Surface Urban Heat Island Variability in Cosenza, Southern Italy: The Roles of Urban Morphology, Extreme-Temperature Conditions, and Local Wind Conditions

Abstract

Mediterranean cities with complex terrain are increasingly exposed to overheating. This study investigates Surface Urban Heat Island Intensity (SUHII) in Cosenza, southern Italy, in relation to land-use, neighborhood-scale morphology, extreme-temperature conditions and local wind. Summer ECOSTRESS Land Surface Temperature (LST) observations from 2018 to 2025 were integrated with CORINE Land Cover, CLC+ Backbone, urban morphological indicators, and meteorological data. Continuous urban fabric exhibited the strongest SUHII, with median values reaching about 6.5 °C at midday, whereas less compact urban areas exhibited lower thermal contrasts. A bin-based analysis revealed clear within-class morphological gradients, particularly in discontinuous urban fabric and industrial/commercial areas, where increasing building height, building surface fraction, and impervious surface fraction were generally associated with higher SUHII; these relationships were weaker and less systematic in compact urban fabric. TX90p conditions were associated with lower SUHII than No-TX90p conditions, likely reflecting stronger warming of rural reference surfaces and a reduced urban–rural thermal gradient. Wind-related differences were generally modest and non-systematic across land-use classes, TX90p categories, and diurnal periods. Overall, SUHII variability reflects both categorical land-use differences and continuous neighborhood-scale morphological gradients, while extreme-temperature and ventilation conditions further modulate urban–rural surface thermal contrasts in this Mediterranean valley city under complex topographic conditions.

1. Introduction

Urban areas are increasingly exposed to urban overheating due to the combined effects of global climate change and localized anthropogenic modifications in land-use and urban form. One of the most widely investigated manifestations of urban overheating is the Urban Heat Island (UHI), which describes thermal differences between urban areas and their rural surroundings. Depending on the variable considered, this phenomenon can be examined through near-surface air temperature or through Land Surface Temperature (LST), the latter being commonly used to characterize the Surface Urban Heat Island (SUHI). UHI formation is mainly driven by the replacement of natural and vegetated surfaces with artificial and impervious materials, together with increased building density, modified urban canyon geometry and reduced evaporative cooling. These changes alter the surface–atmosphere energy balance by enhancing daytime heat storage, reducing evapotranspiration and limiting nocturnal radiative cooling, thereby increasing heat-risk exposure within cities [1].

Among the main factors associated with urban overheating, urban morphology and land-cover composition play a key role in controlling the magnitude and spatial variability in UHI and SUHI patterns. These relationships can be quantified using neighborhood-scale morphological indicators, such as mean building height (HM), building surface fraction (BSF), impervious surface fraction (ISF), and pervious surface fraction (PSF), which provide continuous measures of urban form and surface composition beyond categorical land-use classifications [2,3,4]. As a result, compact urban fabrics generally exhibit stronger surface warming and slower cooling than less urbanized or more permeable areas.

Beyond land cover and morphology, wind circulation is a key dynamic factor that may modulate UHI intensity and its diurnal evolution. In areas characterized by complex terrain, a mountain–valley wind system, cold-air drainage and topographically channeled flows may influence the spatial distribution of heat and, under specific conditions, contribute to reducing nocturnal overheating by favoring air exchange between rural surroundings and the urban core [5,6,7]. During daytime, wind intensity can also affect turbulent mixing and advective heat transport, thereby influencing peak surface temperatures. Recent studies suggest that UHI attenuation may occur as local wind speed increases, although the magnitude of this effect depends on urban morphology, regional climate and topographic setting [8]. This highlights the need to consider local ventilation conditions when interpreting intra-urban thermal variability.

Urban–rural surface thermal contrasts may also vary under extreme air-temperature conditions. During days characterized by unusually high maximum air temperatures, both urban and rural surfaces can experience substantial warming, potentially altering the magnitude of SUHII. These interactions may increase absolute temperatures and modify urban–rural thermal contrasts, with important implications for heat exposure, human health and climate adaptation planning [4,9,10,11]. Previous studies have shown that UHI and SUHI responses to extreme heat are not spatially or climatically uniform [12,13]. Depending on surface moisture, vegetation state, land-cover composition and rural reference characteristics, extreme-temperature conditions may be associated with either stronger or weaker urban–rural surface temperature gradients. Percentile-based temperature classifications therefore provide a useful framework for examining whether SUHII differs between extreme and more typical summer conditions.

