1. Introduction

2. Literature Review

3. Methodology

3.2. Data Sources and Processing

3.2.1. Grid cells versus street blocks

LST exhibit spatial variations influenced by local physical characteristics, and various spatial units are employed to delineate these locations and process local attributes. Geospatial data are commonly distributed in raster or lattice (areal vector) format. Raster-grid-based analysis has been extensively utilized for LST data analysis, owing to its efficiency in handling large-scale continuous geospatial information, particularly since LST images are initially generated as collections of image pixels.

Despite the advantages of grid-based data processing, it involves dividing physical elements into uniformly-sized grid cells, leading to artificial truncation that may introduce biases in LST estimation using physical variables. The grid size is determined by sensor capabilities rather than scientific rationale. Thus, this study explores whether lattice-based analysis yields different results compared to the grid-based approach (Figure 2). Urban landscapes are divided into homogenous blocks by streets and roads, creating coherent spatial units in terms of land use and buildings. These street-enclosed blocks likely represent homogeneous neighborhoods governed by similar land use regulations, including design codes and caps for building coverage ratio (BCR) and floor area ratio (FAR). Consequently, we also utilize street-enclosed neighborhood blocks as alternative spatial units for LST analysis and compare the results obtained from each approach. The Ministry of the Interior and Safety has recently developed a street block unit system known as the state basic district (SBD), where boundaries are delineated based primarily on transportation and river networks. In this study, we select and compare the grid and street blocks as spatial units of analysis.

3.2.2. Land surface temperatures

The seasonal LST values across Seoul were extracted from Landsat 8 Thermal Infrared Sensors (TIRS) Collection-2 Level-2 products, featuring a 30m spatial resolution, accessed through Earth Explorer, the data distribution platform operated by the United States Geological Survey (USGS). This dataset comprises atmospherically corrected surface reflectance and surface temperature values derived from Level-1 inputs meeting specific technical criteria to ensure scientific validity. Remotely sensed (RS) LST data allows for spatially continuous analysis and exhibits a strong correlation with AT measures, which are typically available from only a limited number of meteorological stations. To ensure image quality, it is crucial to select imagery with minimal cloud cover. For this study, clear atmospheric conditions representative of the four seasons in Seoul were available for the year 2017 (Figure 3). Table 1 presents the date and time of data collection.

Seoul boasts extensive nature preserves, with approximately 13% of the city's area covered by public waters, such as rivers and tributaries, and an additional 25.3% covered by forests and woods. In total, approximately 38% of the city comprises uninhabited natural preserves. As these areas lack urban built-up form variables, they were excluded from the subsequent analysis to avoid zero-inflation issues in the dataset. Moreover, the LST patterns of natural features differ significantly from those of urbanized areas. Hence, this study focuses exclusively on the built-up morphology of urban areas and its impact on LST. To address the spillover effects of nearby natural features on the LST of adjacent urban areas, the final dataset includes variables quantifying the proximity to forests, rivers, small green spaces, and water bodies as distinct land-use types.

For the grid-based analysis, the LST cells with a 30m resolution were resampled to 100m by 100m, using bilinear interpolation in ArcGIS to align with building geometry and density information, where the smallest unit of distribution is a 100m grid. On the other hand, for the block-level analysis, the mean LST values were calculated for each individual block. In Seoul, a total of 34,135 grid cells and 4,553 street-blocks were considered for this study.

Figure 3. Seasonal distribution of urban land surface temperature in 2017, Seoul, South Korea: (a) spring, (b) summer, (c) autumn, (d) winter.

Figure 3. Seasonal distribution of urban land surface temperature in 2017, Seoul, South Korea: (a) spring, (b) summer, (c) autumn, (d) winter.

3.2.3. NDVI and albedo

The Normalized Difference Vegetation Index (NDVI) is a widely used indicator of vegetation density or photosynthetic activity, and is calculated as follows:

$$NDVI=\frac{NIR – RED}{NIR+RED}$$

(1)

where RED and NIR are the surface reflectance in the red and near-infrared bands. These SR values are in eq. (2) obtained from Landsat 8 data at a 30m resolution, and NDVI values are calculated for each of the four seasons. NDVI values range from -1.0 to 1.0, with higher values indicating higher vegetation density and health, whereas lower values indicating impervious land with limited vegetation or water content.

Surface albedo quantifies the extent to which a surface reflects solar radiation and is expressed as a percentage ranging from 0 (indicating no reflection) to 100 (indicating complete reflection). Surfaces with high albedo values include white materials like snow and clouds, which remain cooler due to their ability to reflect a significant portion of incoming solar radiation. Conversely, darker surfaces, such as asphalt, dark trees, and soil, exhibit low albedo values and tend to absorb a larger portion of solar energy, leading to increased warmth. Given the influence of surface albedo on surface temperature, it becomes essential to account for this characteristic when modeling LST. The calculation of albedo is based on Landsat SR bands, utilizing the formula proposed by Liang (2001):

$$Shortwave albedo=\frac{0.356\times Blue+0.130\times Red+0.373\times NIR+0.085{SWIR}_1+0.072{SWIR}_2-0.0018 }{0.356+0.130+0.373+0.085+0.072}$$

(2)

where NIR is near-infrared and SWIR 1 and 2 represent short-wave infrared in different bandwidths. Both of the seasonal NDVI and surface albedo values are aggregated at the grid level of 100m and the block level, respectively.

