Urban Heat Islands and Land-Use Patterns in Zagreb: A Composite Analysis Using Remote Sensing and Spatial Statistics

Abstract

Urban heat islands (UHIs) present a growing environmental issue in swiftly urbanizing regions, where impermeable surfaces and a lack of vegetation increase local temperatures. This research analyzes the spatial distribution of urban heat islands in Zagreb, Croatia, utilizing remote sensing data, urban planning metrics, and spatial-statistical analysis. Composite rasters of land surface temperature (LST) and the Normalized Difference Vegetation Index (NDVI) were generated from four cloud-free Landsat 9 images obtained in the summer of 2024. The data were consolidated into regulatory planning units through zonal statistics, facilitating the evaluation of the impact of built-up density and designated green space on surface temperatures. A composite UHI index was developed by combining normalized land surface temperature (LST) and normalized difference vegetation index (NDVI) measurements, while spatial clustering was examined with Local Moran’s I and Getis-Ord Gi*. The results validate spatial patterns of heat intensity, with high temperatures centered in densely built residential areas. This research addresses the gap in past UHI studies by providing a reproducible approach for detecting thermal stress zones, linking satellite data with spatial planning variables. The results support the development of localized climate adaptation methods and highlight the importance of integrating green infrastructure into urban planning methodologies.

1. Introduction

1.1. UHI Drivers and Vegetation Mitigation

Urban heat islands (UHIs) represent one of the most prominent localized consequences of climate change and urbanization. Characterized by elevated temperatures in urban areas compared to their rural surroundings, UHIs arise due to impervious surfaces, building density, and limited vegetation. In urban climatology, the urban heat island (UHI) is defined as a phenomenon in which urban areas exhibit higher surface and air temperatures than their rural surroundings. Sustainable urban planning requires understanding UHI drivers and spatial patterns as climate change intensifies and metropolitan areas densify. Research on UHI generation and mitigation has focused on land use, vegetation dynamics, and urban morphology. Remote sensing and spatial statistics allow for more detailed urban surface temperature studies. However, many issues remain, particularly regarding the link between remotely detected urban heat island measurements and spatial planning indicators, such as zoning and planned green areas. This literature review is organized into four primary topics to provide academic context for the work. The first thematic section examines how land use and urban layout contribute to the formation of urban heat islands. The second focuses on thermal mitigation with vegetation and NDVI. The final segment reviews the methodological frameworks of remote sensing and spatial analysis. The fourth section outlines research shortages in Central European urban areas, including Zagreb.

Land use—especially urbanization, vegetation, and water bodies—shapes surface temperature and urban heat island (UHI) intensity. In Wuhan, China, industrial zones recorded the highest LST due to limited vegetation [1]. Impervious surfaces retain more heat than vegetated or water-covered areas. In Izmir, Turkey, major roads intensified UHI effects, while nearby streets reduced heat through better airflow. [2]. In Wuhan, spatial regression showed that building density and sky view factor (SVF) were key UHI drivers, while floor area ratio (FAR) had context-dependent effects, highlighting the role of urban form over land cover. [3]. In Shenyang, a random forest model using Landsat 8 data showed that public service zones had the greatest impact on surface temperature, followed by residential, industrial, and commercial areas, emphasizing the role of urban function in UHI patterns [4]. Landsat data from 2013 to 2022 showed that the city core and industrial districts in Zagreb had the highest heat levels, matching those of Berlin, Germany, and Istanbul, Türkiye, which have dense urban layouts [5,6]. Zagreb’s urban growth and lack of green infrastructure have likely exacerbated UHI impacts in recent decades [7]. Industrial and commercial zones caused most temperature anomalies, while older areas with narrow streets and poor ventilation had greater nighttime UHI [8]. Despite well-established links between built-up land and LST, few studies integrate zoning data with satellite-derived temperatures. This study bridges the fields of urban climatology and spatial planning. Green spaces help regulate urban heat through evapotranspiration, shading, and heat absorption. Numerous studies show that dense vegetation significantly reduces LST, as confirmed by strong correlations with vegetation indices. For example, urban forestry in Bochum, Germany, effectively reduced heat stress. [9], while green infrastructure in the Andalusia region in Spain improved thermal balance [10]. Tree-covered regions in Skopje, North Macedonia, had much lower LST than urban zones [11], whereas in Naples and Florence, Italy vegetated areas were cooler than built-up or barren areas [12]. From 1961 to 2019, summer temperatures in Dubrovnik, Croatia, climbed, especially over urbanized surfaces [13]. Italian research confirms that tree cover reduces UHI, while impermeable surfaces increase it, especially inland, where water-based cooling is lacking [14]. Extensive Landsat research in Sivas, Turkey (1989-2015), revealed a strong inverse correlation (R² > 0.95) between NDVI and LST. The urbanization of agricultural and vegetated areas increased UHI intensity from 6.77 °C to 9.02 °C [15]. These findings underscore the importance of tree cover, green corridors, and vegetated zones for mitigating UHI. NDVI is widely used to assess vegetation in UHI studies, although some limitations exist. However, future green spaces are rarely integrated into formal land-use plans. Remote sensing and GIS are increasingly supporting UHI assessment by combining land cover, LST, and vegetation indices (NDVI, NDBI), particularly using satellites such as Landsat and Sentinel. For instance, in Seville, Spain, rising urbanization without green infrastructure elevated surface temperatures [16], while in Prishtina, Kosovo, high NDBI values revealed UHI impacts despite relatively low surface temperatures [17]. In Beijing, urbanization of agricultural areas led to expanded thermal hotspots [18]. The hottest UHI hotspots in Tehran, Iran, were industrial, asphalt, and transit regions, while green zones had much lower LST [19]. In Kayseri, Türkiye, Landsat 8/9 (2013–2022) revealed that NDVI was negatively correlated and NDBI was positively correlated with UHI intensity. The creation of the National Garden from a decommissioned airfield highlights the cooling effect of green infrastructure [20]. Other studies found that finer spatial resolution improves localized UHI pattern recognition in Hefei, China [21]. Over half of global satellite-based UHI studies use Landsat, MODIS, or ASTER data [22]. These studies examine LST-NDVI correlations, urban-rural contrasts, and LST-air temperature discrepancies. European Landsat 8 data in Skopje verified the vegetation’s cooling impact [11], whereas Sentinel-3 3 images in Andalusia showed the rural heat and the cooling impact of vegetation [11], whereas Sentinel-3 images in Andalusia showed rural and urban heat retention [23]. Bare soil and urbanization intensified UHI in Ahvaz, Iran [24]. Researchers in Split, Croatia, developed a Combined Thermal Index (CTI) using Landsat 8/9 and in situ data from July to September 2024. The CTI identified dense commercial and industrial zones—especially Split 3—as the most heat-vulnerable, based on LST, NDVI, NDBI, NDWI, and thermal perception metrics. The study recommends combining remote sensing with human thermal perception indicators for more effective UHI mitigation [25]. Addressing urban heat islands (UHIs) requires targeted policy and planning measures. Effective strategies include expanding green infrastructure, applying reflective materials, and enhancing urban design to promote airflow and improve air quality. Urban forestry is especially important for climate adaptation—research indicates that increasing tree canopy cover by 16% can significantly reduce temperatures in European cities [26]. Reflective surfaces and permeable pavements also help reduce heat retention in dense urban areas [10]. In Jakarta, Indonesia, Landsat 8 LST data and LCZ classification identified compact and lightweight low-rise zones (LCZ 3 and 7) as thermal hotspots due to impervious surfaces, heat-absorbing materials, and low vegetation. Cooling measures such as greening, cool roofs, and improved ventilation are recommended for these areas [27]. A study in Hangzhou, China, comparing Landsat-derived LST and air temperature (Tair) metrics found differing UHI patterns by time of day. LST responded more to daytime solar radiation, while Tair captured stronger nighttime UHI. Land cover proved more effective than temperature type in explaining spatial differences, and acquisition time and weather conditions were key for analysis accuracy [28]. In South Korea, MODIS data (2001–2022) showed that urban greening in Seoul, South Korea, through parks, trees, and green planning, led to a steady decline in summer SUHII, whereas industrial growth in Ulsan, South Korea, increased it due to vegetation loss. These findings support targeted greening as an effective heat mitigation strategy [29]. Recent studies also employ composite indices, regression models, and spatial statistics (e.g., LISA, Getis-Ord Gi*) to map temperature patterns. Even so, few associate UHI risk with planning zones and development regulations employing all three methodologies.