Methodologically, urban thermal environments can be investigated using either near-surface air temperature observations, which describe the atmospheric or canopy-layer UHI, or satellite-derived LST, which is used to quantify the SUHI. Meteorological stations provide high temporal resolution and are particularly suitable for assessing nocturnal atmospheric UHI dynamics, but their spatial representativeness is often limited by station distribution and siting conditions. Conversely, thermal remote sensing provides spatially continuous information on surface thermal patterns, although it is constrained by satellite revisit frequency, cloud contamination and overpass time [14].

To capture fine-scale urban thermal heterogeneity, this study uses LST data from ECOSTRESS, the Ecosystem Spaceborne Thermal Radiometer Experiment on Space Station. ECOSTRESS is a thermal infrared sensor developed by NASA’s Jet Propulsion Laboratory and operating aboard the International Space Station since 2018. It provides LST and emissivity observations at approximately 70 m spatial resolution [15]. This resolution is particularly suitable for medium-sized cities and heterogeneous urban–rural transitions [16]. Moreover, its non-sun-synchronous orbit provides observations at varying local times, offering an opportunity to investigate SUHII across different phases of the diurnal cycle rather than at a single fixed satellite overpass time (such as MODIS or Sentinel-3) [17,18].

Despite the growing body of literature on SUHI dynamics, few studies have jointly examined sub-daily SUHII variability, neighborhood-scale urban morphology, extreme air-temperature conditions, and local wind conditions within Mediterranean cities characterized by complex terrain. These factors are often investigated separately, making it difficult to assess how surface thermal contrasts vary across morphological gradients and meteorological conditions within the same observational framework.

Valley and basin cities are particularly relevant in this context because local wind channeling, cold-air drainage, topographic gradients and heterogeneous rural reference surfaces can substantially affect urban–rural thermal contrasts. Cosenza provides a suitable case study because its compact and discontinuous urban fabrics coexist within a narrow valley (i.e., Crati valley) surrounded by complex terrain, allowing morphological and meteorological influences on SUHII to be examined using a multi-year dataset.

This study investigates SUHII dynamics in Cosenza by combining multi-year ECOSTRESS LST observations with land-use classifications, neighborhood-scale urban morphological indicators and meteorological data from the ARPACAL regional monitoring network. Specifically, the objectives are to: (i) quantify the spatial and diurnal variability in SUHII across distinct land-use classes using high-resolution ECOSTRESS LST data; (ii) assess SUHII variability across the three urban land-use classes (CLC1–CLC3) in relation to their neighborhood-scale morphological characteristics, complemented by a bin-based analysis of the relationships between SUHII and HM, BSF and ISF; (iii) compare SUHII behavior between TX90p exceedance and No-TX90p days; and (iv) examine SUHII variability under different local wind-speed conditions.

By coupling multi-year satellite observations with ground-based meteorological measurements, this research provides an integrated characterization of how SUHII varies in relation to urban form, land-use configuration, extreme-temperature conditions and local wind conditions within a complex topographic setting. The results aim to inform the interpretation of urban overheating and contribute to future climate-sensitive planning assessments in Mediterranean cities characterized by complex terrain.

2. Methodology

2.1. Study Area

The study was conducted in the urban area of Cosenza, located in the central–northern part of Calabria, southern Italy (approximately 39.18° N, 16.25° E; Figure 1). The municipality covers an area of about 37 km² and has a population of approximately 64,000 inhabitants [19]. Cosenza is situated within the Crati River valley at an average elevation of about 238 m above sea level (a.s.l.) and is bounded by two major mountain systems: the Sila Massif to the east and the Coastal Range (Catena Costiera) to the west. This valley-bottom setting strongly characterizes the local geographical context and can influence atmospheric ventilation, thermal inversions and local thermally driven circulations.

