Spatiotemporal Regulation of Urban Thermal Environments by Source–Sink Landscapes: Implications for Urban Sustainability in Guangzhou, China

Abstract

Urban thermal environments critically impact human settlements and sustainable urban development. In this study, a multi-index framework integrating Landsat TM/ETM+/OLI observations (2004–2019) is developed to quantify the contributions of “source–sink” landscapes to urban heat island (UHI) dynamics in Guangzhou, China, with direct implications for advancing sustainable development. Urban–rural gradient analysis was combined with emerging spatiotemporal hotspot modeling, revealing the following results: (1) there were thermal spatial heterogeneity with pronounced heat accumulation in core urban zones and improved thermal profiles in northern sectors, reflecting a transition from “more sources, fewer sinks” in the southwest to “fewer sources, more sinks” in the northeast; (2) UHIs were effectively mitigated within 25–35 km of the city center, with the landscape effect index (LI > 1) indicating successful sink-dominated cooling; (3) spatiotemporal hotspots were observed, including persistent UHIs in old urban areas contrasting with environmentally vulnerable coldspots in suburban mountainous regions, highlighting uneven thermal risks. This framework provides actionable strategies for sustainable urban planning, including optimizing green–blue infrastructure in UHI cores, enforcing cool material standards, and zoning expansion based on source–sink dynamics. This study bridges landscape ecology and sustainable development, offering a replicable model for cities worldwide to mitigate UHI effects through evidence-based landscape management.

1. Introduction

The urban heat island (UHI) effect—where urban areas exhibit significantly higher temperatures than surrounding rural zones—has emerged as a pressing urban ecological issue. As a typical ecological challenge in the global urbanization process, UHIs directly threaten the achievement of sustainable urban development goals, exacerbating energy consumption, deteriorating air quality, and increasing public health risks, thereby becoming a key bottleneck for enhancing urban ecological resilience [1]. First documented by Howard in 1833 [2] and later formally defined by Manley in 1958 [3], UHIs have become increasingly pronounced since the turn of the twenty-first century, largely driven by accelerated urbanization and shifting land use patterns [4]. UHIs negatively impact urban ecosystems and residents by intensifying greenhouse effects [5], deteriorating human health, increasing energy consumption, and raising ambient temperatures [6,7,8,9].

In subtropical megacities like Guangzhou—a core city in the Guangdong–Hong Kong–Macao Greater Bay Area—the UHI effect is particularly pronounced due to high-density development and impervious surfaces [10]. Over the past decade, Guangzhou’s urban thermal environment has rapidly evolved, characterized by expanding heat islands and shrinking cooling zones, largely driven by urban sprawl and anthropogenic land transformation [11,12,13]. Similar UHI trends have been documented in other subtropical cities like Shenzhen and Hong Kong [14,15], highlighting the urgent need for improved mitigation strategies.

To address UHI spatial dynamics, landscape ecologists have adopted the concept of the source–sink landscape, a framework initially rooted in global change and atmospheric pollution research [16]. In this context, “source” landscapes are urban features that generate or intensify heat, while “sink” landscapes (e.g., vegetated or water-covered areas) absorb or mitigate thermal energy. The source–sink landscape framework also provides an innovative perspective for balancing urban development with ecological conservation, serving as an important theoretical tool for advancing Sustainable Development Goal 11 (Sustainable Cities and Communities) [17]. This conceptual model aids in interpreting spatial heterogeneity in urban thermal environments. However, defining and classifying source–sink landscapes remains contested. Existing approaches—including mean land surface temperature (LST) thresholds [17,18,19,20,21] and thermal contribution indices [22]—suffer from several limitations, though they are relatively simple and intuitive; they often neglect landscape heterogeneity, oversimplify spatial patterns, and lack adaptability across different urban contexts [16,23]. Moreover, these classification methods are rarely evaluated critically or standardized. This study proposes an improved, integrative classification framework that accounts for landscape composition, structure, and urban–rural gradients, aiming to overcome these methodological constraints and advance the practical application of source–sink theory in thermal landscape research.

Although a growing number of studies apply the source–sink concept to explore the relationship between land use and urban thermal variation, most focus on single factors or specific case cities [17,18,19,20,21,22,24,25]. These works demonstrate the utility of the source–sink framework but often disregard localized climatic influences and spatial–temporal dynamics, limiting generalizability.