1.2. Spatial-Statistical Methods

Anselin [30] developed LISA to decompose global statistics like Moran’s I into local components, enabling the detection of significant hot spots and spatial outliers. Getis and Ord [31] introduced the local Gi and Gi* statistics, highlighting the importance of adjusting for multiple testing when identifying spatial clusters. Chen [32] proposed partial spatial correlation structures and spatial autocorrelation functions (SAFs), advocating for scalable, multidimensional analysis integrating zoning, LST, and NDVI. Ord [31] emphasized the role of spatial weights matrix design and the combined use of LISA and Gi* for improved pattern recognition. Getis [31] distinguished between global and local spatial autocorrelation measures in the context of environmental risks like urban heat islands. These methods are essential for identifying UHI clusters and guiding targeted interventions.

1.3. Regional Research Gap

This study fills that gap by employing zonal statistics, predictive modeling, and local spatial analysis. From 2013 to 2022, Landsat 8 data in Zagreb showed a persistent temperature differential between green and built-up areas, with forests and green zones remaining colder than urban surfaces [33]. Some research [7] found a strong negative relationship between plant cover and surface temperature, with green areas experiencing lower heat stress and built-up zones showing pronounced UHI effects. Urban architecture influences thermal patterns, with central and industrial districts consistently acting as heat hotspots from 1984 to 2014 [8]. The Environmental Criticality Index (ECI) identified year-round temperature anomalies in commercial and transport areas due to impermeable surfaces, limited vegetation, and anthropogenic heat [7]. Thermal field tests confirmed that industrial and commercial zones had the highest heat loads, while residential areas showed moderate levels [8]. Land-cover transitions, especially from vegetated to urbanized surfaces, have exacerbated UHI patterns due to decreased evapotranspiration and increased heat retention [33]. Zagreb’s urban greenery can chill the area by up to 2 °C, especially in parks, forested corridors, and tree-lined avenues [8]. A 16% increase in tree cover could reduce urban temperatures by 1.5 °C [33]. These findings support European sustainability goals by promoting vegetation-based mitigation and supportive planning laws. Central and Eastern Europe are underrepresented in global UHI research despite their importance in urban thermal studies. Integrated assessments of zoning, green spaces, and future land use are lacking in Zagreb. Additionally, the study assesses urban thermal sensitivity using composite UHI indicators, planned green spaces, built-up density, and spatial autocorrelation metrics. Examining geographical patterns necessitates spatial autocorrelation to detect clustering, anomalies, and heterogeneity.

Building on prior research that underscores the impact of urban form, vegetation, and planning designations on surface temperature, this study integrates satellite data with zoning indicators to systematically assess spatial thermal patterns and UHI dynamics in Zagreb. The approach is structured around four main objectives: (1) to assess the relationship between the proportion of built-up areas and surface temperatures, with the expectation that more impervious zones are associated with elevated LST values; (2) to examine the cooling potential of planned green areas by evaluating the correlation between designated vegetation zones and observed surface temperatures; (3) to construct a composite UHI index, based on normalized LST and NDVI values, in order to identify zones of heightened thermal stress; and (4) to explore the spatial clustering of UHI values by applying local spatial autocorrelation techniques to detect statistically significant patterns of hot and cool spots within the study area.

This research addresses the four hypotheses based on the previously stated primary objectives:

H1. 

Higher shares of built-up areas are positively correlated with higher land surface temperatures.

H2. 

Larger amounts of planned green areas are associated with lower land surface temperatures.

H3. 

NDVI and LST values exhibit a statistically significant negative correlation at the pixel level.

H4. 

Composite UHI index values show statistically significant spatial autocorrelation, forming clusters of temperature intensity.

2. Materials and Methods

2.1. Study Area and Climatic Context

According to the Copernicus Climate Change Service (C3S) and the European State of the Climate 2024, 2024 was the warmest year on record, both globally and in Europe, surpassing the exceptional heat of 2023. The global average temperature for the 12 months from September 2023 to August 2024 was 1.64 °C above the pre-industrial baseline (1850–1900), while Europe experienced its hottest summer on record, with land temperatures during June–August 2024 averaging 1.54 °C above the 1991–2020 average [34,35].

In Croatia, July and August 2024 were officially the hottest months on record at many meteorological stations. The mean monthly air temperature anomaly ranged from +2.1 °C to +3.8 °C in July and from +1.8 °C to +4.7 °C in August, compared to the 1991–2020 climatological norm. According to the Croatian Meteorological and Hydrological Service (DHMZ), both months were classified as extremely warm across most of the country, with percentile values placing them at the highest extremes in decades [36].

The motivation and empirical background for this investigation of intra-urban temperature dynamics in Zagreb, which focuses on the spatial organization and planning consequences of UHI creation, are explained by these circumstances. The city of Zagreb, Croatia, offers a relevant geographical setting for examining dynamics. Strong intra-urban land-cover heterogeneity, continuous urbanization, and growing vulnerability to climate extremes are all present in the city. It remains necessary to evaluate the comparative impacts of designed green infrastructure and urban morphology on alleviating urban heat stress.

