Urban heat island (UHI) refers to elevated temperatures in urban areas compared to their rural surroundings. UHI is believed to be primarily caused by pollutants and heat sources emitted during the combustion of industrial and transportation processes (Ulpiani, 2021). In recent years, industrial complexes that emit substantial quantities of these pollutants have garnered attention due to their role as notable contributors to UHI. These industrial complexes are sites where many environmental problems occur concurrently, including water quality degradation due to land development and industrial wastewater and air pollution resulting from material combustion (Ma, 2022). Research has demonstrated that the buildings constituting industrial complexes exert a significant influence on UHI (Khamchiangta and Dhakal, 2019), while the pollutants emanating from these complexes are found to have detrimental effects on urban ecosystems and the health of citizens (Eom et al., 2018). Furthermore, the various environmental problems arising in industrial complexes interact with UHI in a chain reaction, ultimately resulting in amplified environmental challenges. In response, the Korean government is implementing policies, such as the Smart Green Industrial Complex Project, to reduce pollutant emissions by upgrading aging industrial complexes (Government 24, 2022). However, research on the specific components of industrial complexes contributing to UHI is still insufficient.

In the domain of remote sensing, there has been long-standing research in which land surface temperature (LST) is extracted from satellite imagery, such as Landsat, ASTER, and MODIS (Zargari et al., 2024). These studies seek to analyze the spatial patterns of UHI and explore mitigation strategies. However, the majority of these studies have primarily focused on urban areas, while studies specifically examining industrial complexes remain scarce. In recent years, some studies have defined the phenomenon of industrial complexes exhibiting higher temperatures than the surrounding areas as an industrial heat island (IHI) (Gao et al., 2022; Meng et al., 2022), which is classified as a subtype of the UHI. However, studies on the IHI have the following limitations.

According to Oke et al. (2017), urban climates such as UHI are classified into four types according to the spatial scale, as illustrated in Fig. 1. While UHI is studied at the mesoscale (10–100 km) based on a city, IHI is studied at the local scale (0.1–10 km) mainly for individual industrial complexes. This disparity in scale between these two research domains makes it difficult to connect the findings of IHI studies with the results of existing UHI-related research. Furthermore, there is a lack of systematic evaluations to assess whether IHIs are present at a sufficient scale at the city level (mesoscale) to justify their classification as a separate research topic. Therefore, the current approach of focusing only on local scale analysis makes it difficult to consider UHI and IHI comprehensively in policy making.

Figure 1. Horizontal ranges of urban climate scales based on the concept proposed by Oke et al. (2017).

Industrial complexes represent pivotal regions that contribute significantly to UHI, thus necessitating a novel approach to address the limitations of existing research. For instance, given that IHI is a type of UHI, it is possible to initially ascertain whether industrial complexes exhibit elevated UHI relative to non-industrial complexes at the mesoscale, constituting the unit of analysis in existing UHI research. Following this assessment, it will be feasible to proceed with the analysis at the local scale, depending on whether IHI is present at an adequate scale to be defined as an independent phenomenon at the mesoscale.

This hierarchical approach to analyzing IHI at two scales will not only enable the connection with UHI research but also provide a detailed understanding of the spatial patterns of IHI in each industrial complex. Hence, this study aims to present a systematic methodological framework for analyzing IHI to strengthen its connection with UHI and identify the influence of various factors that constitute an IHI. To this end, the methodology is structured into two primary steps. First, the spatial patterns of IHI are analyzed based on the hierarchical approach. Second, the impact of various factors that comprise industrial complexes on IHI is analyzed through spatial regression models.

The structure of this paper is as follows. Section 2 discusses the current status of UHI research. Section 3 meticulously delineates the methodological framework sequentially to systematically analyze IHI, while Section 4 demonstrates its practical application in the context of Incheon Metropolitan City, Korea. In this section, the performance of the spatial regression model is evaluated using several indicators.

2.1. Studies on Spatial Characteristics of UHI and IHI

UHI refers to a phenomenon where the temperature of an urban area is higher than the surrounding area, and research has been continuously conducted to identify it and measure its intensity. In general, UHI identification methods are divided into two approaches. The first uses the difference in absolute temperature, and the second involves defining the surrounding area to measure the difference in relative temperature (Kim and An, 2017).

Absolute temperature methods compare observed values within a study area to identify regions with higher temperatures as UHI. As temperature values vary by region and season, various approaches have been proposed to establish universally applicable thresholds. The most common methods include the grouping of areas exceeding the average LST value of the study area into temperature ranges of 1–7°C (Martin et al., 2015) and classifying regions using standard deviations from the average LST value of the study area (El-Hattab et al., 2018).

Relative temperature methods identify UHI and quantitatively measure its intensity by calculating the difference in LST between urban and surrounding areas (Peng et al., 2012). However, due to the absence of a definitive definition for surrounding areas, various approaches have been proposed to address this ambiguity (Meng et al., 2022). Representative methods include utilizing the local climate zones (LCZ) system to define LCZ D (Low plants) areas as the surrounding region (Stewart and Oke, 2012), using land cover data derived from satellite imagery to classify non-urban areas as surrounding regions (He et al., 2007), and forming buffers at regular intervals based on urban areas to delineate the surrounding region (Peng et al., 2012). Studies have also utilized heat island potential (HIP), which is defined as the average difference between air temperature and surface temperature, to identify areas with high potential for UHI formation (Ahn et al., 2012). Additionally, hotspot analyses using Moran’s I and Getis-Ord Gi* have been conducted to examine areas of significant UHI intensity (Zargari et al., 2024).

Moreover, a methodology for analyzing the spatial characteristics of IHI is also proposed. Previous studies (Gao et al., 2022; Meng et al., 2022) defined surrounding areas by creating buffers at predefined intervals around industrial complexes. The relative temperature difference between the industrial complexes and their surrounding regions was then utilized to identify IHIs and measure their intensity. However, analyzing IHI only at the local scale without prior identification at the mesoscale may result in it being misinterpreted as a phenomenon distinct from UHI. Furthermore, this approach also risks misinterpreting local phenomena as an overarching pattern. As a result, this study first identifies the potential existence of IHI at the mesoscale and subsequently conducts further analysis at the local scale, focusing only on IHI regions. For the local scale analysis, a buffer-based method that was used in previous studies was applied.