The urban structure of Cosenza develops mainly along a north–south axis. The historical and compact urban core is located in the southern part of the city, near the confluence of the Crati and Busento rivers and the slopes of Colle Pancrazio. More recent residential and commercial expansion extends northward across the flatter valley floor, forming an almost continuous urban agglomeration with the neighboring municipality of Rende. The administrative boundary between the two municipalities largely follows the Campagnano River. Industrial and commercial areas are mainly located in the eastern and north-western sectors of the outer urban belt.

According to the Köppen–Geiger climate classification (source for classifications: https://koeppen-geiger.vu-wien.ac.at, URL accessed on 4 September 2026), Cosenza has a Mediterranean climate with hot, dry summers and mild, wetter winters (Csa). However, the valley morphology and the surrounding relief can induce marked local microclimatic effects. During summer, the surrounding mountains may reduce the penetration of regional synoptic winds, favoring episodes of weak ventilation, atmospheric stagnation and intense daytime heating. At the same time, the topographic setting may promote thermally driven circulations, including mountain–valley wind and cold-air drainage processes, which can interact with the urban fabric and influence the spatial distribution of heat.

For the analysis, a 20 km × 20 km study domain centered on Cosenza was defined to include the main urbanized area, the surrounding peri-urban sectors and a sufficiently broad set of rural reference surfaces. The domain also includes the meteorological stations used in this study for wind observations and heatwave identification. Their characteristics and specific roles are described in Section 2.4.

2.2. Land-Use Datasets and ECOSTRESS LST Observations

Land-use information was obtained from the 2018 CORINE Land Cover dataset, produced within the European Union’s Copernicus Land Monitoring Service (https://collections.eurodatacube.com/, accessed on 10 July 2026). The dataset is derived from high-resolution satellite imagery and classifies the European territory into 44 land-cover and land-use categories (https://land.copernicus.eu/content/corine-land-cover-nomenclature-guidelines/html/, accessed on 10 July 2026). The standard product has a spatial resolution of 100 m and is periodically updated to provide spatially consistent information on land-cover changes across Europe. The standard CORINE Land Cover product was used to identify the main land-use classes considered in the analysis. Three urban classes representative of built-up environments were selected: continuous urban fabric (CLC1), discontinuous urban fabric (CLC2), and industrial or commercial units (CLC3).

To improve the identification of rural reference areas (CLC4) CORINE Land Cover information was integrated with the higher-resolution CLC+ Backbone 2023 dataset including all CLC categories 2–9, representing agricultural, forest, semi-natural, and other non-urban land covers. Following the methodology proposed by Esposito et al. [20], to prevent thermal contamination from nearby built-up areas, a 500 m exclusion buffer was applied around the urban classes CLC1–CLC3, and an additional 100 m buffer was applied around sealed artificial surfaces (category 1) identified from the CLC+ Backbone 2023 layer This procedure was adopted to ensure that the rural reference class was minimally affected by nearby urban or impervious surfaces.

Also, to limit topographic heterogeneity within the study area, all areas located above 500 m above sea level were excluded using the TINITALY digital elevation model (DEM), provided by the National Institute of Geophysics and Volcanology (INGV) (https://tinitaly.pi.ingv.it/ accessed on 29 July 2026).

The resulting land-use classification used for the LST and SUHII analyses is shown in Figure 2a. It should be noted that the use of CORINE Land Cover classes introduces a spatial-scale limitation, because CLC is designed for regional land-cover mapping and may not fully capture fine-scale urban heterogeneity or mixed land-cover conditions at the scale of ECOSTRESS pixels.

Because Cosenza is located in a complex valley setting, the rural reference class may include surfaces with different topographic and vegetation characteristics from the urban classes located mainly along the valley floor. To assess the comparability of the urban and rural reference surfaces, the elevation and slope characteristics of each land-use class are summarized in

Table 1. Topographic characteristics of the land-use classes used in the SUHII analysis. Elevation and slope statistics were computed over the pixels belonging to each class within the 20 km × 20 km study domain.

Class	Area (km2)	Elevation Min (m)	Elevation Max (m)	Slope Median (deg)
CLC1	5.35	176.52	346.24	2.81
CLC2	17.66	146.81	388.65	4.89
CLC3	4.71	145.06	420.60	1.72
CLC4	130.10	129.02	468.88	13.16

. This additional characterization was used to support the interpretation of SUHII values and to identify possible topographic influences on the urban–rural LST contrast.