This study applies a high-resolution spatial analysis of LST and NDVI, derived from Landsat imagery, to explore the intensity and spatial clustering of UHI in the city of Zagreb. In doing so, the research integrates satellite-based thermal and vegetation indicators with urban planning data, particularly the extent of planned green areas, the share of built-up zones, and residential land use, aiming to assess their relationship with observed thermal patterns. By combining descriptive statistics, zonal analysis, regression modeling, and spatial autocorrelation techniques (e.g., Local Moran’s I, Getis-Ord Gi*), the study seeks to detect the drivers of thermal variation, evaluate the effectiveness of planned mitigation zones, and identify statistically significant hotspots of thermal stress. Grounded in recent records of extreme summer temperatures, this work is guided by five research hypotheses and four operational objectives, which together structure the methodological framework and empirical analysis.

This research examines the urban landscape of Zagreb, the capital of Croatia, which features a varied urban morphology. The spatial scope aligns with the coverage of the General Urban Plan (GUP), establishing a basis for evaluating the impact of various land uses and planned developments on urban heat patterns. This research focuses on the extreme heat context by examining the summer of 2024, which has been recorded as the hottest globally and in Europe. In Croatia, July and August 2024 were recorded as the hottest months at various weather stations, with anomalies reaching 4.7 °C above the 1991–2020 average [36]. In this context, temperature data were obtained from two distinct meteorological stations: Zagreb Grič, positioned in the city’s dense historical core, and Zagreb Maksimir, located in a greener peripheral area. To explain satellite-derived surface temperature observations and investigate intra-urban thermal variability, the stations offered empirical support. Four Landsat scenes were chosen to match the available cloud-free imagery: 10 July, 27 July, 11 August, and 28 August 2024. The specified dates gave the temporal basis for calculating LST and vegetation greenness (NDVI), establishing a foundation for a spatially detailed analysis of UHI intensity and its correlation with urban planning variables.

2.2. Data Sources and Spatial Layers

This study examines the patterns and drivers of UHI intensity in Zagreb using satellite imagery, weather measurements, and vast urban planning records. LST, NDVI, and planning documents provide an approach, encompassing both physical surface features and structural urban indicators. This study investigates thermal risks and UHI phenomena in Zagreb using high-resolution satellite images, official meteorological data, and spatial urban planning information. The primary remote sensing data is obtained from Landsat-8 and Landsat-9 photos captured on four carefully chosen, cloudless summer dates in 2024 (10 July, 27 July, 11 August, and 28 August). The selected images were chosen for their visual clarity and minimal atmospheric distortion, ensuring optimal conditions for deriving LST and NDVI values. Thermal infrared bands (Band 10) measured LST, while red (Band 4) and near-infrared (Band 5) bands calculated the normalized difference vegetation index (NDVI), which measures vegetation density and health. To show spatially constant plant cover and thermal intensity patterns across the study area, composite mean rasters were created. The analysis utilized daily air temperature data from two Croatian Meteorological and Hydrological Service (DHMZ) stations to validate satellite-based results against empirical surface conditions in Zagreb Grič (an urban center) and Zagreb Maksimir (a suburban, vegetated area). The two sites, with varied urban shapes and land cover, provided ground-truth references for satellite-derived shapes and land cover, as well as ground-truth references for satellite-derived LST thermal anomalies and a temporal framework for urban warming dynamics on the required dates. Additionally, spatial planning data were employed to assess changes in surface temperature related to institutional land-use classifications and rules and regulations. This analysis utilizes the City of Zagreb’s General Urban Plan (GUP) and its 2024 amendments, which delineate zoning typologies and the spatial framework for urban development. Five green spaces—Z (protective greenery), Z1 (public parks), Z2 (urban park-forests), Z3 (thematic parks), and Z4 (public green zones)—were consolidated into a single indicator to show the total designated green area per planning unit. GUP regulations also limit built-up intensity to a certain percentage of construction land. This indicator proxied surface sealing and urban density. These spatial variables provided a framework for analyzing the relationship between formal urban planning objectives, specifically greening and building density, and thermal behavior. The study utilizes satellite-derived indicators and regulatory land-use constraints to create a multidimensional dataset for spatial-statistical modeling, including regression analyses and spatial autocorrelation testing, to assess the efficacy and equity of planning measures for reducing UHI intensity. Urban regulations (hr. urbana pravila) govern construction permits in most consolidated urban areas, as outlined in the General Urban Plan (GUP). The GUP states that these laws provide a sufficient regulatory framework for building activity without comprehensive local plans (such as UPU or DPU), except for less consolidated zones or special urban project regions. Consequently, elements such as the projected built-up ratio have normative significance, indicating spatial constraints on construction density and surface sealing.

2.3. Remote Sensing Preprocessing

This study employed a systematic preprocessing workflow to derive Landsat-8 and Landsat-9 Level-2 LST and Normalized Difference Vegetation Index (NDVI) from Landsat-8 and Landsat-9 Level-2 satellite imagery. All raster operations were conducted in RStudio 2023.09.1+494 “Desert Sunflower”, using packages such as raster, sf, and terra. To generate accurate representations of LST across time, a standardized thermal preprocessing routine was applied to multiple satellite acquisitions. Thermal information was extracted from Band 10 (ST_B10) of each Landsat scene acquired on four cloud-free dates in 2024: 10 July, 27 July, 11 August, and 28 August. The digital numbers were converted to top-of-atmosphere radiance using the scale factor 0.00341802 and an offset of 149.0. These were then transformed to temperature values in degrees Celsius by applying the Planck equation and subtracting 273.15 K for Celsius conversion. Each LST raster was clipped to the urban development boundary, representing the official limits of spatial plans for the City of Zagreb. CRS mismatches were resolved via re-projection, and all rasters were resampled to ensure pixel alignment. A composite LST surface was then produced by stacking the four layers and calculating the pixel-wise mean, which helped smooth out daily thermal anomalies and capture average conditions for the summer period.

To capture spatial variability in vegetation, cover with seasonal consistency, a harmonized NDVI preprocessing approach was applied to multispectral satellite data. Vegetation greenness was assessed using NDVI, calculated from surface reflectance Bands 5 (NIR) and 4 (Red). The index was computed for each date using the standard formula:

$$\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}={\displaystyle \frac{\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{R}\mathrm{e}\mathrm{d}}{\mathrm{N}\mathrm{I}\mathrm{R}+\mathrm{R}\mathrm{e}\mathrm{d}}}$$

(1)

Each NDVI raster was clipped using the same urban boundary shapefile. The resulting rasters were then resampled using bilinear interpolation and aligned to a common reference grid. A composite NDVI raster was derived by computing the pixel-wise mean across all four dates, yielding a stable vegetation surface that mitigates noise from transient vegetation changes or atmospheric disturbances. This dual preprocessing procedure ensured that both thermal and vegetation metrics were harmonized spatially and temporally, thereby providing reliable inputs for subsequent zonal statistics, index construction, and spatial analyses.