3.2.4. Land uses and buildings

Land use classification data available at the land parcel level in vector form is downloaded from the Korean Ministry of Environment’s website, Environmental Geographic Information Service (EGIS). The classification is conducted via manual digitization based on very-high-resolution orthorectified aerial images (0.25m/pixel). In this study, seven land-use categories are considered – residential, commercial, industrial, cultural & recreational, transport, public, and others (non-urban uses), and their proportions within each grid and block are correlated with seasonal LST changes. In addition, the entropy index is used as a metric to measure the level of land-use diversity, and is calculated as follows:

$$Entropy=-\frac{\left(\right[{\sum }_{j=1}^k{p}_{j}ln p_j] }{\ln k}$$

(3)

where $p_j$ is the percentage of land use type j in the grid cell (block) and k is the total number of land-use types considered. It ranges from 0 to 1, higher values indicating a balanced mix of different land uses.

This study gathered GIS data on various building characteristics in Seoul, including the number of buildings, the percentage of aged buildings, building coverage ratio, total floor area, floor area ratio, and building height. The data was obtained from the National Geographic Information Institute (NGII), which provides building information in two different data formats: grid cells of varying sizes and lattice formats encompassing State Basic Districts (SBDs), census tracts, ZIP code areas, villages, and municipalities. To ensure data consistency, the finest spatial unit from each format was selected for data acquisition, resulting in a grid size of 100m and the use of SBDs. While individual building information would have been valuable, it was unfortunately unavailable for the study area. Therefore, building form indicators, as processed by the NGII, were retained along with land-use proportions and land-use diversity variables for subsequent statistical analyses. The study employs six building indicators, three related to two-dimensional aspects, and the remaining three concerning three-dimensional volumes.

3.2.5. Gravity Index

Two gravity indices are calculated for the grid and block level analyses: gravity index for urban forests and for water bodies. The gravity index measures the distance-decay impacts of urban forests and water bodies on LST. First, a 1,000 m buffer is constructed around the centroid of grid cell or street-block i, and the areas of urban forests and water bodies are converted into points for every 30m in GIS. Let $B_i$ be the buffer of centroid i, j a point within this buffer, $d_{ij}$ the distance between the centroid i and point j, ${NF}_j$ the area, either forests or waters, represented by point j (900m2), and α a positive exponent. The gravity index ${GI}_i$ is defined as follows:

$${GI}_i=\underset{j\in B_i}{\Sigma }\frac{N{F}_j}{{d_{ij}}^a}$$

(4)

This gravity index (GI) accounts for the fact that the influence of forests on LST may be reduced by the distances of these areas, and the effect of distance is discounted by exponent α. Higher α values decrease the importance of farther-away forests (waters) while increasing the influence of nearby ones. Several α values (1.0, 1.5, 2.0, 2.5, 3.0) were tried, with α = 1.5 providing the best statistical fit. The mapping of the gravity indices for urban forests (GIUF) and for water bodies (GIWB) is presented in Figure 4 and Figure 5. The final set of variables estimating LST, and their sources are presented in Table 2.

Table 2. List of variables and descriptions.

Table 2. List of variables and descriptions.

Category		Variables		Unit	Source
Dependent		Seasonal land surface temperature (LST)		℃	Landsat 8 TIR
Explanatory	Vegetation	Seasonal normalized difference in vegetation index (NDVI)		[0~1]	Landsat 8 OLI
	Land use	Land use type proportion	Residential	%	Korea Environmental Geographic Information Service (EGIS) 2017
			Commercial	%
			Industrial	%
			Cultural & Recreational	%
			Transport	%
			Public	%
			Other uses	%
		Land use diversity	Entropy index	[0~3]
		Reflectance	Albedo	%	Landsat 8 OLI
	Building	2-D	Number of buildings	-	Korea National Geographic Information Institute (NGII) 2017
			Percent old buildings (+35 years)	%
			Average building coverage ratio (BCR)	%
		3-D	Average floor area ratio (FAR)	%
			Average total floor area	m2
			Average building height	m
	Natural Area	Proximity to natural areas	Gravity index for urban forests	-	EGIS 2017
			Gravity index for rivers and streams	-

Table 2. Descriptive statistics of LST and the explanatory variables aggregated at the grid level (100m×100m) (N=34,136).

Table 2. Descriptive statistics of LST and the explanatory variables aggregated at the grid level (100m×100m) (N=34,136).

Category	Variable		Mean	Std. Dev.	Min.	Max.
LST	LST (℃)	Spring	20.72	1.65	10.99	31.2
		Summer	37.32	2.37	25.25	49.86
		Autumn	13.35	1.09	-2.68	20.22
		Winter	-2.27	1.11	-11.77	3.90
Vegetation	NDVI	Spring	0.31	0.04	0.18	0.71
		Summer	0.43	0.09	0.25	0.93
		Autumn	0.28	0.04	0.18	0.52
		Winter	0.17	0.02	0.11	0.43
Land	Albedo	Spring	10.39	1.59	0.00	32.59
		Summer	12.48	2.08	0.00	45.44
		Autumn	6.66	1.37	0.00	31.23
		Winter	5.23	1.34	0.00	28.54
	Land use type proportion	Residential	19.73	21.6	0.00	97.27
		Commercial	12.45	15.52	0.00	100.0
		Industrial	0.27	2.77	0.00	99.29
		Cultural & Recreational	1.49	5.79	0.00	100.0
		Transport	40.51	19.35	0.00	100.0
		Public	3.35	7.52	0.00	100.0
		Other uses	22.20	26.09	0.00	100.0
	Land use diversity	Entropy Index	0.49	0.16	0.00	0.88
Building	2-D form	Count of buildings	15.43	17.38	0.00	180
		% of old buildings (+35 years)	17.52	24.49	0.00	100
		Average building coverage ratio	33.14	25.96	0.00	90.0
	3-D form	Average floor area ratio	121.17	124.07	0.00	1420.19
		Average total floor area	3071.25	10294.12	0.00	426719.2
		Average building height	14.24	17.26	0.00	284.0
Natural Area	Proximity to nature	Gravity index for urban forests	18.53	26.73	0.00	196.51
		Gravity index for rivers and streams	15.85	36.30	0.00	3290.23

Table 3. Descriptive statistics of LST and the explanatory variables aggregated at the street-block level (N=4,098).