2.2. Studies on Factors Influencing UHI

Numerous studies have analyzed the impact of urban factors on UHI to mitigate UHI. These factors are broadly categorized into physical and socioeconomic environments, with most research focusing on the influence of physical environmental factors. Specifically, land cover has been mainly selected as a key physical environmental factor (Chen et al., 2017). To analyze land cover, researchers have employed normalized indices derived from satellite imagery, including the normalized difference vegetation index (NDVI), normalized difference water index (NDWI), and normalized difference built-up index (NDBI) (Ogashawara and Bastos, 2012). Since physical environmental factors are highly variable and challenging to quantify, these normalized indices were used to represent land cover characteristics (Kim et al., 2015). Furthermore, other indices, such as the enhanced vegetation index (EVI), soil-adjusted vegetation index (SAVI), index-based built-up Index (IBI), and normalized difference bareness index (NDBal), have also been utilized for similar purposes. Also, the topographic and meteorological factors have been analyzed using variables such as elevation, slope direction, solar radiation (Taripanah and Ranjbar, 2021), and air pollutants (Fuladlu and Altan, 2021). More recently, research has incorporated variables such as sky view factor and surface roughness to reflect the threedimensional characteristics of urban environments (Li et al., 2019).

Studies analyzing the influence of the physical environment have focused on establishing mathematical models to elucidate the physical mechanisms of UHI. In contrast, research examining the impact of socioeconomic environments has concentrated on investigating the spatial inequities of UHI and exploring policy measures aimed at alleviating the social burdens experienced by residents in vulnerable areas. For instance, a study has included factors that account for socioeconomic vulnerabilities, such as the proportion of individuals receiving basic livelihood assistance, older adults living alone, children under five, and individuals aged 65 or older (Cho et al., 2019). Additionally, other studies have employed the socioeconomic indexes for areas (SEIFA), an index that measures socioeconomic status, income, employment, education, and housing levels (Sidiqui et al., 2022).

While these studies have enhanced the understanding of factors influencing UHI formation, they primarily adopted variables representative of urban environments, making them insufficient for capturing the unique characteristics of industrial complexes. Additionally, they have predominantly analyzed physical and socioeconomic factors separately. Accordingly, this study considers both physical and socioeconomic environments and selects variables that specifically reflect the characteristics of industrial complexes.

2.3. Studies on Methods for Factor Identification

A quantitative analysis of the relationship between UHI and its influencing factors is categorized into experimental and empirical research (Kim and Brown, 2021). Experimental research explores UHI interactions to predict future change so that they can be effectively applied in urban planning and urban environments (Lui et al., 2020). However, the simulation models in experimental research face notable limitations, such as high modeling costs and challenges in handling multi-scale spatiotemporal variations (Mirzaei and Haghighat, 2010).

Empirical research addresses economic constraints by utilizing statistical models to analyze UHI factors and develop statistical frameworks. Most empirical research has relied on global regression models, such as ordinary least squares (OLS) and neural network models (Luo and Peng, 2016). However, these models inherently assume independence, normality, and homoscedasticity of residuals, limiting their ability to capture spatial effects and explain local variability in spatial data (Brown et al., 2012).

To address these limitations, spatial regression models that account for spatial effects have been increasingly adopted. GWR accounts for spatial non-stationarity by modeling the relationship between dependent and independent variables that vary geographically (Jung and Lee, 2015). Unlike global regression models, which assume a uniform relationship across space, GWR allows for the identification of localized variations in influencing factors. Specifically, given that the underlying natural processes affecting LST and its spatial patterns may operate at different spatial scales (Li et al., 2010), it is essential to consider spatial non-stationarity in IHI research. Consequently, this study employs GWR to spatially analyze the impact of factors constituting industrial complexes on IHI and to estimate the local causes of IHI. By incorporating spatial non-stationarity, this approach improves the representation of the spatial distribution of IHI drivers, which cannot be effectively modeled using traditional global regression techniques.

This section delineates the methodological framework sequentially to analyze IHI. Fig. 2 illustrates the framework proposed in this study, composed of two major components.

Figure 2. Geospatial framework for industrial heat island (IHI) analysis.

In the initial step, the hierarchical approach is introduced to analyze the spatial characteristics of IHI at two scales. At the mesoscale, this study employs UHI regions defined using LST calculated from satellite imagery. Based on UHI regions, the Chi- Square test of independence is performed to identify the spatial distribution differences in UHI between industrial and nonindustrial areas. Suppose a statistically significant difference is identified between the two groups. In that case, the IHI can be considered at an adequate scale to be defined as an independent phenomenon at the city level. The IHI verified at the mesoscale is then analyzed at the local scale with the LST profile method (Meng et al., 2022) to measure its spatial extent and intensity.

In instances where evidence supporting the IHI’s possession of an adequate scale is lacking at any stage of the analysis, the likelihood of the IHI’s existence is considered low, and further analysis is not pursued. The second step in the methodological approach involves the analysis of the influence of factors representing the characteristics of industrial complexes on IHI. This analysis utilizes the GWR model. The dependent variable in the GWR model is the LST, which is calculated in the first step. The independent variables consist of seven factors that have been identified from existing research as representative of industrial complexes. The GWR model is subsequently employed to quantify the influence of these factors on IHI intensity.

3.1. Methods for Analyzing Spatial Characteristics of IHI Based on Hierarchical Approach

3.1.1. Calculating LST and Defining UHI

LST is calculated using Landsat8 OLI/TIRS imagery provided by the United States Geological Survey (USGS). Landsat-8 OLI/TIRS imagery is widely utilized in research due to its improvements in resolution and reliability (Kim et al., 2018). In addition, its observation time, 0220 universal time coordinated (UTC), is appropriate for observing the daily average LST (Jee et al., 2016). This study applied the conversion formula presented by Ihlen and Zanter (2019) (Eq. 1) to transform the image values stored as digital number (DN) into the values of the spectral radiance.

where L_λ: Spectral radiance (W/(m²·sr·μm)), ML: Radiance multiplicative scaling factor for the band, A_L: Radiance additive scaling factor for the band, Q_cal: Level 1 pixel value in DN.

The calculation of LST from spectral radiance values is demonstrated in Eq. (2).

where T_B: Top of atmosphere brightness temperature (K), L_λ: TOA spectral radiance, K₁: Band-specific thermal conversion constant from the metadata, K₂: Band-specific thermal conversion constant from the metadata.

The calculated LST (T_B) is subject to limitations in adequately reflecting the unique characteristics of objects located on the land surface. Therefore, adjustments considering the Land Surface Emissivity (LSE) are required to obtain a more accurate LST (Kim et al., 2014). However, measuring LSE over large areas regularly is challenging due to the spatial variability of ground cover and the temporal dynamics of vegetation, snow cover, and soil moisture. To address this, this study employed the method proposed by Zhang et al. (2006) to estimate LSE using NDVI values, and LST is calculated by applying corrections as specified in Eq. (3).

where T_S: Land surface temperature (LST) after atmospheric and emissivity correction, T_B: Top of atmosphere brightness temperature (K), λ: Wavelength of the thermal infrared band used to retrieve spectral radiance, α: Constant, defined as α = hc/K≈ 1.438 × 10^–2) mk (where: h: Planck’s constant (6.626 × 10^–34) J · s), c: Speed of light (2.998 × 10⁸ m/s), K: Boltzmann constant (1.38 × 10^–23) J/K)), ε: Surface emissivity.