2.4. Zonal Statistics and Variable Derivation

Following the preprocessing of LST and NDVI rasters, spatially explicit indicators were derived through zonal statistics to facilitate urban-level analysis of surface temperature variation and urban land-use characteristics. Also, all spatial operations were conducted in RStudio using the sf, raster, dplyr, and tidyr packages, among others.

To ensure spatial alignment with local planning practices, the analysis was grounded on the regulatory framework of urban rules defined by Zagreb’s General Urban Plan (GUP). Urban regulations as defined by the city of Zagreb’s General Urban Plan (GUP) are represented by the polygons that make up the basic spatial unit of analysis. In the majority of consolidated city zones, urban rules serve as the basic units of direct plan implementation, which means that they provide a suitable regulatory foundation for granting building permits without requiring lower-level detailed plans. Each urban rule polygon was assigned a unique identity (ID) to enable statistical grouping.

This was achieved by calculating the mean LST for each planning unit using the composite raster. Zonal mean values were extracted using the extract() function, resulting in a new variable mean temperature assigned to each urban rule polygon, effectively capturing the average thermal behavior across regulatory spatial units.

To assess the spatial extent of land-use categories relevant to urban climate mitigation, a shapefile containing planned land-use variables from the General Urban Plan (GUP) was used to quantify land-use composition. This dataset includes official zoning categories, which were spatially intersected with urban rule polygons to extract relevant information. The analysis focused on categories with environmental and residential significance, including recreational zones (R1, R2), green spaces (Z, Z1, Z2, Z3, Z4), and water-related zones (V1, V2, V3), such as stream corridors and water protection areas. For each intersected land-use category, area values were calculated using the `st_area()` function and then aggregated per planning unit. This produced a set of variables (e.g., Z1\_Area, R2\_Area, V3\_Area) representing the absolute area (in square meters) of each GUP land-use type within each urban rule polygon.

Multiple composite indicators were built to minimize dimensionality and increase interpretability. Total projected green space within each urban rule polygon is determined as the sum of all green land-use categories (Z, Z1–Z4). POST, generated from GUP attribute data, represents structural intensity and surface imperviousness as a percentage of authorized built-up coverage for each unit. From attribute data, ZG_TOT calculates building coverage for each planning unit. To ensure uniformity, missing values were replaced with zeros and all values were rounded. These indicators are crucial independent variables in regression analysis, allowing the study of surface temperature patterns and formal urban planning designations. Saving the completed dataset allowed hypothesis testing and spatial regression modeling.

This study created a composite UHI index to examine the correlation between surface temperature and plant cover, employing normalized LST and the Normalized Difference Vegetation Index (NDVI). This composite metric enables the identification of micro-areas that concurrently display elevated surface temperatures and diminished vegetation, hence micro-areas that display elevated surface temperatures and reduced vegetation, thereby revealing localized thermal stress hotspots. The derivation process began by transforming the mean NDVI raster into vector polygons using the rasterToPolygons() function and then converting them into an sf object (ndvi_sf). These polygons were then spatially intersected with the enriched planning dataset to produce a combined spatial layer containing both LST and NDVI values for each intersected unit. Both LST and NDVI were normalized to a [0, 1] scale using min-max scaling to ensure comparability and remove scale-induced bias:

$$\mathrm{N}\mathrm{o}\mathrm{r}\mathrm{m}\_\mathrm{L}\mathrm{S}\mathrm{T}={\displaystyle \frac{(\mathrm{L}\mathrm{S}\mathrm{T}-\mathrm{m}\mathrm{i}\mathrm{n}(\mathrm{L}\mathrm{S}\mathrm{T}\left)\right)}{\left(\mathrm{m}\mathrm{a}\mathrm{x}\right(\mathrm{L}\mathrm{S}\mathrm{T})-\mathrm{m}\mathrm{i}\mathrm{n}(\mathrm{L}\mathrm{S}\mathrm{T}\left)\right)}}$$

(2)

$$\mathrm{N}\mathrm{o}\mathrm{r}\mathrm{m}\_\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}={\displaystyle \frac{(\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}-\mathrm{m}\mathrm{i}\mathrm{n}(\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}\left)\right)}{\left(\mathrm{m}\mathrm{a}\mathrm{x}\right(\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I})-\mathrm{m}\mathrm{i}\mathrm{n}(\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}\left)\right)}}$$

(3)

The composite index was calculated as a weighted linear combination of the two normalized values:

Composite UHI Index = Norm_LST − 0.5 × Norm_NDVI

(4)

Normalized LST and inverse NDVI measurements were weighted equally to generate the composite UHI index. The equal weighting (0.5 × Norm_LST − 0.5 × Norm_NDVI) reflects the theory that excessive thermal buildup and insufficient vegetation contribute equally to surface heat stress. The substantial inverse link between LST and NDVI is supported by empirical investigations [12].

The composite UHI index developed in this study differs from conventional LST-based assessments by integrating both surface temperature (LST) and vegetation cover (NDVI), which are known to act as thermal stressors and mitigators, respectively. By assigning equal weights (0.5) to normalized LST and inverted NDVI, the index highlights areas experiencing both high heat exposure and insufficient vegetative cooling. Compared to more complex models, such as the Combined Thermal Index (CTI), which incorporates additional metrics like NDBI, NDWI, and thermal perception indicators, our approach remains methodologically streamlined. This makes it more transparent and adaptable for planning-related applications, where clear associations with zoning and green infrastructure are critical. While CTI provides a broader thermal profile, the Composite UHI Index is specifically optimized to support zoning analyses and spatial prioritization for urban cooling interventions.

Given the significant absolute correlation between the two variables (r = −0.69 in this study), assigning equal weight prevents subjective prioritization or bias and allows for a concise, precise composite formulation for risk detection and geographical prioritization.

The composite UHI index captures the combined effects of elevated land surface temperatures and insufficient vegetation cover. Instead of analyzing temperature and vegetation separately, the index integrates both variables to identify areas where high surface heat and low ecological buffering co-occur. These zones represent compounded stress environments that are particularly relevant for urban climate adaptation planning. The dual-input structure of the index allows for a systematic classification of planning units according to their thermal exposure, providing a practical framework for the spatial prioritization of mitigation interventions. As such, the composite index is most effective.