Table 3. Descriptive statistics of LST and the explanatory variables aggregated at the street-block level (N=4,098).

Category	Variable		Mean	Std. Dev.	Min.	Max.
Block	Street-block area (m2)		65,611.1	71,397.9	3,071.2	2,494,292.3
LST	LST (℃)	Spring	20.44	1.17	15.32	25.93
		Summer	37.62	1.81	28.95	45.21
		Autumn	13.30	0.80	8.70	17.04
		Winter	-2.51	0.78	-6.29	1.07
Vegetation	NDVI	Spring	0.30	0.03	0.21	0.47
		Summer	0.41	0.06	0.31	0.74
		Autumn	0.27	0.02	0.21	0.41
		Winter	0.17	0.01	0.13	0.26
Land	Albedo (%)	Spring	9.99	0.85	7.39	15.00
		Summer	11.99	1.16	8.05	22.38
		Autumn	6.33	0.71	3.53	11.49
		Winter	4.90	0.70	2.83	12.53
	Land use type proportion (%)	Residential	24.87	18.64	0.00	83.98
		Commercial	15.71	13.35	0.00	74.37
		Industrial	0.20	1.82	0.00	49.00
		Cultural & Recreational	1.10	2.36	0.00	31.85
		Transport	40.07	13.26	0.18	91.96
		Public	3.08	4.08	0.00	48.96
		Other uses	14.97	18.06	0.00	99.82
	Land use diversity	Entropy Index	0.60	0.11	0.01	0.91
Building	2-D form	Building density (km2)	20.57	15.99	0.00	95.80
		% of old buildings (+35 years)	21.43	19.30	0.00	100.00
		Average building coverage ratio	49.13	14.68	0.00	80.23
	3-D form	Average floor area ratio	183.80	113.95	0.00	1409.8
		Average total floor area	3554.87	13344.02	0.00	426719.2
		Average building height	18.26	15.32	0.00	250.73
Natural Area	Proximity to nature	Gravity index for urban forests	23.67	39.64	0.00	1057.1
		Gravity index for rivers and streams	7.27	12.78	0.00	98.85

4. Results

4.1. Quadrant analysis

Based on the average summer LST (37.3℃) and the average winter LST (-2.2℃), grid cells and blocks are categorized into four quadrants. Quadrant 1 (Q1) represents higher LST in both summer and winter, while Quadrant 3 (Q3) exhibits lower LST in both seasons. Ideally, Quadrant 2 (Q2) is the most favored, displaying cooler summer LST and warmer winter LST, while Quadrant 4 (Q4) represents the least favored combination, with hotter summer and colder winter temperatures.

The results of the summer and winter LST quadrant analysis presented in Table 4 reveal that Quadrant 1 is the most prevalent type, encompassing the largest number of grid cells (31.5%) and blocks (37.9%), followed by Q3, which includes 27.7% of the grid cells and 29.3% of the street blocks. This indicates that approximately 60% to 66% of the entire city consistently experiences warmer or colder LST in both summer and winter compared to other areas. Areas with relatively cooler summer and warmer winter temperatures (Q2) account for approximately 17.3% at the grid level and 14.6% at the block level. Meanwhile, areas with the least favorable LST conditions, being warmer in summer and colder in winter (Q4), account for approximately 23.6% at the grid level and 18.3% at the block level.

The spatial distribution of these four quadrants across Seoul's 25 local districts shows significant clustering, with spatial disparities among the districts in terms of quadrant proportions (Figure 7). Some districts located northeast and southwest of Seoul exhibit disproportionately larger numbers of Q4, indicating a large contiguous cluster of grid cells and blocks with unfavorable LST conditions. On the other hand, some other districts located southeast of Seoul show very few traces of Q4. The three districts with the fewest number of Q4 consist of socioeconomically well-to-do communities and high-end commercial centers in Seoul, typically characterized by higher property values and well-maintained green infrastructure. These districts predominantly comprise Q2, Q3, and Q1.

Notably, Q2 (cooler summer and warmer winter LST) tends to be more located along the rivers and streams, particularly the Han River that cuts through Seoul. Q3 (cooler summer and colder winter LST) also exhibits a similar spatial pattern, being situated close to water bodies, but it is more densely clustered near urban forests and mountains north and northwest of Seoul. Additionally, some districts to the east and northwest of Seoul are predominantly composed of Q1 and Q4, indicating vulnerable thermal conditions in summer and winter. These areas coincide with large local clusters of affordable residential areas in Seoul, indicating an inequitable social distribution of summer heat and winter cold.

The comparison of mean values of urban form variables, including NDVI, building form indicators, and proximity to forests and water bodies, provides further insights into the characteristics of each quadrant (Table 4). Quadrant 2 (Q2) exhibits the highest vegetation density in both summer and winter. It also has greater access to large water bodies in its vicinity compared to the other quadrants, as well as moderate access to urban forests nearby. Quadrant 3 (Q3) ranks second to Q2 in terms of vegetation density in both seasons and access to water bodies, but it excels in its proximity to urban forests. It appears that urban nature elements such as greenery and waterfront areas are closely associated with pleasant changes in LST between summer and winter.

Table 4. Comparison of means among LST quadrants: grid-level versus block-level.

Table 4. Comparison of means among LST quadrants: grid-level versus block-level.