Another popular research area for interdisciplinary collaboration is thermal environments under an urban–rural differential. By situating this analysis within an urban–rural gradient framework [26,27,28,29,30,31,32,33], studies have further investigated how these landscape elements interact to regulate urban thermal conditions, drawing on regional examples such as Wuhan [34,35] and employing spatiotemporal visualization techniques [35,36].

Furthermore, terms like “hot environment” and “hotspot” are frequently used interchangeably, yet they represent distinct phenomena. In this research, “hot environment” denotes general thermal stress in built environments [37], and “hotspot” identifies localized high-temperature zones within the urban matrix [38]. Standardizing these definitions improves conceptual clarity and analytical consistency.

To fill these research gaps, the present study integrates multiple indices to categorize “source” and “sink” landscapes [22], providing an operational advance through explicit coupling of multidimensional landscape metrics with thermal dynamics across urban–rural gradients—a linkage rarely quantified in prior source–sink studies. While this enhances diagnostic specificity over single-index methods, the framework’s conceptual novelty is constrained by its adherence to established paradigms. This study examines the role of “source–sink” landscapes in regulating urban thermal dynamics under the urban–rural gradient context, potentially offering a scientific foundation for enhancing urban habitats.

The remainder of this paper is organized as follows. Section 2 describes the study area and the multiple sources of the remote sensing data. Section 3 subsequently introduces the relevant methods for analyzing land surface temperature, source–sink landscapes, urban–rural gradients, and emerging spatiotemporal hotspots. The results are then presented in Section 4, followed by a discussion in Section 5. Finally, the paper concludes in Section 6. The study’s framework is shown in Figure 1.

2. Study Area and Data Sources

2.1. Study Area

The research was carried out in Guangzhou (Figure 2), situated in the southern region of mainland China in the south-central part of Guangdong Province. It is the provincial capital and a center for political, economic, scientific, technical, educational, and cultural endeavors. Guangzhou extends from 112°57′ to 114°03′ E and 22°26′ to 23°56′ N. The urban heat island intensity in Guangzhou in 2022 was 1.0 °C, 0.3 °C lower than that in 2021 [39]. The heat island intensity is the strongest in spring and autumn, and the heat island intensity in summer in all locations meets these requirements [40].

2.2. Data Sources

The research data primarily comprised remote sensing image data from the Thematic Mapper (TM) on board Landsat 5, the Enhanced TM Plus (ETM+) on board Landsat 7, the Operational and Land Imager (OLI) on board Landsat 8. Remote sensing image data from the Landsat TM5/ETM+/OLI series were sourced from the Google Earth Engine (GEE) platform (http://earthengine.google.com (accessed on 10 August 2024)) for the period 2004–2019, specifically from October to December, with less than 8% cloud cover and a spatial resolution of 30 m. Four representative years (2004, 2009, 2014, and 2019) were selected from the continuous image series from 2004 to 2019 based on comprehensive considerations of data availability and integrity. The remote sensing images from October to December were chosen because Guangzhou has less precipitation and more clear and cloudless weather, which makes the data results more desirable; secondly, the effects of anthropogenic heat emissions from summer heat islands and sudden changes in meteorological conditions in summer were eliminated, so that the study could focus on the analysis of thermal “source–sink” dynamics caused by landscape changes [41]. To ensure data quality, we implemented rigorous quality control through the visual screening of substandard data and cross-validation of LST retrieval results against concurrent moderate-resolution imaging spectroradiometer (MODIS) land surface temperature products to address potential impacts from various acquisition-related factors on remote sensing imagery.

3. Methods

3.1. Spatiotemporal Characteristics of Land Surface Temperature

3.1.1. Retrieval of Land Surface Temperature

Land surface temperature (LST) is the result of multiple factors, such as the natural environment [1,42,43,44] and land use [45,46,47,48]. Considering the available parameters, a single-channel algorithm [49,50] was selected in the current study for retrieving LST. The basic principles of the inversion algorithm were determined using a model outlined in the Landsat user manual with revised calibration parameters from Chander [51]:

$$L_6=gain\times DN+bias$$

(1)

$$T=K_2/ln\left(K_1/L_6+1\right)$$

(2)

In this context, L₆ represents the radiance measurement of the image element at the sensor for the enhanced thematic mapper plus (ETM+) thermal infrared band. The gray value of the image element is denoted as a digital number (DN), and the gain and bias refer to the specific values associated with the thermal infrared band. The sensor-recorded temperature is denoted as T. Calibration coefficients K₁ and K₂ are defined as 606.09 W/(m²·sr·µm) and 1282.71 K, respectively, where K₁ characterizes radiative sensitivity and K₂ represents blackbody reference temperature.