To examine the spatial structure of UHI intensity in Zagreb, the study applied global and local spatial autocorrelation methods to the composite UHI index, revealing patterns of thermal clustering across the city. The spatial contiguity of units was determined using a Queen contiguity criterion based on polygon adjacency. The neighborhood structure was created using ‘poly2nb()‘, and spatial weights were normalized using ‘nb2listw()‘ with row-standardized (W) weight. Global Moran’s I was derived for the composite UHI index (C\_UHI\_I) to assess spatial dependency and test the null hypothesis of spatial randomness vs. spatial grouping. Permutation-based inference determined significance. To identify local clustering patterns, ‘localmoran()‘ was employed to compute local Ii values, Z-scores, and pseudo p-values for each spatial unit. Spatial units were classified as High-High (high values adjacent to high values), Low-Low (low values adjacent to low values), or deemed insignificant based on statistical significance (p < 0.05) and the nature of the connection. In parallel, the Getis-Ord Gi\* statistic was used with ‘localG()‘ to find substantial UHI intensity hot and cold locations, with Z-scores above 1.96 and below −1.96 indicating hot and cold places, respectively. This strategy cross-validated LISA results using a different clustering method.

2.5. Regression Models and Hypothesis Assessment

The final element of the analytical framework concentrated on assessing the statistical correlations between urban thermal conditions and spatial planning attributes, with objectives O1−O4 and hypotheses H1–H4. To do this, several linear regression models were constructed with the composite UHI index as the dependent variable. The principal model specification comprised three independent variables: one indicating the proportion of built-up area within each planning unit (reflecting construction intensity); a second representing the planned green area surface; and a third, the composite mean of NDVI, which captures the average normalized vegetation index derived from remote sensing data. This model formulation directly facilitates the examination of Hypotheses H1–H3, which assert: (H1) a positive correlation between built-up intensity and surface temperature; (H2) a negative correlation between green zones and surface temperature; and (H3) a significant inverse relationship between NDVI and LST. The regression coefficients, significance levels, and residual diagnostics were analyzed to assess the explanatory contribution of each spatial variable. Before model estimation, an interquartile range (IQR) filter was utilized to eliminate extreme outliers in both dependent and independent variables, hence enhancing the model’s robustness and representativeness. All variables were normalized to guarantee the comparability of coefficient magnitudes. The evaluation of model performance was conducted utilizing the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). These criteria facilitated a quantitative evaluation of model fit and predicted accuracy throughout the study area. The findings validated that urbanized regions are markedly and positively correlated with UHI intensity, whereas vegetated areas, whether designed or naturally occurring, are consistently associated with reduced surface heat values. The regression model outputs were subsequently utilized in spatial studies for Objective 4 (O4), specifically to identify spatial clusters of residual heat through Local Indicators of Spatial Association (LISA) and Getis-Ord Gi* statistics. These spatial approaches to the existence of statistically significant clustering in the spatial distribution of urban heat. A comprehensive description of the regression model framework and variable transformations is provided in Appendix A, while spatial autocorrelation scripts and output visualizations are elaborated on in Appendix B.

3. Results

3.1. Spatial and Thermal Characterization of the Urban Environment

To contextualize the satellite-derived LST values, daily air temperature data from two meteorological stations—Zagreb Grič, located in the dense urban core, and Zagreb Maksimir, representing a more peripheral and vegetated setting—were analyzed for the summer of 2024. Four cloud-free dates, which corresponded to satellite acquisitions, were given particular attention: 10 July, 27 July, 11 August, and 28 August. The average temperatures for each day were calculated as z-scores and then compared to the seasonal averages using percentage deviation. This enabled an objective assessment of whether these dates reflected thermally typical or anomalous conditions.

Based on the outcomes, 10 July and 11 August were among the warmest days of the season, with z-scores exceeding +1.0 at both stations. On the other hand, the z-scores were nearly zero, particularly at Maksimir, during the remaining two acquisition dates, 27 July and 28 August, indicating that they were thermally neutral (Figure 1a). This temporal dispersion provides a balanced temporal framework for studying surface heat, ensuring that satellite data records both extreme and average thermal states which are remarkably stable. This supports the notion of regional climatic coherence and enhances confidence in the representativeness of the selected satellite dates. The urban Grič station recorded higher mean temperatures (27.2 °C) than Maksimir (25.1 °C) across the four dates, which supports the anticipated microclimatic disparity (Figure 1b).

To provide a representative overview of average thermal and vegetative conditions during the summer of 2024, composite rasters were generated for LST and the normalized difference vegetation index (NDVI) using four cloud-free Landsat 8/9 acquisitions (10 July, 27 July, 11 August, and 28 August). These rasters demonstrate the average surface thermal conditions and vegetation cover across the urban expanse. A significant thermal variation is evident in the composite LST raster, with higher values concentrated in densely built-up urban cores and lower values in vegetated or open zones. This variation is evident throughout the study area. The NDVI composite, on the other hand, suggests that the lowest vegetation index values are found in areas with minimal vegetative cover, such as urban cores. Conversely, peripheral zones demonstrate higher NDVI and correspondingly lower LST, which reinforces the cooling role of vegetation. Confirming the substantial accumulation of surface heat during the analyzed period, the composite LST raster indicated an overall mean surface temperature for the area to be 37.83 °C. The spatial distribution of both indicators illustrates the inverse relationship between vegetation density and surface temperature (Figure 2).

A composite raster of the Normalized Difference Vegetation Index (NDVI) was created to improve thermal analysis by consolidating values from four Landsat acquisition dates: 10 July, 27 July, 11 August, and 28 August. Values are captured by the NDVI index, which measures the relative density and health of vegetation. The resulting composite map (Figure 3) demonstrates an inverse spatial pattern with respect to LST: zones with elevated NDVI values—primarily situated in peripheral areas, and within urban green corridors—coincide with regions of reduced surface temperature. Conversely, the urban core exhibits low NDVI values and elevated LST, which are indicative of sparse vegetation and intensive built-up land cover. The distribution of vegetation is characterized by a distinct spatial gradient, as illustrated in Figure 4. NDVI clusters around forested areas and parks. These data support the concept that vegetation influences local temperature conditions by reducing surface heat accumulation. The temporal stability of the data is markedly enhanced by employing a composite NDVI raster, which mitigates the effects of phenological change and transient atmospheric conditions. The composite raster revealed an average NDVI value of 0.279 for the whole research area, signifying moderate yet spatially heterogeneous vegetation cover throughout the peak summer season.