Variables	Summer-winter LST quadrants
	Quadrant 1 (hot-warm)	Quadrant 2( cool-warm)	Quadrant 3 (cool-cold)	Quadrant 4 (hot-cold)
Grid-level (N=34,135)
Count	10,747 (31.5%)	5,891 (17.3%)	9,457 (27.7%)	8,040 (23.6%)
Summer LST *	39.32**	35.09**	35.42**	38.50**
Winter LST *	-1.45**	-1.36**	-3.31**	-2.82**
Summer NDVI *	0.40**	0.52**	0.45**	0.39**
Winter NDVI *	0.17**	0.19**	0.17**	0.16**
Count of buildings *	21.05**	3.14**	7.41**	26.38**
Percent old buildings *	23.30**	7.90**	11.65**	23.75**
Average BCR *	40.10**	12.73**	23.71**	49.89**
Average FAR *	128.84**	54.29**	112.80**	169.79**
Average TFA *	1487.5**	2495.1**	6362.3**	1739.3**
Average BH *	11.04**	8.62**	21.23**	14.40**
GIUF *	14.77**	17.82**	24.59**	16.94**
GIWB *	12.48**	35.60**	13.94**	8.12**
Block-level (N=4,553)
Count	1,727 (37.9%)	663 (14.6%)	1,332 (29.3%)	831 (18.3%)
Summer LST *	39.07**	36.30**	35.84**	38.52**
Winter LST *	-1.94**	-1.98**	-3.29**	-2.89**
Summer NDVI *	0.39**	0.45**	0.44**	0.39**
Winter NDVI *	0.17**	0.18**	0.17**	0.16**
Building density *	28.44**	10.32**	10.54**	28.45**
Percent old buildings *	27.27**	16.35**	14.08**	25.12**
Average BCR *	53.75**	43.26**	42.74**	54.48**
Average FAR *	172.46**	178.84**	199.84**	185.62**
Average TFA *	1250.9**	4906.4**	7260.7**	1324.6**
Average BH *	13.61**	19.48**	25.95**	14.65**
GIUF *	13.67**	33.19**	37.18**	15.20**
GIWB *	8.01**	11.00**	5.68**	5.29**
* According to one-way ANOVA test, mean differences between quadrants are statistically significant (p<0.01). ** The mean differences against at least two other groups are statistically significant (p<0.05)

Furthermore, the quadrants significantly differ in terms of building form and geometry. Q2 and Q3 have lower building density, percent of aged buildings, and building coverage ratio (BCR) than Q1 and Q4, but they boast higher or comparable average building height and volumes. This indicates that Q1 and Q4 include areas with high population density, characterized by mid-to-low rise buildings situated close to one another and located farther away from rivers and forests within the city. On the other hand, Q2 and Q3 areas exhibit more spaced-out, taller buildings and have better access to urban green spaces and water features.

4.2. Regression analyses of LST

4.2.1. Model Selection

Multivariate regression models are employed to investigate the impact of individual urban built-up form variables on seasonal LST variations. Given the observed spatial correlation between LST and urban form indicators, spatial regression models are explored. The Queen contiguity criterion is adopted for the spatial weight matrix, defining adjacency between areas when they share a vertex or line segment. Moran's I statistics indicate significant SA in the error term of the OLS models for both grid- and block-level analyses across all four seasons. This points to the necessity of controlling for spatial autocorrelation (Table 5). The Lagrange multiplier (LM) test confirms that both lambda (λ) in the spatial error term (λWω) and rho (ρ) in the spatial lag term (ρWY) have statistically significant values (λ≠0 and ρ≠0), indicating the suitability of SEM and SLM over OLS at both grid and block scales for all seasons. To determine the relative fit of SEM and SLM, the Robust LM test is applied. Robust LM error assesses whether the spatial error term remains statistically significant (λ≠0) when the spatial lag term is included in the model (ρ≠0), while Robust LM lag assesses whether it is still appropriate to include the spatial lag term when the spatial error term is controlled for. The Robust LM test of the four-season LST models at the grid and block levels shows that both Robust LM error and lag statistics are significant, with the former higher than the latter in all cases. This suggests that the GSM that integrates both error and lag terms is preferred. Consequently, subsequent regression analysis of seasonal LST at grid and block scales is conducted using GSM.

Table 5. Statistics for spatial regression model selection at grid and block levels.

Table 5. Statistics for spatial regression model selection at grid and block levels.

Unit	Statistics	Spring LST	Summer LST	Autumn LST	Winter LST
Grid	Moran’s I (error)	231.6***	242.7***	252.9***	221.1**
	Lagrange Multiplier (lag)	23347.6***	17587.1***	28046.9***	43904.2**
	Robust LM (lag)	1185.5***	867.7***	198.6***	2364.5**
	Lagrange Multiplier (error)	53517.2***	58793.9***	63801.3***	48794.3**
	Robust LM (error)	31355.2***	42074.5***	35953.0***	7254.6**
Block	Moran’s I (error)	47.8***	49.9***	53.5***	45.3***
	Lagrange Multiplier (lag)	249.7***	198.4***	368.0***	1432.9***
	Robust LM (lag)	6.7***	16.3***	3.2*	42.2***
	Lagrange Multiplier (error)	2263.0***	2466.2***	2827.1***	2027.0***
	Robust LM (error)	2020.0***	2284.2***	2462.3***	636.3***
***: p-value <0.01, **: p-value<0.05, *: p-value<0.1

4.2.2. Model Estimation: Grid Level

The GSM is employed to explain LST variations across all four seasons, utilizing both grid and block-level databases (Table 6 and Table 7). The spatial error and lag terms demonstrate statistical significance with substantial coefficients, affirming the suitability of spatial regression over OLS. The results indicate that the model for summer LST exhibits the highest goodness of fit statistics, including R² and AIC, with all explanatory variables being statistically significant, followed by the spring LST model. In contrast, the winter LST model shows the lowest model fit compared to the other seasons, suggesting that winter temperatures are more challenging to explain based on urban built-up form indicators, such as vegetation, land use, building characteristics, and proximity to nature. The variance in LST is much smaller in winter than in other seasons, indicating a stronger association between urban form and heat rather than cold (Table 3 and Table 4). Summer experiences the most substantial impact from urban natural and artificial environments.