Table 1 . Normalized difference vegetation index (NDVI) ranges and corresponding emissivity values (Zhang et al. 2006).

NDVI ranges	Emissivity
NDVI < –0.185	0.995
–0.185 ≤ NDVI < 0.157	0.970
0.157 ≤ NDVI ≤ 0.727	1.0094 + 0.047 ln (NDVI)
NDVI > 0.727	0.990

The classification of UHI is based on the average LST value and standard deviation of the study area (Yang et al., 2017; El- Hattab et al., 2018). Setting the threshold using the standard deviation minimizes the potential for subjectivity, allowing for a more accurate identification of UHI regions. Since the range of LST varies by region and season, normalized LST (NLST) is calculated through min-max normalization (Eq. 4) and classified into seven levels, as presented in Table 2. The UHI is defined as high, higher, and highest.

Table 2 . The classification of normalized land surface temperature (NLST) zones based on the NLST range (Yang et al., 2017).

NLST grade	NLST rangea)
lowest	NLST < NLSTmean – 1.5S
lower	NLSTmean – 1.5S ≤ NLST < NLSTmean – 1.0S
low	NLSTmean – 1.0S ≤ NLST < NLSTmean – 0.5S
medium	NLSTmean – 0.5S ≤ NLST < NLSTmean + 0.5S
high	NLSTmean + 0.5S ≤ NLST < NLSTmean + 1.0S
higher	NLSTmean + 1.0S ≤ NLST < NLSTmean + 1.5S
highest	NLST ≥ NLSTmean + 1.5S

^a) NLST_mean and S respectively denote the mean value and standard deviation of NLST..

3.1.2. Mesoscale Analysis: Chi-Square Test of Independence

Before analyzing at the local scale, the potential existence of IHI at the mesoscale is examined using UHI. This step aims to evaluate whether IHI possesses an adequate scale to be classified as an independent phenomenon. Specifically, the area under investigation is divided into industrial and non-industrial zones. Pearson’s Chi-Square Test of Independence is conducted to compare the distribution of UHI between the two zones. The results of the Chi-Square Test are used to assess the likelihood of a relatively strong UHI in the industrial complex.

The Chi-Square Test of Independence is a non-parametric method used to evaluate the association between two categorical variables. It assesses the statistical significance of the differences between zones and the extent to which each category contributes to the differences (Franke et al., 2012). In this study, the row variable is a binary variable (1: industrial zone, 0: non-industrial zone) indicating the presence or absence of an industrial complex, and the column variable is a categorical variable with three levels (high, higher, highest) representing UHI areas and other areas (No UHI). The process of summing each row and column to calculate the expected frequencies for testing the independence between the two zones is described in Eq. (5) (McHugh, 2013).

where E_ij: Expected frequency in the ith row and jth column, Ri, Sum: Sum of the frequencies in the ith row, C_{j, Sum}: Sum of the frequencies in the jth columns, N: Overall total of all frequencies (grand total).

The expected frequency refers to the estimated frequency of each UHI class under the assumption that no difference exists based on the presence or absence of an industrial complex. The chi-squared test statistic, as shown in Eq. (6), is calculated using the difference between the expected frequency and the observed frequency.

where O_ij: Observed frequency in the ith row and jth column, E_ij: Expected frequency in the ith row and jth column, r: Number of rows, c: Number of columns, df: Degrees of freedom.

Suppose a statistically significant difference is observed in the UHI distribution between industrial and non-industrial zones (p<0.05). In that case, there is a statistically significant association between the industrial zones and the UHI classification. The strength of association is measured using Cramer’s V coefficient. The Cramer’s V coefficient ranges from 0 to 1, with the degree of association corresponding to the coefficient values presented in Table 3 (Akoglu, 2018).

Table 3 . Interpretation of Cramer’s V (Akoglu, 2018).

Cramer’s V range	Interpretation
0.25 ≤ Cramer’s V < 1.00	Very strong
0.15 ≤ Cramer’s V < 0.25	Strong
0.10 ≤ Cramer’s V < 0.15	Moderate
0.05 ≤ Cramer’s V < 0.10	Weak
0 ≤ Cramer’s V < 0.05	No or very weak

If the Cramer’s V coefficient exceeds 0.15, it indicates a statistically significant (p < 0.05) and strong association (strong or very strong) between the industrial zones and the UHI. This suggests that the IHI is potentially present at the mesoscale, making it appropriate for measuring the intensity of IHI at the local scale. Conversely, if no statistically significant difference (p ≥ 0.05) is observed or if the Cramer’s V coefficient is below 0.15, the likelihood of IHI existing at the mesoscale is extremely low. This is because there is verified independence between industrial complexes and UHI. Accordingly, this study conducts local scale analysis only if a statistically significant association (p < 0.05) and strong association (Cramer’s V≥ 0.15) are identified at the mesoscale.

3.1.3. Local Scale Analysis: LST Profile

At the local scale, the IHI for each industrial complex is analyzed using the LST Profile method (Meng et al., 2022), which quantitatively describes and tracks the LST of industrial complexes and surrounding areas. Specifically, buffers with predefined intervals are generated around each industrial complex, and the average LST within each buffer zone is calculated. In previous studies, the buffer radius and interval have been determined differently depending on the region. Subsequently, LST profile graphs are generated to depict the variation in LST with increasing distance from the industrial complex, as illustrated in Fig. 3. LST_{Industrialpark} denotes the average LST of the industrial complex, while LST_Turningpoint denotes the average LST of the buffer corresponding to the first extreme value.

Figure 3. Land surface temperature (LST) profiles and slope analysis for IHI Identification adapted from Meng et al. (2022). Case of (a) negative slope and (b) flat slope.

If the slope between these two points is negative (f(x) < 0), as shown in case (a), the temperature decreases as the distance from the industrial complex increases, thereby confirming the presence of IHI. Conversely, if the slope is positive or flat (f(x) ≥ 0), as shown in case (b), the temperature increases or remains constant as the distance from the industrial complex increases, thereby indicating the absence of IHI. In this study, the range up to the buffer corresponding to the first extreme value is defined as the IHI occurrence area when IHI is identified. Furthermore, the average LST of the industrial complex (LST_Turningpoint) is defined as the maximum value of the IHI, while the first extreme value (LST_Turningpoint) is defined as the minimum value. The difference between these two values is utilized as an indicator to quantify the strength of the IHI.