Temperature (T) was corrected for a specific radiance [52] and converted to land surface temperature:

$$LST=T/[1+(\lambda T/\rho )ln \epsilon ]$$

(3)

where λ is the ETM+6 band’s center wavelength ($\lambda$ = 11.45 $\mathsf{\mu }\mathrm{m}$), $\rho$ = 1.438 × 10⁻² mK, and $\epsilon$ is the surface-specific emissivity [53].

3.1.2. Classification of LST

The heat island intensity of the study area was classified using a mean–standard deviation grading approach by determining the mean and standard deviation of the inverted ground temperatures for 2004, 2009, 2014, and 2019. Five categories were created based on the surface temperature inversion results: low, secondary low, medium, secondary high, and high LST (

Table 1. Calculation of the ranges of different surface temperature intervals.

Temperature Zone		Breakpoint Description
Low LST zone	Non-UHI	U − STD > T
Secondary-low LST zone		U − 0.5STD > T ≥ U − STD
Medium LST zone		U + 0.5STD > T ≥ U − 0.5STD
Secondary-high LST zone	UHI	U + STD > T ≥ U + 0.5STD
High LST zone		T ≥ U + STD

U: mean value; STD: standard deviation; LST: land surface temperature; UHI: urban heat island.

). This categorization facilitated an analysis comparing the geographical distribution of the heat island effect across various time points.

3.1.3. Trajectory of the Change in Center of Gravity

Analysis of the high/low-temperature spatial center of gravity using LST can be applied to investigate the dynamics of both cold and hot environments across various levels [54]. In the current study, the changing trajectory of the center of gravity in a thermal environment was studied using a weighted center of gravity model. The elements related to the change in the pattern of space in that environment were emphasized. The trajectory direction of the surface thermal centroid migration reveals the spatial evolution pattern and dominant trend of temperature distribution, while the migration distance quantifies the dynamic intensity of thermal spatial redistribution [55]. The weighted center of gravity model was calculated as follows:

$$X_{Ti}={\displaystyle \frac{{\sum }_{i=1}^n{T}_{ti}{X}_{ti}}{{\sum }_{i=1}^n{T}_{ti}}}$$

(4)

$$Y_{Ti}={\displaystyle \frac{{\sum }_{i=1}^n{T}_{ti}{Y}_{ti}}{{\sum }_{i=1}^n{T}_{ti}}}$$

(5)

where X_Tt and Y_Tt represent the longitude and latitude of the gravitational center located in the high/low-temperature areas of the LST in Guangzhou, respectively, and T_ti denotes the surface temperature of raster I in year t. In year t, the longitude and latitude coordinates of raster I are denoted as X_ti and Y_ti, and n represents the number of rasters within the high/low-temperature zone.

3.2. “Source–Sink” Landscapes

3.2.1. Identification of “Source–Sink” Landscapes

The standardized preprocessing workflow encompassed radiative normalization (Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes atmospheric correction), sensor-specific calibration (radiance-to-reflectance conversion), and automated orthorectification, executed through the integrated ENVI 5.3.1 and ArcGIS Pro 3.0.1 platforms to ensure radiometric consistency and geometric fidelity. The inversions of LST, normalized difference vegetation index (NDVI), normalized difference moisture index (NDMI), and impervious surface area ratio (ISA) were conducted. These four indices were selected on their well-documented effectiveness in characterizing key drivers of urban thermal environments and their thermal effects, demonstrated in numerous studies. The formula for LST inversion is presented above, while the main calculation formulae for the other indices are below [56,57]:

$$NDVI={\displaystyle \frac{{band}_{NIR}-{band}_{Red}}{{band}_{NIR}+{band}_{Red}}}$$

(6)

$$NDMI={\displaystyle \frac{{band}_{Green}-{band}_{NIR}}{{band}_{Green}+{band}_{NIR}}}$$

(7)

$$Fr={\displaystyle \frac{NDVI-{NDVI}_{soil}}{{NDVI}_{veg}-{NDVI}_{soil}}}$$

(8)

$$ISA=(1-Fr{)}_{dev}$$

(9)

where band_NIR, band_Red, and band_Green denote the near-infrared, red, and green spectral bands of the images, respectively; Fr is the fractional vegetation cover; NDVI_soil and NDVI_veg refer to the normalized vegetation indices for bare soil and vegetation in the study area, typically taking the minimum and maximum values as representatives; ISA is the impervious surface area ratio; and dev indicates that the quantity is defined exclusively for regions classified as developed [58].