To examine how surface heat relates to urban spatial structure, composite LST raster values were aggregated to individual planning units through zonal statistics, enabling the analysis of thermal variation at a scale relevant for urban management. By aggregating the composite LST raster to individual planning units, zonal statistics were calculated to measure surface thermal fluctuation at the scale of urban spatial management. As a result, mean surface temperature per polygon may be extracted, efficiently converting thermal data at the pixel level into spatial units that are important to policy. The obtained values were then associated with the General Urban Plan (GUP)-derived regulated land-use categories, such as green zones (Z–Z4), water-related areas (V1–V3), and recreational and sport zones (R1–R2). Through spatial intersection, the total surface area of each land-use category for each planning unit was determined and then normalized for interpretation. This process produced a thorough dataset that connects surface temperature at the zonal level with urban form, defined by indicators of green spaces, recreational areas, and water-designated areas. The average LST per planning unit is represented by the derived variable mean_temperature, which is also the dependent variable in the following hypothesis testing. Built-up coverage and the cumulative extent of climate-relevant land-use categories—namely recreational, green, and water-designated zones—were calculated as composite indicators to capture structural density and mitigation potential across planning units. In an urban setting, these are the primary indicators of surface temperature. The regional distribution of zonal mean LST values among the planning units is shown in the visualization (Figure 4). The results indicate distinct thermal gradients, with elevated temperatures centered in southern and central urban areas. These regions have limited or scattered planning for recreational and vegetative uses. Conversely, reduced values are observed in continuous green corridors and forested hills, validating the designations’ efficacy in mitigating urban heat.

As a novel contribution of this study, the index was designed to reflect the joint influence of surface heat and vegetation on urban thermal conditions. It was constructed by combining normalized LST and NDVI values for each regulatory planning unit. This was achieved by intersecting polygonized NDVI raster data with planning zones that already contained aggregated LST values. Both variables were normalized to ensure comparability, and the resulting index identifies areas of increased thermal stress where high surface temperatures coincide with low-vegetation density. This composite measure represents an original methodological step in the study, providing a spatially explicit tool for pinpointing critical zones of urban heat exposure. The spatial distribution that came out (Figure 5) shows localized concentrations of thermal stress areas with high LST and low NDVI at the same time—usually in dense, built-up parts of the city. On the flip side, places with a lot of vegetation always have lower composite UHI values, which supports the idea that urban greenery helps cool the air. The composite index gives a more detailed picture of urban heat exposure.

3.2. Statistical Validation of Hypotheses

The analytical framework was structured around five hypotheses (H1–H5) and four operational objectives (O1–O4), aiming to comprehensively assess the spatial dynamics of urban heat accumulation and its primary drivers. The hypotheses addressed key spatial determinants of land surface temperature (LST), including the effect of built-up density (H1), the cooling contribution of formally designated green and recreational areas (H2), the pixel-level relationship between vegetation health and surface heat (H3), the spatial alignment between thermally stressed areas and regulatory planning indicators (H4), and the presence of statistically significant spatial clustering in thermal exposure (H5). In parallel, the operational objectives guided the spatial delineation of residential thermal risk zones (O1), the development of a composite UHI index integrating NDVI and LST (O2), the exploration of key statistical predictors via multivariate regression (O3), and the detection of significant spatial clusters using LISA and Getis-Ord Gi* statistics (O4). Together, these components provide an integrated understanding of where and why urban heat islands emerge, and how spatial planning may mitigate their impact.

Hypothesis 1 (H1) tested whether a higher share of built-up land within planning units is associated with elevated surface temperatures. This hypothesis was supported by a strong positive Pearson correlation (r = 0.68) and a coefficient of determination of R² = 0.46, indicating that nearly half of the temperature variability can be explained by built-up density alone. Figure 6a further illustrates that areas with higher structural coverage consistently exhibit elevated LST values, underscoring the thermal implications of impervious urban morphology.

Hypothesis 2 (H2) posited that planning units with greater green and recreational land use would exhibit lower surface temperatures. This hypothesis was partially supported: a negative correlation was observed between the proportion of green recreational area and mean LST (r = –0.32), and the associated univariate regression model accounted for approximately 10% of the temperature variance (R² = 0.10). While the cooling effect of planned green infrastructure was evident, it was less pronounced than the warming effect of built-up coverage. This finding suggests that while formal green zones can contribute to localized thermal mitigation, their efficacy may depend on factors such as vegetation density, spatial distribution, and maintenance practices (Figure 6b).

Hypothesis 3 (H3) aimed to determine whether vegetated surfaces are systematically associated with lower surface temperatures at the pixel level. To assess this relationship, a correlation analysis was conducted between the composite NDVI and LST rasters after spatial alignment and masking. The results revealed a strong negative Pearson correlation coefficient (r = −0.69), indicating that higher vegetation health is consistently linked to reduced surface heat intensity. This pixel-level association reinforces the earlier findings at the planning unit scale (see H2), suggesting that the cooling influence of vegetation is evident across spatial resolutions. The strength and magnitude of this inverse relationship confirm the critical role of vegetation in moderating urban surface temperatures and underscore the importance of ecological infrastructure in urban heat mitigation. (Figure 7).

The fourth hypothesis (H4) assesses whether the composite UHI index is spatially clustered in a statistically significant manner, as measured by local indicators of spatial association. This hypothesis tests the presence of spatially dependent patterns in surface thermal stress, identifying localized concentrations of high or low index values. To evaluate spatial clustering in thermal exposure, two complementary local spatial statistics were applied: Local Moran’s I and the Getis-Ord Gi* statistic. The Local Moran’s I results revealed numerous statistically significant high–high and low–low clusters, indicating spatial concentration of both elevated and suppressed UHI index values. In parallel, the Getis-Ord Gi* analysis identified contiguous “hot spots” and “cold spots” of composite thermal stress across the urban fabric (Figure 8a). At the global level, the Moran’s I coefficient reached 0.94 (p < 0.001), providing statistical evidence that the spatial distribution of UHI intensity is highly autocorrelated rather than random. These results confirm that thermally stressed areas are not isolated but tend to form cohesive spatial clusters that align with specific land-use and morphological patterns (Figure 8b). For spatial planning, this spatial coherence is a crucial realization, particularly when guiding targeted mitigation efforts.

3.3. Analytical Objectives

The first implementation objective aimed to determine the extent to which zones identified as thermal-priority areas—defined by NDVI values below 0.2 and surface temperatures exceeding 30 °C—spatially intersect with residential planning zones. The NDVI threshold of <0.2 used to define low-vegetation zones follows commonly accepted classification schemes in remote sensing literature, where values in the 0.1–0.2 range typically indicate sparse or degraded vegetation. This aligns with previous urban UHI studies [12,15], which identified NDVI < 0.2 as representative of built-up or impervious areas with limited vegetative cover. By combining this with a surface temperature threshold (>30 °C), the analysis focuses on zones of concurrent thermal and vegetative stress. Using spatial intersection with the official urban plan layer, a total of 1644 residential parcels (classified as S—residential, or M—mixed-use) were found to overlap with priority zones. This result confirms that many areas facing the greatest surface heat stress are located within zones intended for housing and mixed-use development, underscoring the need for targeted greening strategies in these environments (Figure 9).