3.2. Methods for Analyzing Correlation Between IHI and Industrial Factors

3.2.1. GWR Model Specification

This study employs the GWR model to examine the correlation between IHI and various factors constituting industrial complexes. The spatial scope of the analysis is defined as the IHI-affected areas identified in the preceding step. Unlike OLS, which captures only global trends, the GWR model accounts for spatial heterogeneity, allowing relationships to vary across geographic locations. In other words, it enables spatial analysis of the differing effects of industrial factors on each industrial complex. The basic formula of the GWR model, which uses industrial factors (x_ik) to explain LST (y_i) that varies with geographic location i, is presented in Eq. (7) (Fotheringham et al., 2002).

where y_i: Dependent variable at location i, x_ik: kth independent variable at location i, m: Number of Independent variables, β_i0: Intercept parameter at location i, β_ik: Local regression coefficient for the kth independent variable at location i, εi: Random error at location i.

The GWR model accounts for spatial dependence by assigning weights based on proximity to geographic locations while estimating regression coefficients (Li et al., 2010). Spatial dependence, a concept defined by Tobler’s “First Law of Geography,” refers to the property of spatial objects whereby closer objects are more closely related than those farther apart (Tobler, 1970). Accordingly, greater weight is assigned to observations nearer to the ith location, while lesser weight is given to those farther away.

Traditional GWR determines these weights using a kernel function, where the bandwidth of the kernel function is optimized through leave-one-out cross-validation (LOOCV) (Murakami et al., 2020). This process iteratively adjusts the matrix to achieve an optimal bandwidth. However, as dataset size increases, computational complexity grows exponentially and presents challenges for processing large datasets. To address these computational constraints and facilitate scalable analysis across a range of urban areas of disparate sizes, this study utilizes the Scalable GWR algorithm proposed by Murakami et al. (2020).

The Scalable GWR algorithm mitigates the limitations of traditional GWR by replacing the bandwidth with the number of neighbors (k-nearest neighbors) and calculating weights using a linear polynomial kernel. This method does not require parallel processing, thereby significantly improving computational efficiency and accuracy for large datasets (N> 3,000). In this study, the number of neighbors was determined using the corrected Akaike information criterion (AICc) (Akaike, 2011), which is a relative performance evaluation metric that minimizes the difference between observed and estimated values while considering model fit.

3.2.2. Variable Selection and Data Processing

The GWR model’s dependent variable is LST, while the independent variables are categorized into three groups, namely industrial activity factors, industrial space factors, and industrial workforce factors. These groups are derived from existing research and the three core components of industrial complexes, which are industry, space, and people (Korea Industrial Complex Corporation, 2025).

The air pollutants NO₂ and SO₂ are selected as industrial activity factors. These primary pollutants are generated during the combustion of fossil fuels and contribute to the formation of secondary pollutants, including PM and O (Hwang et al., 2020). As Cao et al. (2016) have noted, the correlation between LST and air pollutants is influenced by regional meteorological conditions and atmospheric properties, which necessitates interpretation based on actual analytical results. Accordingly, this study utilizes air environment statistics from public data platforms to develop NO₂ and SO₂ variables. Since the data is collected at fixed points, interpolation is conducted using Ordinary Kriging. Ordinary Kriging calculates a weighted average of observed values from surrounding stations to predict values at unobserved locations. The basic formula for Ordinary Kriging is in Eq. (9), as described by Srivastava (1989) and cited in Gia Pham et al. (2019).

where, Z * (x₀): Predicted value at the unmeasured position x₀, Z(x_i): Measured value at position x_i, λ_i: Weighting coefficient from the measured position to x₀, n: Number of positions within the neighborhood searching.

Elevation and land cover are selected as industrial space factors. Elevation is developed based on existing studies using the 1:5000 digital topographic map provided by the National Geographic Information Institute (NGII). The Digital Elevation Model (DEM) is generated by extracting contour lines from the digital topographic map, creating a triangulated irregular network (TIN), and converting the TIN into raster data. According to Woo et al. (2001), a spatial resolution of 10 m is most suitable for mountainous and hilly areas with significant elevation differences, while a resolution of 30 m is recommended for urban and agricultural areas with minimal elevation differences. Thus, this study adopts a spatial resolution of 30 m, as it concentrates on urban areas where industrial complexes are located.

Additionally, the soil-adjusted vegetation index (SAVI), normalized difference built-up index (NDBI), and automated water extraction index (AWEI) are selected as industrial space factors to represent land cover in urban areas where industrial complexes are located. SAVI, an index that modifies NDVI, minimizes the influence of soil on the primary spectral response (Huete, 1988) and provides more accurate results in regions with low vegetation density. NDBI is a widely used index for identifying urban and built-up areas (Guha et al., 2018). AWEI, an index for detecting water bodies, demonstrates higher accuracy and stability compared to traditional indices like MNDWI in urban areas with man-made surfaces (Feyisa et al., 2010). Although AWEIsh effectively removes shadows, it introduces additional noise, which can result in the omission or overestimation of water bodies (Senel et al., 2020). Thus, selecting the appropriate type of AWEI requires consideration of regional characteristics. The process for calculating each normalized index is shown in Table 4.

Table 4 . Equations used for SAVI, NDBI, and AWEI calculation.

Index name and abbreviation	Formula for Landsat-8a)	Reference
Soil-adjusted vegetation index (SAVI)	SAVI=NIR−REDNIR−RED+0.5⋅1.5	Huete (1988)
Normalized difference built-up index (NDBI)	NDBI=SWIR−NIRSWIR+NIR	Zha et al. (2003)
Automated water extraction index (AWEI)	AWEInsh=4⋅GREEN−SWIR1−0.25⋅NIR+2.75⋅SWIR2AWEIsh=BLUE+2.5⋅GREEN−1.5⋅NIR+SWIR1−.025⋅SWIR2	Feyisa et al. (2010)

^a) In Landsat 8, NIR (Band 5), RED (Band 4), GREEN (Band 3), BLUE (Band 2), SWIR1 (Band 6), and SWIR2 (Band 7) are used for SAVI, NDBI, and AWEI calculations..

The number of workers is selected as the industrial workforce factor. Studies analyzing the correlation between population density and UHI indicate that higher active population density is associated with greater UHI intensity, with strong UHI particularly observed in urban centers and commercial or industrial zones (Cui and Shi, 2012). Based on this observation, this study identifies workers who remain in industrial complexes during the daytime and engage in industrial activities as the industrial workforce factor. Worker statistics provided by Statistics Korea are used to quantify the number of workers. Finally, the selected independent variables are summarized in Table 5.

Table 5 . Summary of dependent and independent variables.