Landscape type is represented by the high (H) and low (L) values of the four surface features in a fixed order (LST, NDVI, NDMI, and ISA), such that landscape type HLLH denotes a landscape with high LST, low NDVI, low NDMI, and high ISA; the remaining landscape types follow the same naming conventions [59]. Based on the average values of four key indicators (LST, NDVI, NDMI, and ISA), a comprehensive comparison was performed. Then, the threshold for distinguishing high (H) and low (L) values of each indicator was determined as follows: considering the 15-year study period, the threshold for each of the four indicators (LST, NDVI, NDMI, and ISA) was calculated by first computing the annual average value of the indicator for each of the 15 years, then summing these 15 annual average values, and finally taking the mean of this sum. Specifically, for each indicator, if its value was higher than the corresponding threshold, it was classified as high (H); if lower, it was classified as low (L). The specific thresholds are shown in

Table 2. Specific values of thresholds.

LST	NDVI	NDMI	ISA
20.296	0.194	0.163	0.775

.

Using this decision method, the landscape types were categorized into 16 distinct classes (Figure 3). HLLH and HLHH were identified as “source” landscapes, whereas the “sink” landscapes were LLLH, LLHL, LLHH, LHLL, LHLH, LHHL, LHHH, HLHL, and HHHH.

3.2.2. Landscape Contribution Index (CI) and Landscape Effect Index (LI)

In this study, the contribution index (CI) and landscape effect index (LI) were used to evaluate how heat island effects influence “source” and “sink” landscapes. The impact of heat islands on these landscapes was assessed using CI and LI metrics. The thermal contribution of source–sink landscapes was quantified through the temperature deviation (individual landscape’s mean surface temperature minus regional baseline) scaled by its proportional area within the total region [17].

$$CI=\left(T_i-T_{mean}\right)\times S_i/S$$

(10)

In this context, Ti denotes the mean surface temperature of the “source” or “sink” landscape within study unit i, whereas Tmean signifies the average surface temperature across the entire study unit. In spatial unit i, the source–sink land use category occupies area Sᵢ, while S denotes the aggregate spatial extent of the study unit. Generally, the average surface temperature of “source” landscapes was higher than the regional average surface temperature, resulting in CI > 0; similarly, the CI of “sink” landscapes was <0. The contribution index (CI) reflects the effect of a landscape on the urban thermal environment: a positive CI denotes a warming “source” effect, whereas a negative CI signifies a cooling “sink” effect. The absolute value of CI indicates the strength of this influence.

The landscape effect index between “source” and “sink” landscapes is represented by the ratio of the contribution from “sink” landscapes to that of “source” landscapes [60]:

$$LI=\left|{CI}_{sink}/{CI}_{source}\right|$$

(11)

In this context, CI_sink and CI_source denote the contributions of “sink” and “source” landscapes, respectively. The landscape effect index (LI) was calculated by dividing CI_sink by CI_source. LI > 1 suggests that the “source–sink” landscape configuration in the study area effectively mitigated the heat island effect. Conversely, LI < 1 indicates that the landscape arrangement did not alleviate the UHI effect. LI = 1 implies that the “source–sink” landscape configuration had a neutral impact on mitigating the heat island effect.

3.3. Analysis of Urban–Rural Gradient

Examining the urban-to-rural continuum is crucial for comprehending the spatial variations in urban heat islands and discerning the distinct patterns of “source–sink” landscapes along this gradient. In recent years, scholars both domestically and internationally have employed various metrics—such as population density, road network density, residential density, and urban land cover ratio—to delineate urban–rural gradients. Building on these approaches, this study selects population density as the primary indicator for urban gradient classification, based on a comprehensive consideration of the research context and data availability. A density core distribution map was generated by aggregating 16 years of population density data through kernel density analysis. With this density core as the center, concentric buffers were established: the initial buffer had a radius of 15 km, followed by additional gradients extending outward in 10 km increments, resulting in a total of seven distinct zones: ≤15 km, >15–25 km, >25–35 km, >35–45 km, >45–55 km, >55–65 km, and >65–75 km (Figure 4).