The results of the grid-based GSM estimation reveal that two urban built-up form factors play a crucial role in the city's adaptive capacity to temperature changes across the four seasons: vegetation density, represented by NDVI, and proximity to water bodies, represented by the gravity index for rivers and streams (GIWB) (Table 6). These two variables exhibit opposite coefficient directions between summer and other seasons. Higher vegetation density leads to cooler summer and warmer winter LSTs. However, a higher winter NDVI, suggesting a higher density of evergreen trees and lawns, increases LST significantly. Additionally, closer proximity to water bodies reduces LST in summer and spring but increases LST in winter and fall. This indicates that areas located near rivers, streams, and other water bodies are more likely to experience lower temperatures in summer and warmer temperatures in winter compared to areas without such proximity (within 1km). Careful management of vegetation health and water content in urban areas holds significant potential to mitigate temperature extremes and naturally regulate heat and cold. However, proximity to urban forests and mountains consistently decreases LST across all seasons. Unlike urban greenery, hilly mountains covered by forests block sunlight and trap cold air near the ground, leading to overall colder conditions. Moreover, forests contribute to increased humidity due to evaporation, intensifying the perception of cold temperatures. Seoul's urban forests, usually situated at higher altitudes, lead to temperature drops both within the forests and in adjacent areas.

Regarding other urban form variables, except for NDVI and GIWB, their effects on LST remain consistent across all seasons without flexible switching. Concerning land use types, all urban land uses increase LST in all seasons. Industrial and cultural/recreational uses tend to have a more significant impact on increasing LST, likely due to the expansive building interior spaces required for industrial production and recreational facilities. Residential and commercial uses, as well as roads and parking lots, also contribute to elevated LST in all seasons compared to non-urban land uses, such as agricultural open spaces. However, the level of land use mixture is associated with LST reductions. The entropy index, representing land use diversity, shows that LST is likely to decrease as the land use mix increases in all seasons. Increasing land use diversity at a compact scale (100m×100m) may help reduce LST in summer and spring but may not be as favorable in fall and winter.

Regarding building morphology, two- and three-dimensional factors exert opposing impacts on LST. Horizontal expansion of building space increases LST in all seasons. The count of buildings is positively associated with higher LST. A higher BCR, indicating less open space and limited spacing between buildings, also leads to increased LST in all seasons. Moreover, an increased number of buildings aged over 35 years results in a further increase in LST, possibly due to heat-prone materials and designs. Conversely, vertical expansion of buildings consistently reduces LST across seasons, with a more significant decrease observed in warmer seasons. A higher FAR and TFA, representing larger building floors and volumes, and greater building height, all contribute to a significant decrease in LST, possibly because taller buildings cast long shadows on streets and surrounding areas, blocking sunlight. The effects of building morphology on LST remain consistent throughout all seasons.

Table 6. General spatial model (GSM) estimation results at the grid level (N=34,135).

Table 6. General spatial model (GSM) estimation results at the grid level (N=34,135).

Variable		Spring LST	Summer LST	Autumn LST	Winter LST
Constant		2.0475***	11.759***	1.4901***	-3.4028***
NDVI		3.5077***	-3.6698***	4.0788***	9.1573***
Albedo		0.1341***	0.0573***	0.1316***	0.1714***
Land use type proportion	Residential	0.0005	0.0093***	0.0060***	0.0028***
	Commercial	0.0065***	0.0153***	0.0071***	0.0036***
	Industrial	0.0186***	0.0340***	0.0159***	0.0067***
	Cultural & Recreational	0.0193***	0.0303***	0.0115***	0.0076***
	Transport	0.0031***	0.0101***	0.0073***	0.0062***
	Public	0.0071***	0.0184***	0.0064***	0.0020***
Land use mix	Entropy Index	-0.5538***	-0.2345***	-0.2339***	-0.2575***
Building 2-D form	Count of buildings	0.0072***	0.0100***	0.0050***	0.0035***
	% of old buildings (+35 years)	0.0017***	0.0032***	0.0009***	0.0006***
	Average building coverage ratio	0.0034***	0.0051***	0.0013***	0.0008***
Building 3-D form	Average floor area ratio	-0.0004***	-0.0006***	-0.0001***	-0.0001
	Average total floor area	-1.0e-06**	-1.0e-06**	-2.0e-06***	-2.0e-06***
	Average building height	-0.0039***	-0.0052***	-0.0005**	-0.0003
Natural area	Gravity index for urban forests	-0.0018***	-0.0113***	-0.0022***	-0.0030***
	Gravity index for water bodies	-0.0002**	-0.0018***	0.0001*	0.0009***
Rho (ρ)		0.777***	0.685***	0.701***	0.702***
Lambda (λ)		0.647***	0.769***	0.748***	0.662***
Adj. R2		0.892	0.923	0.857	0.841
Std. Err.		0.54	0.66	0.41	0.44
AIC		37,325	51,139	19,431	23,078
***: p-value <0.01, **: p-value<0.05, *: p-value<0.1

4.2.3. Model Estimation: Block Level

The R² values of the GSM models for the block-based approach indicate a lower explanatory power compared to the grid-based models. This difference is attributed to the smaller level of spatial autocorrelation observed in the block-based data. Street blocks, being larger in size and enclosed by roads, experience reduced interaction with neighboring blocks, leading to lower spatial autocorrelation. Despite this, the coefficient signs and relative magnitudes of the individual variables in both the block- and grid-based approaches produce similar results. This suggests that grid-based data processing does not lead to distorted results compared to zonal approaches like street blocks. Hence, the choice of spatial unit of analysis does not significantly bias the analysis of LST.