Type			Variable	Unit
Dependent variable	Land surface temperature			C
Independent variable	Industrial activity factor	Major air pollution	Nitrogen Dioxide (NO2)	μg/m3
			Sulfur Dioxide (SO2)	μg/m3
	Industrial space factor	Digital Elevation Model (DEM)		m
		Land cover	Soil Adjusted Vegetation Index (SAVI)	None (index)
			Normalized Difference Built-up Index (NDBI)	None (index)
			Automated Water Extraction Index non-shadow (AWEInsh)	None (index)
	Industrial workforce factor	Workers		Number of people

4.1. Study Area

This study’s methodology, which analyzes the spatial characteristics of IHI based on a hierarchical approach and examines its influencing factors using the GWR model, was applied to the inland region of Incheon Metropolitan City in northwestern South Korea. Incheon is one of Korea’s major industrial cities. Thus, its lowlands are primarily allocated for industrial complexes and residential areas (Kong et al., 2016). According to Korea Industrial Complex Corporation (2025), this area includes 12 industrial complexes, which cover a diverse range of industrial types. In addition, it contains aging national industrial complexes, including Namdong, Bupyeong, and Juan, which have been in operation for over 40 years. These industrial complexes are a focus of governmental attention and require significant improvements. Given these characteristics, Incheon offers a suitable context for a detailed assessment of IHI patterns and influencing factors due to its diversity in industrial complexes.

In this study, geographically adjacent industrial complexes were consolidated and classified into nine industrial complexes for the unbiased measurement of IHI at the local scale (Fig. 4). Table 6 shows the establishment year, area, and type of these nine industrial complexes. Due to differences in size and shape between industrial complexes and administrative divisions, the national standard grid (100 × 100 m) provided by the NGII was adopted as the spatial unit to consider the geographical location of the industrial complex. The study area comprised a total of 38,838 grids.

Figure 4. Spatial distribution of industrial complexes within study area, Incheon.

Table 6 . Summary of industrial complexes within the study area.

Industrial complex		Date of designation	Area (m2)		Type
					National	General	Urban high-tech
(A) Beautiful Park		2006-12-26	2,250,719			○
(B) Incheon Western District Industrial Park		1992-07-29	994,661			○
(C) Cheongna District 1 General Industrial Park		1965-06-16	194,361			○
(D) IHP Urban High-Tech Industrial Park		1997-08-06	495,144				○
(E) Bupyeong Industrial Park		2015-02-09	609,361		○
(F) Seowoon General Industrial Park		2015-06-01	524,970			○
(G) Juan	Juan Industrial Park	1969-08-05	2,663,283	1,176,829	○
	Incheon National Industrial Park	1973-04-01		1,136,269		○
	Incheon Machinery Industrial Park	1967-11-23		350,185		○
(H) Songdo Knowledge Information Industrial Park		2000-09-18	2,401,745				○
(I) Namdong Industrial Park		1980-09-02	9,504,046		○		○

4.2. Analysis of Spatial Characteristics of IHI

4.2.1. Analysis of LST Distribution

The LST was calculated using Landsat 8 OLI/TIRS imagery provided by the USGS. As LST serves as the foundational indicator for IHI analysis, images with minimal cloud cover were selected to enhance accuracy. For temporal analysis, one image per season was selected, resulting in a total of four images representing the four seasons. The acquisition dates of the selected images are April 2, June 21, October 27, and December 30, 2022, all with a Path/Row of 116/034.

Fig. 5 illustrates the seasonal distribution of Normalized LST (NLST) in Incheon. LST exhibits variations in range and distribution across seasons, even within the same region. To ensure objectivity in the seasonal comparison of UHI regions, NLST was derived by classifying LST into seven levels using Min-Max normalization. The top three levels (high, higher, and highest) were defined as UHI. As a result, UHI was most extensive in summer, followed by winter, fall, and spring in terms of intensity. Moreover, elevated UHI levels were observed at the core of most industrial complexes, except for IHP and Songdo, which are categorized as urban high-tech industrial complexes. The level of NLST exhibited a decreasing trend with increasing distance from the boundaries of the industrial complexes. Urban high-tech industrial complexes, where relatively low levels of NLST were observed, predominantly accommodate knowledge, culture, and information and communication industries. In contrast to manufacturing-focused industrial complexes, these are interpreted as having relatively low NLST levels due to their industrial structures, which generate fewer artificial heat sources.

Figure 5. Spatial distribution of normalized LST (NLST) in 2022.

4.2.2. Mesoscale Analysis

This study analyzed the relationship between industrial complexes and UHI levels at the mesoscale to evaluate whether IHI exists at an adequate scale to be classified as an independent phenomenon. Table 7 presents the results of the chi-square independence test conducted for the row variable (industrial zone: 1 for industrial zones, 0 for non-industrial zones) and the column variable (UHI levels: high, higher, highest, no UHI). Across all seasons, the pvalue was 2.2 × 10^–16 (p < 0.05), indicating a statistically significant association between industrial complexes in Incheon and UHI levels.

Table 7 . Summary of industrial complexes within the study area.

Seasons	Chi-squarea)	p-value	Cramer’s V
Spring (2022/04/02)	6682.805***	< 2.2e-16	0.415
Summer (2022/06/21)	7403.103***	< 2.2e-16	0.437
Fall (2022/10/27)	6428.851***	< 2.2e-16	0.407
Winter (2022/12/30)	1885.619***	< 2.2e-16	0.22

^a) Significance level: *p<0.05; **p<0.01; ***p<0.001..

This finding suggests that the presence of industrial complexes may influence the distribution of UHI levels. Specifically, the strength of the association between the two variables was measured by calculating the Cramer’s V coefficient. The results revealed a statistically significant and strong association between industrial complexes in Incheon and UHI levels across all four seasons (Cramer’s V ≥ 0.15). In other words, the analysis confirms that the distribution of UHI levels between industrial and non-industrial areas exhibits statistically significant differences, suggesting that the IHI potentially exists at the city level.

4.2.3. Local Scale Analysis

Following the confirmation of IHI at the mesoscale, the analysis was extended to investigate IHI generated by individual industrial complexes at the local scale. The range and intensity of IHI for each industrial complex were quantified using the LST profile method (Meng et al., 2022). Meng et al. (2022) set the buffer size and radius for LST profile analysis at 50 m and 5 km, respectively, as spatial units. However, the high density of industrial complexes in Incheon rendered these parameters unsuitable. Therefore, the buffer size was set to 30 m to match the resolution of Landsat 8 imagery, and the radius was set to half of the shortest distance to the nearest neighboring industrial complex. The findings of the IHI analysis under these conditions are summarized in Table 8.

Table 8 . Seasonal analysis of IHI by industrial complex.