To capture the variations in landscape contribution and effect indices across “source–sink” landscapes and urban–rural gradients, landscape indices were extracted from each gradient zone. The mean values of these indices were calculated as a percentage of the region’s area and the density of “source–sink” landscapes. This approach facilitated the analysis of landscape index characteristics across different gradient zones to be able to assess the contribution and spatial heterogeneity of various “source–sink” landscape compositions to LST.

3.4. Emerging Spatiotemporal Hotspot Analysis

The spatiotemporal cube [61] is a visualization model that simultaneously represents the spatial distribution and temporal changes in spatiotemporal data. It effectively conveys both time series and spatial patterns using the X- and Y-axes to denote spatial coordinates (latitude and longitude) and the Z-axis to represent the temporal dimension. In this model, space–time cubes aligned along the same temporal axis form time slices corresponded to specific time points, whereas those sharing identical spatial coordinates constituted the time series for particular geographic locations.

In this study, emergent hotspot spatiotemporal analysis was used to identify temporal evolutionary trends and the spatial distribution of spatiotemporal patterns in the thermal environmental spatiotemporal data. This method is based on spatial hotspots and time series analysis using a spatiotemporal cube model. Temporal sequence analysis was incorporated into spatial hotspot analysis to increase the calculation of temporal hotspots, thereby detecting and determining the existence of hotspots and coldspots spatiotemporally. Emerging hotspot spatiotemporal analyses use Getis–Ord Gi* statistics [36] to detect hot- and coldspots in the spatial distribution of thermal environmental elements, whereas the Mann–Kendall trend analysis method [62] is used to detect trends in time series changes in hot- or coldspots. Based on statistical z-scores and p-values, landscape contribution patterns are classified into 16 distinct hot/coldspot categories: historical coldspots, oscillating cold, sporadic cold, diminishing cold, persistent cold, intensifying cold, consecutive cold, historical hot and cold, oscillating hot, sporadic hot, diminishing hot, persistent hot, intensifying hot, consecutive hot, and new hot [63].

4. Results

4.1. Temporal and Spatial Variations in the Thermal Environment

4.1.1. Variations in the Thermal Environment’s Spatial Pattern

Using statistical calculations of LST for different years in Guangzhou, the distribution of LST for other years was obtained according to geothermal temperature division criteria. The corresponding statistical outcomes are presented in

Table 3. Calculation of land surface temperature (°C) interval ranges.

Year	Low LST	Secondary Low LST	Medium LST	Secondary High LST	High LST
2004	T < 20.603	20.603 ≤ T < 21.434	21.434 ≤ T < 23.096	23.096 ≤ T < 23.927	23.927 ≤ T
2009	T < 20.816	20.816 ≤ T < 21.902	21.902 ≤ T < 24.074	24.074 ≤ T < 25.160	25.160 ≤ T
2014	T < 21.999	21.999 ≤ T < 23.175	23.175 ≤ T < 25.527	25.527 ≤ T < 26.703	26.703 ≤ T
2019	T < 21.418	21.418 ≤ T < 22.643	22.643 ≤ T < 25.096	25.096 ≤ T < 26.322	26.322 ≤ T

, while Figure 5 displays the geographical distribution characteristics.

Table 2 demonstrates that the threshold values for each land surface temperature (LST) classification level show a persistent upward progression from 2004 to 2014. Specifically, the upper boundary of the low LST zones increased sequentially from 20.603 °C (2004) to 20.816 °C (2009) and further to 21.999 °C (2014), with analogous progression patterns observed across other LST zones. This evidence reveals a pronounced warming trajectory in Guangzhou’s surface thermal environment throughout this decade. While 2019 witnessed a modest temperature decrease, recorded values consistently maintained elevations above pre-2009 baselines. As illustrated in Figure 4, since 2004, the secondary-low and medium LST zones in central Guangzhou have progressively shifted to secondary-high or high LST zones, whereas the low LST zones have expanded in all directions.

4.1.2. Spatial Center of Gravity Transfer Distance Analysis of the Thermal Environment

The results of the central coordinates (latitude and longitude) for geothermal temperature centers across regions with similar geothermal classifications in Guangzhou over the years are shown in

Table 4. Latitudinal and longitudinal coordinates of the center of gravity for different geothermal class regions.