To objectively identify regions susceptible to intensified surface heat, a composite UHI index was created by integrating normalized LST and normalized NDVI values obtained from the composite summer rasters, as outlined in Objective 2. To account for both thermal load and a lack of vegetation, each was given equal weight (0.5). The index, generated at the raster-polygon level across the entire research area, shows UHI stress levels ranging from low to high. There were 3537 grid cells categorized as high-risk zones because their composite UHI value exceeded 0.3. These locations are of particular importance for targeted urban cooling interventions (Figure 10).

The third objective is to evaluate the effectiveness of planning indicators—specifically, the ratio of built-up area inside the urban planning unit and the total planned area allocated for green usage, in conjunction with remotely sensed NDVI—in explaining regional variance in the composite UHI index. The model includes three predictors: built-up intensity, planned green space area, and NDVI. A multiple linear regression model showed that all three predictors significantly influence the UHI index (p < 0.001). NDVI had the strongest negative effect (β = −1.10), confirming its critical role in mitigating surface heat accumulation. The share of built-up area had a positive effect (β = 0.006), indicating that more built-up zones are associated with more intense heat. Meanwhile, the total area of planned green zones also had a negative effect (β = −2.4 × 10⁻⁹), though its contribution was notably smaller in magnitude. The model demonstrated excellent explanatory power (R² = 0.893), with low prediction error (RMSE = 0.057; MAE = 0.046). As shown in Figure 11, predicted values align closely with observed UHI values along the 1:1 line, confirming the model’s strength. These results are driven by a combination of land-use planning intensity and vegetation characteristics, suggesting that spatial variations in surface heat intensity are primarily driven by these factors.

The last goal was to find out if the spatial patterns of urban heat buildup show statistically significant clustering, using the composite UHI index as the target variable. We used two types of local spatial autocorrelation: Local Moran’s I (LISA) and Getis-Ord Gi*. We created a spatial weights matrix over the polygon units with determined UHI index values by using Queen contiguity. Local Moran’s I found 3173 important spatial clusters, which were either high-high (hot spots) or low-low (cool places). The Getis-Ord Gi* statistic also found 3166 areas with very high (GiZ ≥ 1.96) or very low (GiZ < −1.96) intensity. The results support Hypothesis 4, indicating that urban heat stress is not randomly distributed but is instead structured in space, with distinct groups of high and low values. These clusters align with distinct patterns in the city’s shape, such as areas where buildings are densely packed and there is limited green infrastructure. The fact that there are spatial clusters like these supports the argument that urban heat mitigation programs should be tailored to specific locations, especially in areas that have consistently high temperatures, which may be areas of compounded sensitivity. To better visualize the spatial intensity of these clusters, we generated a kernel density map using the centroids of LISA-defined high-high zones. By highlighting the city’s central areas of concentrated thermal stress, this heatmap indicates where mitigation resources could be utilized most effectively (Figure 12).

3.4. Synthesis of Analytical Objectives and Key Findings

To strengthen the alignment between the research objectives, tested hypotheses, and corresponding results,

Table 1. Alignment of Research Objectives, Hypotheses, Analytical Methods, and Key Results.

Objective	Hypothesis	Analytical Approach	Key Metrics	Key Results
O1	H1	Bivariate correlation and linear regression at the planning unit level	r = 0.68, R2 = 0.46, p < 0.001	Strong positive correlation between built-up share and mean LST
O2	H2	Bivariate correlation and linear regression at the planning unit level	r = –0.32, R2 = 0.10, p < 0.01	Weak but significant negative correlation between green space and LST
O3	H3	Raster-based Pearson correlation	r = –0.69, p < 0.001	Strong negative correlation between NDVI and LST
O4	H4	Local Moran’s I and Getis-Ord Gi* on composite UHI index	Moran’s I = 0.94, z = 12.9, p < 0.001	Significant spatial clustering of high and low UHI index values

provides a concise summary of the analytical approaches, key statistical metrics (e.g., correlation coefficients, p-values), and main interpretations derived from the findings.

4. Discussion

This study examined the spatial dynamics of UHI in Zagreb using a comprehensive methodology that integrates remote sensing, spatial statistics, and urban planning data. The results confirmed that increased built-up intensity is associated with higher LST, whereas vegetation—especially designated green spaces—provides a quantifiable albeit very minor cooling effect. The research presented a composite UHI measure that integrates normalized LST and NDVI readings, identifying areas where high temperatures and low vegetation coincide. This composite index facilitated a more focused comprehension of thermal vulnerability, particularly in residential areas with inadequate green infrastructure. In addition, spatial autocorrelation techniques, including Local Moran’s I and Getis-Ord Gi*, revealed more than 3000 statistically significant clusters of severe thermal values, substantiating the assertion that urban heat in Zagreb is spatially organized rather than randomly dispersed. The findings possess practical significance for urban planners, as they identify regions that would gain the most from localized greening initiatives and adaptive zoning approaches. The utilization of regulatory urban norms as fundamental spatial units enhances the policy significance of the findings, indicating that green space mandates might be immediately integrated into current planning instruments. A study establishes that the spatial distribution of urban heat in Zagreb is not random; rather, it adheres to a systematic pattern closely associated with land-use intensity and vegetation cover. The results robustly validate H1 and H2, establishing that the proportion of impermeability is a significant determinant of increased LST. In contrast, the implementation of planned green infrastructure demonstrates a mitigating but comparatively lesser impact. These findings align with extensive urban climatology literature [37,38], which consistently emphasizes the dual influence of built-up density and vegetation on urban thermal conditions. The evident negative correlation between NDVI and LST (H3) substantiates the recognized cooling effect of vegetation, affirming that greening initiatives can significantly contribute to thermal regulation, even in densely populated areas, as previously recorded in comparative urban studies throughout Europe and East Asia [39]. Furthermore, the pixel-level correlation aligns with zonal aggregations, providing a crucial layer of methodological validation that is frequently neglected in remote sensing UHI research, where scale discrepancies can lead to significant bias [40]. This research’s primary contribution is the development of a composite UHI index that amalgamates normalized LST and NDVI into a singular metric. This index identifies areas of compounded vulnerability—regions that are simultaneously hot and lacking in vegetation. Unlike studies that analyze thermal and vegetation indicators in isolation, this approach provides a more pertinent risk typology for policy-making. This study builds upon previous research on multi-indicator UHI mapping in cities such as Vienna and Barcelona by directly aligning the composite metric with Zagreb’s planning units, thereby facilitating the practical application of urban regulation and design interventions. Local spatial autocorrelation approaches substantiate Hypothesis 4, demonstrating statistically substantial clustering of extreme values using both Local Moran’s I and Getis-Ord Gi* methods. More than 3000 high-high and low-low clusters were discovered, indicating that heat stress is unevenly distributed and concentrated in spatially persistent zones, especially in residential areas with limited green infrastructure. These findings correspond with other research conducted in cities such as Paris, France [41] and Tokyo, Japan [42], where heat hotspots were observed to coincide with developed, vegetation-deficient areas significantly. The regression model findings related to Objective 3 and Hypothesis 1 exhibited a notably strong explanatory power (R² = 0.89), highlighting the efficacy of integrating regulatory planning variables with remote sensing data. This aligns with recent research highlighting the incorporation of planning datasets in urban climate evaluations [38,43]. Although the green space variable exhibited a lower standardized coefficient than built-up intensity, its normative significance in urban planning is essential. The findings suggest that current spatial plans, particularly those adhering to the urban rules framework, may be lacking in mandatory thresholds for climate-sensitive land use. Consequently, incorporating minimum green space quotas per planning unit could represent a rational progression in policy reform.