Industrial Complex	Seasons
	Spring				Summer				Fall				Winter
	TIPa)	TTPb)	ΔTmaxc)	EBufferd)	TIP	TTP	ΔTmax	EBuffer	TIP	TTP	ΔTmax	EBuffer	TIP	TTP	ΔTmax	EBuffer
(A) Beautiful Park	20.69	17	3.69	180	29.15	25.37	3.78	240	18.03	15.76	2.27	240	-0.09	-1.78	1.69	180
(B) Western	22.78	17.85	4.93	450	30.55	25.59	4.96	390	19.88	16.73	3.15	180	1.98	-1.43	3.41	390
(C) Cheongna	19.14	19.05	0.09	270	29.81	28.83	0.98	330	16.92	17.13	-0.21	-	-0.56	0.17	-0.73	-
(D) IHP	19.99	19.79	0.2	210	30.52	30.42	0.1	180	18.08	18.30	-0.22	-	0.08	0.61	-0.53	-
(E) Bupyeong	20.13	16.94	3.19	180	28.53	23.53	5	180	18.71	16.48	2.23	180	1.32	0.41	0.91	270
(F) Seowoon	18.02	17.93	0.09	240	29.28	28.64	0.64	300	16.93	16.73	0.2	240	-0.71	-0.51	-0.2	-
(E) Juan	21.64	18.3	3.34	480	32.39	29.76	2.63	480	19.49	17	2.49	420	0.64	-1.14	1.78	420
(H) Songdo	20.94	17.74	3.2	210	31	27.77	3.23	210	19.05	16.57	2.48	210	-0.15	-1.34	1.19	180
(I) Namdong	17.46	16.67	0.79	210	27.21	26.37	0.84	210	16.31	15.71	0.6	210	-2.32	-2.53	0.21	180

^a) The mean land surface temperature measured within the industrial park (°C)..

^b) The mean land surface temperature of the buffer corresponding to the first extreme temperature value observed at the turning point (°C)..

^c) The intensity of the industrial heat island, calculated as the temperature difference between T_IP and T_TP (°C)..

^d) The spatial extent of the industrial heat island, defined as the distance from the industrial park to the first turning point (m)..

The intensity of the IHI (ΔT_max) is defined as the difference between the average LST of the industrial complex (T_IP) and the average LST of the buffer corresponding to the first extreme value (T_TP). Overall, IHI was strongest in summer, while the results varied according to the industrial complex in other seasons. In particular, the intensity of IHI was negative in winter for Seowoon, and in both fall and winter for Cheongna and IHP. This indicates that the temperature tends to remain constant or increase with increasing distance from these industrial complexes. Accordingly, the IHI in these industrial complexes is interpreted as very weak or non-existent. The tendency for the IHI to be stronger in warm seasons and weaker in cold seasons is consistent with the findings of Meng et al. (2022). In contrast, the UHI is typically stronger in fall and winter (Lee et al., 2017), confirming that the seasonal pattern of the IHI is different from that of the UHI.

Fig. 6 shows the LST profile graph for summer, the season with the strongest IHI. In all industrial complexes, LST decreased, satisfying the IHI existence criteria outlined in Section 2.3. Accordingly, the distance to the buffer corresponding to the first extreme value (E_Buffer) was defined as the IHI range. For instance, in the Western industrial complex, where LST declined sharply, the IHI range was 390 m, and the intensity of IHI (ΔT_max) reached 4.96°C, the highest among the industrial complexes analyzed. In addition, Seowoon, Cheongna, and IHP, which had very weak or non-existent IHI during colder seasons, displayed weak levels of IHI in summer. These results demonstrate that IHI, like UHI, is influenced by seasonal changes and is most pronounced in summer. Furthermore, the results indicate that the type of industries within the industrial complex cannot fully explain the formation of IHI. For example, Cheongna and Seowoon, hosting a diverse range of manufacturing enterprises, exhibited very weak or non-existent IHI. Conversely, Songdo, the urban high-tech industrial complex with companies focused on research and development, demonstrated the fifth strongest IHI among all industrial complexes.

Figure 6. LST profiles of industrial complexes in summer.

Consequently, IHI arises from a complex interplay of various factors that constitute an industrial complex. Thus, to better understand IHI formation, further analysis is necessary to account for additional factors such as topography and land use. In this study, the influencing factors of IHI were analyzed based on the IHI regions identified in summer.

4.3. Analysis of Correlation Between IHI and Industrial Factors

4.3.1. GWR Analysis

Before analyzing the factors influencing IHI based on the independent variables selected in this study, potential multicollinearity was assessed using the variance inflation factor (VIF). While there is no universally accepted threshold for determining multicollinearity, a VIF value exceeding 10 is generally considered to be of concern (Senaviratna and Cooray, 2019). As presented in Table 9, the VIF values of all independent variables were less than 10, indicating that multicollinearity was not an issue.

Table 9 . Results of the multicollinearity test.

Variable		VIF
Industrial Activity Factor	NO2	3.232810
	SO2	3.274802
Industrial Space Factor	DEM	1.475150
	SAVI	4.102683
	NDBI	5.732950
	AWEInsh	5.117545
Industrial Workforce Factor	Workers	1.048237

The GWR analysis in this study was performed using the Scalable GWR algorithm proposed by Murakami et al. (2020). This method improves computational efficiency and accuracy for large datasets (N > 3,000) by replacing bandwidth selection with a k-nearest neighbors approach. The optimal number of neighbors was determined as 52 based on theAICc.

The summary of GWR coefficient estimates is presented in Table 10. The regression coefficients of each variable and the Local R² and residuals of the GWR model were spatially visualized, as shown in Fig. 9, to examine the influence of each independent variable on the IHI regions. The regression coefficient represents the impact of independent variables on the dependent variable (LST). Local R² indicates how much the independent variables explain the dependent variable for each grid, the spatial unit in this study.

Figure 9. Satellite images of (A) Beautiful Park (source: Esri base map imagery provided by Maxar).

Table 10 . Summary of GWR coefficient estimates.

Variable		Min.	1st Qu.	Median	3rd Qu.	Max.
Intercept		-123.29786	27.95332	46.36831	69.06586	278.8111
Industrial activity factor	NO2	-13.6387	-3.36	-2.5706	-1.58575	7.4406
	SO2	-20.54573	4.93211	8.01562	10.77334	32.04726
Industrial space factor	DEM	-27.68494	-10.28235	-5.61662	-0.47933	33.59111
	SAVI	-39.88124	-2.94592	0.21062	3.52672	33.79574
	NDBI	-63.39799	-13.62452	-9.55001	-5.26499	53.04452
	AWEInsh	-120.0681	-3.90086	4.13424	11.34932	90.11169
Industrial workforce factor	Workers	-200.86498	-44.80516	-23.74561	-5.65557	146.21861

The most significant factor influencing the LST in the IHI regions was the industrial activity factor, represented by air pollutants. The influence of SO₂ surpassed that of NO₂. The correlation between both variables was exclusively positive in (F) Seowoon, the thirdstrongest IHI region. In other industrial complexes, contrasting patterns were observed. In (G) Juan, the largest IHI region, NO₂ exhibited a very strong negative (–) correlation, while SO₂ exhibited a very strong positive (+) correlation. Overall, this influence of NO₂ and SO₂ on IHI formation exhibited dominant positive (+) effects in the northwest and southeast, respectively.