		2004	2009	2014	2019
Low LST	XTt	113.684	113.731	113.757	113.751
	YTt	23.652	23.639	23.648	23.634
Secondary low LST	XTt	113.489	113.569	113.616	113.616
	YTt	23.391	23.450	23.379	23.415
Medium LST	XTt	113.524	113.528	113.525	113.531
	YTt	23.275	23.269	23.279	23.260
Secondary high LST	XTt	113.525	113.465	113.429	113.425
	YTt	23.272	23.263	23.232	23.243
High LST	XTt	113.484	113.431	113.374	113.375
	YTt	23.183	23.184	23.195	23.211

LST, land surface temperature.

, with the migration trajectory illustrated in Figure 6.

Between 2004 and 2019, the centroids of low, secondary-low, secondary-high, and high LST zones exhibited pronounced migratory patterns, while the medium-LST zone displayed relatively limited displacement. The low LST zone centroid, persistently anchored in central Conghua District, demonstrated a consistent eastward trajectory, with longitudinal coordinates shifting from 113.684° E to 113.751° E and latitudinal coordinates moving from 23.652° N to 23.634° N. In contrast, the secondary-low LST zone followed a more complex path: initially migrating northeastward (2004–2009: 113.489° E → 113.569° E; 23.391° N → 23.450° N), then shifting southeastward (2009–2014: 113.569° E → 113.616° E; 23.450° N → 23.379° N), before finally moving northward with a stable longitude (2014–2019: 23.379° N → 23.415° N). As evidenced in Figure 5 and Table 4, the secondary-high LST zone centroid oscillated recurrently between Huangpu and Baiyun Districts, alternating between northeast and southwest trajectories. Meanwhile, the high LST zone centroid progressed steadily westward, transitioning from central Huangpu District to northern Tianhe District. These differential migration patterns highlight spatially heterogeneous thermal changes across Guangzhou’s urban landscape over the 15-year period.

4.2. Changes in the “Source–Sink” Landscape Pattern of the Thermal Environment

For the analysis, a five-year interval was adopted, focusing on Guangzhou’s “source–sink” landscapes in 2004, 2009, 2014, and 2019. The evolution of these “source–sink” landscapes across different years is illustrated in Figure 7. As time progressed and urbanization intensified, both the “source” and “sink” landscapes tended to increase in their areas, indicating a rise in the extent of both landscape types. The “source” landscapes were concentrated in the west, center, and south, whereas the “sink” landscapes were concentrated in the east and north. From southwest to northeast, the trend was from “more sources and fewer sinks” to “fewer sources and more sinks.”

Changes in the proportion of the “source–sink” landscape area in Guangzhou are shown in Figure 8; the “source” landscape area increased from 2134.807 km² to 2782.122 km² and the area share from 28.94% to 37.70%. The area of the “sink” landscape expanded from 3617.167 km² to 3956.255 km², increasing its proportion from 49.04% to 53.61%; the area of other landscapes tended to decrease, with an area of 1623.481 km² decreasing to 641.595 km², with an area share of 22.02% to 53.61%. This percentage decreased from 22.02% to 8.69%.

4.3. Analysis and Computation of “Source–Sink” Landscape Contribution

Figure 9 illustrates the distribution of CI and LI conditions across the various districts in Guangzhou. The results indicate that only Nansha, Zengcheng, and Conghua successfully mitigated the heat island phenomenon. At the regional scale, the overall warming effect from Guangzhou’s “source” landscapes exceeded the mean warming contribution observed in each individual district’s “source” landscapes. Similarly, the cooling effect of “sink” landscapes was greater than the average cooling contribution from “sink” landscapes in all study areas. Furthermore, the cooling effect of the “sink” landscapes is greater than the average cooling contribution from the “source” landscapes within every district.

4.4. “Source–Sink” Landscape Effect of the Thermal Environment in Urban–Rural Gradients

The spatial distribution of LI indicated a decline from Guangzhou’s center toward the outskirts, prompting the use of an equal-distance buffer analysis approach. The average LI for each buffer zone was determined using a weighted average calculation (excluding the sections of the buffer beyond Guangzhou’s administrative area).

