation patterns of SUHII for all cities but also enhances the interpretability of the fitting results, and it can be applied to estimate the overall SUHII change trend in other urban agglomerations in the future. Terrain has a relatively small impact on the urban heat island effect on the city scale. Land use change inhibits the weakening of the SUHI effect, and population change contributes to the formation of this phenomenon. Urbanization and population growth may lead to increased energy demand, thereby affecting energy release and the UHI effect. It is noteworthy that temperatures in urban expansion areas are rapidly increasing, as urban expansion leads to the expansion of surface heat flux. Meanwhile, the temperature growth rate in city centers is the slowest, indicating that urban centers have reached saturation, causing temperatures to spread outward, and resulting in rising temperatures in the suburbs. Consequently, the temperature difference between urban and suburban areas gradually decreases, leading to a reduction in the observed SUHII change trend. The mechanism of the urban heat island effect is highly complex. Our study can only provide general explanations, and specific reasons require more scientifically effective methods for verification and explanation.