Table 7. General spatial model (GSM) estimation results at the block level (N=4,553) .

Table 7. General spatial model (GSM) estimation results at the block level (N=4,553) .

Variable		Spring LST	Summer LST	Autumn LST	Winter LST
Constant		6.6525***	21.347***	4.0591***	-6.499***
NDVI		2.9606***	-5.3722***	6.8056***	9.1059***
Albedo		0.4429***	0.1354***	0.3110***	0.4796***
Land use type proportion	Residential	0.0006***	0.0179***	0.0104***	0.0061***
	Commercial	0.0123***	0.0291***	0.0160***	0.0095***
	Industrial	0.0420***	0.0972***	0.0394***	0.0160***
	Cultural & Recreational	0.0463***	0.0651***	0.0226***	0.0192***
	Transport	0.0029***	0.0212***	0.0114***	0.0097***
	Public	0.0052*	0.0407***	0.0066***	-0.0071***
Land use mix	Entropy Index	-0.4084***	0.1690	-0.1940**	-0.1257
Building 2-D form	Building density (100m2)	0.0169***	0.0268***	0.0094***	0.0060***
	% of old buildings (+35 years)	0.0046***	0.0074***	0.0009**	0.0003
	Average building coverage ratio	0.0044***	0.0044***	0.0019***	0.0010
Building 3-D form	Average floor area ratio	-0.0007***	-0.0010***	-0.0002	1.0e-05
	Average total floor area	4.0e-06***	4.0e-06***	0.000	-1.0e-06
	Average building height	-0.0083***	-0.0108***	-0.0023***	-0.0029***
Natural area	Gravity index for urban forests	-0.0005*	-0.0059***	-0.0016***	-0.0021***
	Gravity index for water bodies	0.0013	-0.0044***	0.0025***	0.0075***
Rho (ρ)		0.388***	0.381***	0.323***	0.259***
Lambda (λ)		0.407***	0.439***	0.539***	0.494***
Adj. R2		0.731	0.826	0.691	0.638
Std. Err.		0.61	0.75	0.45	0.47
AIC		5,655	7,664	2,962	3,312
***: p-value <0.01, **: p-value<0.05, *: p-value<0.1

In most cases, the grid- and block-based analyses yield similar findings, particularly for variables related to vegetation density and proximity to large water bodies, which help areas mitigate temperature extremes by reducing LST in warm seasons while increasing LST in cold months. However, there are differences in the results, particularly concerning variables related to building morphology. In the block-based results, the LST effects of several building-related variables become insignificant for fall and winter, with only building density and height remaining significant for LST. Higher building density leads to elevated LST, while increased average building height reduces LST consistently across all seasons.

Another disparity between the grid- and block-based results involves the sign of TFA, which switches to positive in summer, while it remains negative in the grid-based analysis. This discrepancy can be explained by the fact that block-level calculations of TFA capture the net effect of gross building floor area, incorporating shade casting, wind channeling, and thermal mass effects of the buildings within blocks. On the other hand, grid-level analysis involves the truncation of building mass into two or more grid cells, making it challenging for a single cell to capture all potential effects of those buildings. As a result, different cells may reflect different influences of building volume on LST. The block-level analysis, which conserves building shape without truncation, suggests that the net effect of building volume is likely to increase LST rather than decrease it, although this effect is only valid for spring and summer.

Overall, the block-based analysis reveals a smaller number of significant variables compared to the grid-based counterpart. This may be due to local variations in LST attributable to urban form geometry, which are likely averaged out with block-level calculations that are better suited to identify the net effect of urban form.