Zoning regulations could integrate thermal risk layers to guide land-use intensity, ensuring that compact or high-density developments are balanced with adequate vegetated buffers or ventilation corridors. Minimum green space standards—defined as a percentage of parcel or block area—should be mandated within zoning ordinances, particularly in districts exhibiting high LST and low NDVI values. Urban renewal policies could incorporate composite UHI metrics when assessing redevelopment proposals, ensuring that thermal comfort and climate resilience are prioritized. These policy directions offer a framework for climate-adaptive urban planning in Zagreb and other Central European cities, where traditional planning approaches often lack integrated strategies for heat mitigation.

Additionally, the kernel density mapping of statistically significant high-high clusters offered a crucial visual representation of thermal risk areas. These clusters often align with southern and central residential areas, underscoring prior concerns over spatial thermal imbalance and the need for localized adaptation strategies. The work addresses a significant research gap in Croatia and the wider Central European setting, where comprehensive spatial evaluations of UHI have been infrequent, and planning processes have not completely incorporated climate resilience criteria. The proposed composite method, based on empirical strength and planning relevance, offers a scalable and adaptable framework for cities seeking to identify, assess, and mitigate urban heat vulnerability.

5. Conclusions

This study demonstrated the value of an integrated spatial methodology combining urban planning indicators, remote sensing data, and spatial-statistical analysis to assess the configuration and drivers of urban heat islands in the city of Zagreb. The use of a composite UHI index—derived from normalized LST and NDVI—alongside high-resolution spatial clustering techniques enabled the identification of statistically significant zones of thermal stress that align with patterns of dense construction and insufficient vegetation. Results confirmed that UHI intensity is not randomly distributed but forms spatially coherent clusters, particularly in residential and mixed-use zones with low ecological buffering capacity. Regression models further substantiated the predictive role of land-use intensity and vegetated surfaces, supporting the hypothesis that both built-up coverage and green infrastructure significantly influence thermal exposure. The analytical framework proved effective for pinpointing spatial vulnerabilities and offers a transferable tool for targeting climate adaptation interventions at the local scale. However, certain limitations must be acknowledged. The analysis was based on satellite data from a single summer season (2024), which constrains the ability to infer temporal variability or inter-annual trends. While the analysis is based on four cloud-free Landsat acquisitions during the summer of 2024, this limited temporal scope constrains broader generalization. Urban heat patterns can fluctuate across seasons and years due to atmospheric variability, vegetation phenology, and ongoing land-use change. However, previous research in Zagreb [33] has shown that UHI intensity—particularly the contrast between densely built-up and vegetated areas—remains relatively consistent across multiple summer periods from 2013 to 2022. This suggests that the spatial structure of thermal hotspots is relatively stable over time. Nonetheless, future analyses should incorporate multi-year and seasonal datasets to capture temporal persistence, assess inter-annual climatic variation, and improve the robustness of planning recommendations. Future research could adopt a multi-seasonal, longitudinal approach to assess the persistence and dynamics of UHI formation. Additionally, while urban planning data enriched the spatial interpretation of results, microclimatic variables such as building height, materials, and surface albedo were not included. Incorporating such data could enhance the explanatory depth of the models and refine location-specific mitigation strategies. The spatial configuration of urban heat exposure identified in this study demonstrates not only the concentration of elevated surface temperatures in densely built areas with limited vegetation, but also the structural persistence of thermal risk within specific regulatory planning units. High–high clusters of UHI intensity are predominantly located in the southern and central parts of the city, especially within residential and mixed-use zones that lack sufficient planned green infrastructure and display low NDVI values, despite their formal designation as green areas. The composite UHI index captured these compounded vulnerabilities by integrating vegetation scarcity and thermal intensity into a single, interpretable indicator. Regression analysis further confirmed that built-up density and planned green area coverage significantly influence surface temperature variability. These results indicate that urban climate adaptation in Zagreb should move beyond general greening strategies and instead prioritize targeted mitigation efforts based on spatial risk assessments. This includes the integration of thermal risk into zoning instruments and the implementation of mandatory green space thresholds at the level of urban regulatory units. Finally, the clustering patterns observed point to persistent spatial disparities in exposure, particularly affecting residential areas, which underscores the need to address thermal inequity within broader frameworks of climate resilience and urban planning. The composite UHI index developed in this study functions as a practical and spatially explicit tool for identifying urban areas most exposed to thermal stress, thereby informing targeted climate adaptation strategies. The spatial overlap between residential zones and thermal hotspots underscores a critical issue of spatial equity, as the most heat-vulnerable populations often reside in areas with the least ecological protection and adaptive capacity. This highlights the necessity for integrating thermal vulnerability assessments into urban planning and zoning regulations. While the temporal scope of the analysis is limited to a single summer season, it is important to contextualize this within the extreme climatic conditions of 2024, which was the warmest summer on record both globally and in Europe, according to the Copernicus Climate Change Service and the Croatian Meteorological and Hydrological Service (DHMZ). These exceptional conditions likely intensified surface temperature contrasts in Zagreb and may have amplified the observed spatial patterns of UHI distribution. Therefore, caution is warranted when extrapolating these results beyond the analyzed temporal frame. Future research should aim to expand the spatial coverage to include the broader metropolitan region and enhance the generalizability of findings through integration of harmonized land-use/land-cover (LULC) classifications, such as those from Copernicus. Additionally, the methodological framework could be refined by incorporating alternative or supplementary remote sensing indices, including the Enhanced Vegetation Index 2 (EVI2) and the Normalized Difference Built-up Index (NDBI), to better capture the complex interplay between built surfaces and ecological buffers in shaping urban thermal dynamics.