The landscape effect index from the center of the population core to the 15 km range was <1 in all four years and was dominated by the contribution of the “source” area, whereas in the 15–25 km range, the landscape effect index was >1 in 2004 and 2009 and was dominated by the mitigation of the heat island effect. The landscape effect index of the 15–25 km range was <1 in 2014 and 2019, indicating that the area was gradually dominated by the “source” landscape effect over time. In 2019, the landscape effect index of the 15–25 km area was already <1, indicating that the area gradually changed to a “source” landscape role over time.

In the buffer range from the center to 45 km, the “sink–source” landscape area ratio decreased continuously over the four years; in the buffer range beyond 55 km, the “sink–source” landscape area ratio tended to increase and then decrease (Figure 10 and Figure 11). In the buffer zone > 55 km, the “sink–source” landscape area ratio increased and then decreased. The ratio of the “sink–source” landscape area showed an upward trend with increasing distance from the urban core, aligning with the negative relationship observed between landscape patterns and urban development intensity. In addition, combining the “sink–source” landscape area ratio map and the landscape effect index evaluation map indicated that most of the areas with a landscape effect index >1 had a “sink–source” landscape area ratio of >0.57. The “sink–source” landscape area ratio at 25–35 km was >0.57 in 2019. The “sink” and “source” landscape area ratio at 35 km was 0.478, which is <0.57, but its landscape effect index was >1.

4.5. Analysis of Emerging Spatiotemporal Hotspots in the Thermal Environment

Regarding the spatial arrangement of hot- and coldspot patterns (Figure 12), grid areas identified as hotspots were predominantly found in Yuexiu, Haizhu, Panyu, western Baiyun, southern Tianhe, southern Huangpu, southern Zengcheng, southern Huadu, northern Liwan, northern Nansha, and the urban core of Conghua District in Guangzhou. On the other hand, coldspots were predominantly found in the mountainous areas of central Guangzhou and the suburban regions of Conghua and Zengcheng districts, where development levels were relatively lower.

The predominant hotspot categories were mostly oscillating (18.41%), sporadic (9.00%), or continuous (7.97%), whereas the coldspots were mostly oscillating (21.47%) and continuous (8.59%). This suggests that the UHI effect is unevenly distributed, exhibiting specific fluctuations that could be associated with multiple factors, such as seasonal variations, weather patterns, and the intensity of urban activities.

5. Discussion

5.1. Exploration of the Causes of Changes in the Thermal Environment of Guangzhou

The accelerated increase and subsequent slower decrease in Guangzhou’s annual mean surface temperature (2004–2019) aligns with the findings in [64] and mirrors patterns observed in rapidly developing megacities globally, where intense land cover change drives initial warming [65]. Urban expansion, primarily southward and eastward [66], fueled this trend. Post-2001, rapid heat island growth in Huadu and Panyu significantly increased core urban surface temperatures, exemplifying the classic “urban sprawl” phase seen in some cities in Thailand [67].

Advancements in clean energy utilization demonstrably mitigated this effect. Achieving “coal-free zone” status in Yuexiu, Haizhu, and four other districts by 2014 [68] directly reduced the UHI effect and surface temperatures. This is consistent with findings in 113 key cities in China [69]. Industrial agglomeration enabled centralized control of heat sources, reducing dissipation and limiting heat island expansion in new districts [70]. Crucially, the increase in green space coverage in populated areas robustly explains the post-2014 decrease in heat island areas and expansion of cold islands, strongly supported by the established cooling role of urban green infrastructure [71].

Shifts in the centers of gravity for secondary-low and high land surface temperature (LST) zones illustrate urban development dynamics. The trajectory of the secondary-low temperature zone (Huangpu North → Conghua South → Zengcheng West) reflects suburban expansion. Conversely, the westward shift in the high LST zone center of gravity is directly linked to intensifying the UHI in the highly urbanized Tianhe District economic hub [72] and demonstrably connected to the city’s strategic reorganization of industrial spaces [73,74].

Changes in “source” (heat-contributing) and “sink” (heat-absorbing) landscapes increased over time. The transition from “more sources, fewer sinks” (southwest) to “fewer sources, more sinks” (northeast) correlates with the topography (mountainous north, flat south/center). The “fewer sources, more sinks” pattern in the northern mountains is demonstrably linked to ecological restoration efforts, including the Guangdong Ecological Forest Protection Program and large-scale afforestation [75], which increased vegetation cover and enhanced heat reflection [76]. Large, concentrated “source” patches correspond spatially to dense industrial/population centers with high impervious surface cover [77]. Inadequate protection allows outward expansion of impervious surfaces to encroach on “sink” landscapes [78].