5. Discussion

6. Conclusions

References

1. Adilkhanova, I.; Santamouris, M.; Yun, G.Y. Coupling urban climate modeling and city-scale building energy simulations with the statistical analysis: Climate and energy implications of high albedo materials in Seoul. Energy Build. 2023, 290, 113092. [Google Scholar] [CrossRef]
2. Alavipanah, S.; Schreyer, J.; Haase, D.; Lakes, T.; Qureshi, S. The effect of multi-dimensional indicators on urban thermal conditions. J. Clean. Prod. 2018, 177, 115–123. [Google Scholar] [CrossRef]
3. Anselin, L. Exploring spatial data with GeoDaTM: A workbook. Center for spatially integrated social science 2005, 165–223. [Google Scholar]
4. Baek, J.; Kim, S.; Kang, D.; Ko, A.; Aida, A.; Choi, J.; Park, C. Prediction of Micro-climate Impact by Floor Height Change Scenarios in Housing Renewal Site: Focusing on the Temperature, Particulate Matter (PM10), Fine Particulate Matter (PM2. 5). Journal of Korea Planning Association 2022, 57, 124–137. [Google Scholar] [CrossRef]
5. Boeing, G. Measuring the complexity of urban form and design. Urban Des. Int. 2018, 23, 281–292. [Google Scholar] [CrossRef]
6. Cai, Z.; Han, G.; Chen, M. Do water bodies play an important role in the relationship between urban form and land surface temperature? Sustainable Cities and Society 2018, 39, 487–498. [Google Scholar]
7. Chen, Y.; Yang, J.; Yu, W.; Ren, J.; Xiao, X.; Xia, J.C. Relationship between urban spatial form and seasonal land surface temperature under different grid scales. Sustain. Cities Soc. 2023, 89, 104374. [Google Scholar] [CrossRef]
8. Choi, Y.; Kim, J. & Lim, U. An analysis on the spatial patterns of heat wave vulnerable areas and adaptive capacity vulnerable areas in Seoul. J. Korea Plan. Assoc. 2018, 53, 87–107. [Google Scholar] [CrossRef]
9. Chun, B.; Guldmann, J.M. Impact of greening on the urban heat island: Seasonal variations and mitigation strategies. Computers, Environment and Urban Systems 2018, 71, 165–176. [Google Scholar] [CrossRef]
10. Cureau, R.J.; Pigliautile, I. and Pisello, A.L. Seasonal and diurnal variability of a water body's effects on the urban microclimate in a coastal city in Italy. Urban Clim. 2023, 49, 101437. [Google Scholar] [CrossRef]
11. Dai, Z.; Guldmann, J.M.; Hu, Y. Spatial regression models of park and land-use impacts on the urban heat island in central Beijing. Sci. Total Environ. 2018, 626, 1136–1147. [Google Scholar] [CrossRef]
12. Fan, C.; Que, X.; Wang, Z.; Ma, X. Land Cover Impacts on Surface Temperatures: Evaluation and Application of a Novel Spatiotemporal Weighted Regression Approach. ISPRS Int. J. Geo-Inf. 2023, 12, 151. [Google Scholar] [CrossRef]
13. Firozjaei, M.K.; Sedighi, A.; Mijani, N.; Kazemi, Y.; Amiraslani, F. Seasonal and daily effects of the sea on the surface urban heat island intensity: A case study of cities in the Caspian Sea Plain. Urban Clim. 2023, 51, 101603. [Google Scholar] [CrossRef]
14. Ge, X.; Mauree, D.; Castello, R.; Scartezzini, J.L. Spatio-temporal relationship between land cover and land surface temperature in urban areas: A case study in Geneva and Paris. ISPRS Int. J. Geo-Inf. 2020, 9, 593. [Google Scholar] [CrossRef]
15. Guo, F.; Schlink, U.; Wu, W.; Hu, D.; Sun, J. (2023). Scale-dependent and season-dependent impacts of 2D/3D building morphology on land surface temperature. Sustainable Cities and Society, 104788.
16. Han, D.; An, H.; Wang, F.; Xu, X.; Qiao, Z.; Wang, M. . & Liu, Y. Understanding seasonal contributions of urban morphology to thermal environment based on boosted regression tree approach. Build. Environ. 2022, 226, 109770. [Google Scholar]
17. Han, S.; Hou, H.; Estoque, R.C.; Zheng, Y.; Shen, C.; Murayama, Y.; Pan, J.; Wang, B. and Hu, T. Seasonal effects of urban morphology on land surface temperature in a three-dimensional perspective: A case study in Hangzhou, China. Build. Environ. 2023, 228, 109913. [Google Scholar] [CrossRef]
18. HosseiniHaghighi, S.; Izadi, F.; Padsala, R.; Eicker, U. Using climate-sensitive 3D city modeling to analyze outdoor thermal comfort in urban areas. ISPRS Int. J. Geo-Inf. 2020, 9, 688. [Google Scholar] [CrossRef]
19. Hu, Y.; Dai, Z.; Guldmann, J.M. Modeling the impact of 2D/3D urban indicators on the urban heat island over different seasons: A boosted regression tree approach. J. Environ. Manag. 2020, 266, 110424. [Google Scholar] [CrossRef]
20. Hwang, R.L.; Lin, T.P.; Matzarakis, A. Seasonal effects of urban street shading on long-term outdoor thermal comfort. Build. Environ. 2011, 46, 863–870. [Google Scholar] [CrossRef]
21. Kafy, A.A.; Shuvo, R.M.; Naim, M.N.H.; Sikdar, M.S.; Chowdhury, R.R.; Islam, M.A.; Sarker, M.H.S.; Khan, M.H.H. and Kona, M.A. Remote sensing approach to simulate the land use/land cover and seasonal land surface temperature change using machine learning algorithms in a fastest-growing megacity of Bangladesh. Remote Sens. Appl. : Soc. Environ. 2021, 21, 100463. [Google Scholar]
22. Kang. C-D. Effects of urban form indicators on land prices in Seoul, Republic of Korea: An urban morphometric approach. J. Real Estate Anal. 2022, 8, 73–101. [Google Scholar] [CrossRef]
23. Kim, M.; Kwon, I.; Yoo, S. A Study on the Typological Characteristics of Deteriorated Low-rise Residential Areas in Seoul. J. Korea Plan. Assoc. 2022, 57, 5–25. [Google Scholar] [CrossRef]
24. KMA 2022 Extreme climate report. Korea Meteorological Administration (KMA): Daejeon, Korea, 2020.
25. Li, H.; Li, Y.; Wang, T.; Wang, Z.; Gao, M.; Shen, H. (2021a). Quantifying 3D building form effects on urban land surface temperature and modeling seasonal correlation patterns. Build. Environ. 2023, 204, 108132. [Google Scholar] [CrossRef]