In the industrial space factor, the DEM exhibited a strong negative (–) correlation across most IHI regions. This finding is consistent with previous studies demonstrating that the intensity of IHI tends to increase as elevation decreases. In particular, a very strong negative (–) correlation was observed in (G) Juan and (I) Songdo, both located near high-elevation mountains. The relatively lower elevation of these regions compared to their surroundings is likely to restrict air circulation, thereby contributing to the accumulation of air pollutants. Consequently, the increased relative elevation disparity may have triggered a cascade of environmental phenomena that ultimately intensified the IHI.

SAVI exhibited a moderate negative (–) correlation overall. Fig. 7 illustrates the vegetation distribution in (B) Western and (E) Bupyeong, both of which demonstrated a strong negative (–) correlation. In particular, (E) Bupyeong exhibited the smallest IHI range but the highest intensity and demonstrated SAVI values of 0.2 to 0.4. This region is characterized by a high density of buildings and a scarcity of vegetation. Thus, this case indicates that as vegetation density decreases, IHI intensity increases. However, the SAVI functions solely as an indicator of vegetation density, neglecting to consider the structure or type of vegetation. Consequently, the simple correlation proposing that diminished SAVI values engender heightened IHI intensity cannot be universally implemented across all regions. (B) Western exemplifies this limitation. In (B) Western, SAVI values were very high in the northern area and the southern area outside the industrial complex. However, a negative (–) correlation was observed in the north and a positive (+) correlation in the south. As illustrated in Fig. 7, the vegetation in the south consists of artificial grass within a golf course, differing from the natural vegetation in the north. These findings suggest that even in regions with high SAVI values, the impact on IHI varies depending on the type of vegetation. Furthermore, the green zone located in the northwest of the industrial complex exhibited a cooling effect, suggesting that appropriate vegetation was selected to mitigate IHI.

Figure 7. Spatial distribution of the local coefficient for independent variables and GWR parameters (residual and local R2).

NDBI and AWEInsh demonstrated results that contradicted the trends identified in numerous existing studies, including Ogashawara and Bastos (2012), in specific regions. Specifically, NDBI exhibited a positive (+) correlation in (C) Cheongna and (E) Bupyeong, while other industrial complexes displayed a mix of moderate negative (–) and positive (+) correlations. These findings are a reflection of the distinctive regional characteristics of Incheon. For instance, the western part of Incheon borders the coast, so prevailing west winds during summer may have diminished the influence of NDBI due to the sea breeze. Furthermore, characteristics of industrial complexes, such as the shadow effect from large factory buildings and the utilization of building materials with high light reflectivity, may have contributed to the mitigation of IHI.

AWEInsh exhibited a negative correlation with (C) Cheongna, (E) Bupyeong, and (G) Juan, which is consistent with previous studies. However, it displayed varying degrees of positive (+) correlation in other industrial complexes. These findings may be attributed to the reduced cooling effect of water bodies surrounding industrial complexes, caused by the discharge of pollutants and heat from industrial activities. Fig. 8 presents the satellite image of (A) Beautiful Park, demonstrating a very strong positive (+) correlation. Geomdan Stream, located northwest of (A) Beautiful Park, is adjacent to Gimpo City’s Hagun Industrial Complex and connects to Anam Lake on the left. According to the Sudokwon Landfill Site Management Corporation (2022), the water quality of Geomdan Stream and Anam Lake has consistently remained at the lowest level (Grade 6) of COD environmental standards since 2019. This pollution is primarily attributed to the industrial wastewater from nearby industrial complexes, including (A) Beautiful Park. Consequently, it can be inferred that the continuous inflow of discharge water has deteriorated water quality, thereby further exacerbating the intensity of IHI.

Figure 8. Satellite images of (B) Western and (E) Bupyeong (source: Esri base map imagery provided by Maxar).

Finally, Workers exhibited a strong positive (+) correlation across most industrial complexes. The northwest region of (G) Juan observed a very strong positive (+) correlation. The maximum regression coefficient of workers in this region was 57.124418, which is identified as the greatest influence variable on the formation of IHI in (G) Juan. An increase in the number of workers in industrial complexes is associated with higher energy consumption, increased industrial activity, and greater traffic volume. Consequently, it is hypothesized that these factors contribute to elevated artificial heat emissions, potentially intensifying the IHI.

4.3.2. Model Evaluation: Comparison of OLS and GWR

In this study, the OLS and GWR model results were compared to evaluate the fit of the GWR model to IHI. Table 11 presents the summary of OLS coefficient estimates and their statistical significance. The results indicate that all independent variables are highly significant (p < 0.001).

Table 11 . Summary of OLS coefficient estimates and p-value.

Variable		Coefficient	Std. Error	t-value	p-value
Intercept		35.2623	1.04815	33.64243	1.332062e–244
Industrial activity factor	NO2	–2.56771	0.04904	–52.36465	0.000000e+00
	SO2	7.32698	0.12933	56.6541	0.000000e+00
Industrial space factor	DEM	–3.91649	0.16355	–23.9465	8.237424e–126
	SAVI	4.00459	0.16401	24.41689	1.098445e–13
	NDBI	–17.6364	0.21685	–81.32926	0.000000e+00
	AWEInsh	15.83505	0.39152	40.44554	0.000000e+00
Industrial workforce factor	Workers	–19.48746	0.97759	–19.9341	5.677136e–88

Fig. 10 illustrates the residual distributions used to compare the prediction accuracy of the OLS and GWR models. The x-axis represents the observed LST values in summer, while the y-axis represents the LST values predicted by the two models, with the slope of the trend line fixed at 1. The residuals of the OLS model were widely scattered from the trend line, whereas those of the GWR model were concentrated near the trend line and exhibited a narrower range. The global Moran’s I statistic and its significance level were calculated to assess the spatial autocorrelation of the residuals. The OLS model demonstrated strong positive spatial autocorrelation, with a Moran’s I value of 0.686 (p < 0.001). In contrast, the GWR model exhibited a significantly reduced Moran’s I value of 0.050 (p < 0.001), indicating a much weaker spatial autocorrelation. These findings demonstrate that the GWR model provides more accurate predictions than the OLS model.

Figure 10. Residual distributions of models. (a) OLS model. (b) GWR model.