5.2. The Influence of Urban Planning on Alterations in the Thermal Environment

Analysis combining the landscape effect index (LI) and landscape density reveals a key relationship: buffer zones with LI > 1 typically had a “sink–source” area ratio > 0.57. The exception in 2019 (25–35 km buffer: ratio = 0.478 < 0.57, LI > 1) strongly indicates that rational landscape pattern planning enhanced this zone’s UHI mitigation capacity beyond simple area metrics, underscoring the critical importance of green space configuration [79].

Guangzhou’s planning framework (“expanding south, refining north, progressing east, linking west, adjusting centrally”) facilitated a shift from expansion to optimization. The 25–35 km perimeter, encompassing Huadu, Huangpu, Nansha, and the central area, is a policy priority zone [80]. The explicit “livable city” goal within the Guangzhou Plan draft [81] demonstrably contributed to UHI alleviation, echoing the effectiveness of integrating climate goals seen in six global cities [82].

Thermal contrasts exist between urban cores. The old urban areas, with early development, high density, and industrial agglomeration, exhibit stable, strong UHI effects (persistent hotspots), akin to historic European cores [83]. Development in southern Huadu, Huangpu, and Zengcheng, guided by outward industry transfer strategies, resulted in dispersed hotspot clusters [84]. Crucially, protected forested land near Baiyun Mountain and in suburban districts (Huadu, Conghua, Zengcheng) forms stable coldspots that suppress UHIs and foster surrounding cooling, highlighting the vital ecosystem service of peri-urban forests [85].

5.3. Limitations and Future Research

This study has several limitations requiring consideration. The exclusive focus on October–December data introduces seasonal bias, as vegetation phenology significantly alters “source–sink” dynamics—summer vegetation typically acts as a stronger heat sink, suggesting warm-season cooling potential may be underestimated in our landscape classification. Additionally, reliance on Landsat-derived surface temperature (LST) leads to the capture of skin temperature rather than the near-surface air temperature (Ta) experienced by residents, with the LST–Ta relationship varying across landscapes and timescales [86]. While the 15-year period captures urbanization impacts, it may not fully resolve long-term climate trends. Future work should expand the temporal scope to capture annual and diurnal thermal cycles, incorporating real-time sensor networks, employ multi-seasonal remote sensing to quantify intra-annual thermal dynamics, integrate mesoscale climate modeling with land use scenarios to provide predictive insights [87], and conduct high-resolution microclimate studies using Unmanned Aerial Vehicles or sensor networks to inform localized mitigation design [88].

The resolution of remote sensing data may mask fine-grained thermal variations, while anthropogenic heat sources—such as traffic emissions and air conditioning exhaust—were not explicitly incorporated into the landscape effect model. Future research should address these gaps and deepen insights by (a) quantifying the differential cooling/warming efficiencies of specific source–sink landscape types using controlled experiments or high-resolution thermal imaging, (b) integrating socioeconomic data and anthropogenic heat flux models to refine the framework’s predictive capability under diverse urban scenarios, and (c) validating the framework’s transferability through rigorous application in contrasting city types (e.g., arid, coastal, and high-latitude cities) to develop context-specific adaptation strategies.

6. Conclusions

This study establishes a novel framework integrating multi-source remote sensing, landscape ecology metrics, and spatiotemporal cube analysis to regulate urban thermal environment dynamics. The framework was applied to Guangzhou, revealing the critical impact of the “source–sink” landscape configuration on urban heat island (UHI) development and severity. A pronounced UHI dominated the central region, while the northern cold island effect was milder. The landscape transitioned spatially from “more sources, fewer sinks” in the southwest to “fewer sources, more sinks” in the northeast. Crucially, the landscape effect index indicates that urban master planning can be a potent tool for UHI mitigation; specific policy implications include prioritizing green–blue infrastructure networks in core UHI zones, enforcing cool material standards in redevelopment, and strategically zoning future expansion areas based on source–sink optimization principles.

These findings offer actionable strategies for optimizing Guangzhou’s ecological planning and thermal resilience. The framework’s core methodology—systematic multi-source data fusion, standardized landscape metric calculation, and spatiotemporal cube visualization—is inherently transferable. Thus, this study offers a replicable framework for diagnosing thermal challenges and assessing planning interventions, aimed at balancing sustainability with livability, for global cities that are facing dual climatic and urbanization pressures.