26. Li, T.; Xu, Y.; Yao, L. Detecting urban landscape factors controlling seasonal land surface temperature: From the perspective of urban function zones. Environ. Sci. Pollut. Res. 2021, 28, 41191–41206. [Google Scholar] [CrossRef] [PubMed]
27. Liang, S. Narrowband to broadband conversions of land surface albedo I: Algorithms. Remote sensing of environment 2001, 76, 213–238. [Google Scholar] [CrossRef]
28. Liu, X.; Ming, Y.; Liu, Y.; Yue, W.; Han, G. Influences of landform and urban form factors on urban heat island: Comparative case study between Chengdu and Chongqing. Sci. Total Environ. 2022, 820, 153395. [Google Scholar] [CrossRef] [PubMed]
29. Liu, B.; Guo, X.; Jiang, J. How Urban Morphology Relates to the Urban Heat Island Effect: A Multi-Indicator Study. Sustainability 2023, 15, 10787. [Google Scholar] [CrossRef]
30. Moazzam, M.F. U.; Doh, Y.H.; Lee, B.G. Impact of urbanization on land surface temperature and surface urban heat Island using optical remote sensing data: A case study of Jeju Island, Republic of Korea. Build. Environ. 2022, 222, 109368. [Google Scholar] [CrossRef]
31. Park, H.; Song, J. Relationship between Flood Damage and Flood Vulnerability Focusing on Property Damage and Human Casualties. J. Korea Plan. Assoc. 2023, 58, 149–166. [Google Scholar] [CrossRef]
32. Park, Y.; Guldmann, J.M.; Liu, D. Impacts of tree and building shades on the urban heat island: Combining remote sensing, 3D digital city and spatial regression approaches. Computers, Environment and Urban Systems 2021, 88, 101655. [Google Scholar] [CrossRef]
33. Park, Y.; Zhao, Q.; Guldmann, J.M.; Wentz, E. (2022). Quantifying the Cumulative Cooling Effects of 3D Building and Tree Shades with High Resolution Thermal Imagery in a Hot Arid Urban Climate. Preprint.
34. Peng, J.; Jia, J.; Liu, Y.; Li, H.; Wu, J. Seasonal contrast of the dominant factors for spatial distribution of land surface temperature in urban areas. Remote Sens. Environ. 2018, 215, 255–267. [Google Scholar] [CrossRef]
35. Qiao, Z.; Tian, G.; Xiao, L. Diurnal and seasonal impacts of urbanization on the urban thermal environment: A case study of Beijing using MODIS data. ISPRS J. Photogramm. Remote Sens. 2013, 85, 93–101. [Google Scholar] [CrossRef]
36. Shao, L.; Liao, W.; Li, P.; Luo, M.; Xiong, X.; Liu, X. Drivers of global surface urban heat islands: Surface property, climate background, and 2D/3D urban morphologies. Building and Environment 2023, 110581. [Google Scholar] [CrossRef]
37. Tang, L.; Zhan, Q.; Fan, Y.; Liu, H. and Fan, Z. Exploring the impacts of greenspace spatial patterns on land surface temperature across different urban functional zones: A case study in Wuhan metropolitan area, China. Ecol. Indic. 2023, 146, 109787. [Google Scholar] [CrossRef]
38. Turner, K.V.; Rogers, M.L.; Zhang, Y.; Middel, A.; Schneider, F.A.; Ocón, J.P. . &Dialesandro, J. More than surface temperature: Mitigating thermal exposure in hyper-local land system. J. Land Use Sci. 2022, 17, 79–99. [Google Scholar]
39. Wang, C.; Li, Y.; Myint, S.W.; Zhao, Q.; Wentz, E.A. Impacts of spatial clustering of urban land cover on land surface temperature across Köppen climate zones in the contiguous United States. Landsc. Urban Plan. 2019, 192, 103668. [Google Scholar] [CrossRef]
40. World Health Organization. (2014). Quantitative risk assessment of the effects of climate change on selected causes of death, 2030s and 2050s. World Health Organization.
41. Wu, W.; Li, L.; Li, C. Seasonal variation in the effects of urban environmental factors on land surface temperature in a winter city. J. Clean. Prod. 2021, 299, 126897. [Google Scholar] [CrossRef]
42. Yang, J.; Shi, Q.; Menenti, M.; Wong, M.S.; Wu, Z.; Zhao, Q.; Abbas, S. and Xu, Y. Observing the impact of urban morphology and building geometry on thermal environment by high spatial resolution thermal images. Urban Clim. 2021, 39, 100937. [Google Scholar] [CrossRef]
43. Yao, L.; Li, T.; Xu, M.; Xu, Y. How the landscape features of urban green space impact seasonal land surface temperatures at a city-block-scale: An urban heat island study in Beijing, China. Urban For. Urban Green. 2020, 52, 126704. [Google Scholar] [CrossRef]
44. Yu, K.; Chen, Y.; Wang, D.; Chen, Z.; Gong, A.; Li, J. Study of the seasonal effect of building shadows on urban land surface temperatures based on remote sensing data. remote sensing 2019, 11, 497. [Google Scholar] [CrossRef]
45. Zhang, Z.; Luan, W.; Yang, J.; Guo, A.; Su, M.; Tian, C. The influences of 2D/3D urban morphology on land surface temperature at the block scale in Chinese megacities. Urban Clim. 2023, 49, 101553. [Google Scholar] [CrossRef]
46. Zhao, L.; Li, T.; Przybysz, A.; Liu, H.; Zhang, B.; An, W. and Zhu, C. Effects of urban lakes and neighbouring green spaces on air temperature and humidity and seasonal variabilities. Sustain. Cities Soc. 2023, 91, 104438. [Google Scholar] [CrossRef]
47. Zhao, Y.; Sen, S.; Susca, T.; Iaria, J.; Kubilay, A.; Gunawarden, K.; Zhou, X.; Takane, Y.; Park, Y.; Wang, X. and Rubin, A. Beating the Heat: Solution Sets for Developed Cities. 2023, Preprint.
48. Zhou, W.; Cao, W.; Wu, T.; Zhang, T. The win-win interaction between integrated blue and green space on urban cooling. Sci. Total Environ. 2023, 863, 160712. [Google Scholar] [CrossRef]
49. Zhu, Z.; Shen, Y.; Fu, W.; Zheng, D.; Huang, P.; Li, J.; Lan, Y.; Chen, Z.; Liu, Q.; Xu, X. and Yao, X. How does 2D and 3D of urban morphology affect the seasonal land surface temperature in Island City? A block-scale perspective. Ecol. Indic. 2023, 150, 110221. [Google Scholar] [CrossRef] [PubMed]