The $R_{adj}^2$. (Adjusted R-squared) is a measure of how well a model explains the variability of the data. $R_{adj}^2$. values ranging from 0 to 1, with values closer to 1 indicating that the model explains a larger portion of the variance in the dependent variable. The $R_{adj}^2$. value of the GWR model was 0.9126, which is 0.5449 higher than that of the OLS model (0.3677). Additionally, a lower AICc value indicates a better-fitting model. Generally, a difference of 3 or more in AICc values between two models is considered significant. The AICc value of the GWR model was 111,146.4, which is 60,459 lower than that of the OLS model (171,605.4). These findings demonstrate that the GWR model outperforms the OLS model in terms of explanatory power and model fit.

4.4. Findings and Discussion

The analysis of the IHI at the mesoscale and local scale for the inland region of Incheon revealed the following findings. In the mesoscale analysis, the chi-square independence test indicated that the distribution of UHI grades between industrial and nonindustrial areas was statistically significant. The evaluation of the association using Cramer’s V demonstrated a strong correlation across all four seasons, confirming that industrial complexes perform a pivotal role in the formation of UHI. This finding suggests that the IHI of Incheon is of adequate scale to be defined as an independent phenomenon at the mesoscale. Following the initial findings, the results of the local scale analysis revealed that the spatial extent and intensity of IHI were strongest in summer. In some industrial complexes, IHI weakened significantly in fall and winter to the point of being almost indiscernible. Therefore, IHI exhibits a different seasonal pattern than UHI, deviating from previous research (Lee et al., 2017), which identified UHI as more pronounced in fall and winter.

The analysis of IHI impact factors using the GWR model revealed that the influence of each variable varied across industrial complexes. The most influential variable was the Industrial Activity Factor, represented by air pollutants (NO₂ and SO₂). SO₂ exhibited a strong positive (+) correlation, while NO₂ exhibited a negative correlation. The Industrial Space Factor, DEM, demonstrated a strong negative (–) correlation by confirming the association between lower elevations and higher IHI intensity in line with previous studies. SAVI exhibited a moderate negative correlation, but its influence differed based on vegetation type. For example, in (B) Western, natural vegetation mitigated IHI, while artificially created lawns did not have the same effect. To effectively mitigate IHI, accounting for both the quantitative and qualitative characteristics of vegetation, such as structure and type. NDBI and AWEInsh demonstrated results that contradicted the trends identified in numerous existing UHI studies.

According to Ogashawara and Bastos (2012), buildings and roads tend to intensify UHI, while water tends to mitigate UHI. However, NDBI displayed a mix of moderate negative (–) and positive (+) correlations across all but two industrial complexes. Since Incheon borders the coast, prevailing west winds during summer may have diminished the influence of NDBI due to the sea breeze. AWEInsh exhibited varying degrees of positive (+) correlation in most industrial complexes. This pattern can be interpreted as the continuous discharge of industrial wastewater into nearby streams, which degrades water quality and, in turn, intensifies IHI. The Industrial Workforce Factor shows that workers exhibited a strong positive (+) correlation in most industrial complexes. Notably, it had the greatest influence on IHI formation in the northwest region of (G) Juan. An increase in the number of workers leads to a rise in artificial heat emissions, suggesting the necessity of implementing policies to reduce heat emissions to alleviate IHI in the region.

A comparison of the OLS and GWR models for Incheon revealed that the GWR model outperformed the OLS model. The global Moran’s I statistic and its significance level for residuals were 0.686 (p < 0.001) in the OLS model, but it decreased substantially to 0.050 (p < 0.001) in the GWR model. This indicates a significant reduction in spatial autocorrelation in the GWR model. Additionally, the $R_{adj}^2$. value of the GWR model was 0.9126, which was 0.5449 higher than that of the OLS model (0.3677). This value demonstrates that the GWR model better accounts for the variability of LST. The AICc value of the GWR model was 111,146.4, which was 60,459 lower than that of the OLS model (171,605.4). This discrepancy confirms that the performance difference between the two models is statistically significant. Consequently, the GWR model is more effective in explaining the spatial patterns of IHI than the OLS model.

Industrial complexes have been recognized as significant contributors to UHI due to their high pollutant emissions, and their environmental impacts often reinforce UHI. While recent studies have explored industrial heat islands (IHI) as a subset of UHI, most have been limited to local-scale analyses, making it difficult to establish broader connections with UHI research. Additionally, the specific industrial characteristics influencing IHI remain largely unexplored.

To address these gaps, this study proposes a geospatial framework for analyzing the spatial characteristics and influencing factors of IHI within industrial complexes. To validate this framework, spatial patterns of IHI were examined at two scales (mesoscale and local scale) in the inland region of Incheon Metropolitan City. Furthermore, a quantitative assessment of the factors contributing to Incheon’s IHI formation was performed using a GWR model. The GWR model’s performance was also evaluated by comparing it with the OLS model, demonstrating its superior ability to explain IHI patterns.

The academic significance of this study lies in redefining IHI as a subtype of UHI and proposing a geospatial framework for analyzing IHI as an independent environmental phenomenon. By introducing a hierarchical framework that integrates mesoscale and local scale analyses, this study strengthens the connection between IHI and UHI, a relationship that existing research has insufficiently explored. This strengthened connection provides a foundation for expanding IHI research.

In addition, the GWR analysis revealed that the geographical characteristics of Incheon Metropolitan City (coastal region), along with the physical and social environments of individual industrial complexes, had a significant impact on IHI. These factors influenced both the strength and direction of the correlation. For instance, the influence of NDBI appears to have been diminished by sea breezes. At the same time, AWEInsh is interpreted to have contributed to the intensification of IHI because of water quality deterioration caused by industrial wastewater inflow. These findings indicate that the complex interplay of various factors within industrial complexes forms IHI. Consequently, customized environmental improvement measures that reflect the unique characteristics of each industrial complex are necessary.

However, the methodological framework proposed in this study has several limitations. Although the newly introduced mesoscale analysis served as an important tool for assessing the potential independence of IHI prior to the local-scale analysis, it did not quantitatively connect the results between the two scales or examine their interactions. Future research should focus on integrating the patterns observed at both scales and investigating how local scale IHI patterns influence mesoscale UHI phenomena.

Additionally, due to constraints in data collection, this study was unable to consider qualitative factors such as energy consumption patterns and industrial structure. Furthermore, while the GWR model proved effective in quantitatively analyzing the influence factors forming IHI, it falls short of explaining the mechanisms through which these factors affect IHI. Addressing this limitation, simulation models such as CFD and ENVI-met could be employed to provide deeper insights into the IHI formation process based on GWR results. Finally, analyzing IHI across diverse regions and industrial complexes is expected to refine and expand the framework proposed in this study further.

None.

No potential conflict of interest relevant to this article was reported.