CLC4 showed a wider elevation range than the urban classes and included surfaces located outside the main valley floor. Therefore, SUHII values should be interpreted as urban–rural thermal contrasts within a complex topographic domain rather than as effects exclusively attributable to urbanization. This aspect is explicitly considered in the interpretation of the results and discussed as a potential source of uncertainty.

LST data were obtained from the ECOSTRESS Level-2 Land Surface Temperature and Emissivity product, distributed by the NASA Land Processes Distributed Active Archive Center (LP DAAC) (https://lpdaac.usgs.gov/, accessed on 16 July 2026). ECOSTRESS is a thermal infrared sensor developed by NASA’s Jet Propulsion Laboratory and operating aboard the International Space Station since 2018. The product provides instantaneous LST and emissivity observations at approximately 70 m spatial resolution across five thermal infrared bands, making it suitable for resolving intra-urban thermal variability in medium-sized cities characterized by heterogeneous land-cover mosaics [15].

Approximately 250 ECOSTRESS scenes acquired during the summer months, June, July and August (JJA), over the period 2018–2025 were initially collected for the Cosenza study domain. A multi-step quality-control procedure was then applied to retain only high-quality observations. First, scenes with total cloud cover exceeding 10% within the study domain were excluded. Second, residual cloud contamination was further reduced through a spatial screening procedure: for each scene, a 3 km buffer was generated around cloud-contaminated pixels, and scenes affected by clouds within this buffer were discarded. After the filtering procedure, 127 high-quality ECOSTRESS scenes were retained for the subsequent analyses. Of these, 19 scenes were acquired between 03:00 and 06:00, 37 between 07:00 and 10:00, 26 between 11:00 and 14:00, 26 between 15:00 and 18:00, 10 between 19:00 and 22:00, and 9 between 23:00 and 02:00. For each retained scene, LST values were extracted for the four land-use classes. Prior to aggregation, a percentile-based filtering procedure was applied by excluding pixels outside the 1st–99th percentile range of the LST distribution within each scene and land-use class. This step was introduced to reduce the influence of residual anomalous pixels, and possible artifacts related to undetected cloud contamination or surface emissivity uncertainty. The same percentile-based filtering was applied consistently to all classes and scenes to avoid introducing class-specific biases.

For the LST and SUHII boxplot analyses, valid ECOSTRESS pixels were pooled according to land-use class and satellite overpass time interval. For each ECOSTRESS scene, the rural reference LST used for SUHII calculation was defined as the median LST of valid rural pixels characterized by a pervious surface fraction (PSF) ≥0.95. Following the workflow presented in Esposito et al. [20], pixel-level SUHII values were then calculated by subtracting this scene-specific rural reference LST from the LST of each valid urban pixel belonging to CLC1, CLC2 and CLC3. Because the resulting boxplots are based on pooled pixel-level observations, they are used here to describe spatial and temporal distributions of LST and SUHII rather than to provide formal statistical inference. Pixels belonging to the same ECOSTRESS scene may be spatially autocorrelated and share the same meteorological forcing; therefore, they should not be interpreted as fully independent temporal observations.

Because ECOSTRESS acquires observations at variable local times due to the non-sun-synchronous orbit of the International Space Station, the retained scenes were grouped into six time intervals representative of different phases of the diurnal cycle [20]. For each time interval and land-use class, LST distributions were summarized using boxplots showing the median, interquartile range and dispersion of the data, as illustrated in Figure 2b. This procedure provided the basis for the subsequent estimation of SUHII, calculated by comparing urban LST values with the rural reference class.

2.3. Urban Morphological Indicators

To characterize the built environment and its potential influence on surface thermal patterns, four neighborhood-scale morphological indicators were derived for the study area, HM, BSF, ISF and PSF, following the methodological framework of Esposito et al. [4,21]. The indicators were calculated using a regular 500 m × 500 m grid, which was adopted to represent neighborhood-scale variations in urban form and surface-cover composition. This grid size allows the main morphological differences among compact urban fabric, discontinuous urban areas, industrial or commercial zones, and rural surroundings to be described consistently.

The geometric indicators HM and BSF were derived from the Global Building Atlas dataset [22], which provides spatially explicit information on building footprints and estimated building heights. HM was calculated as the average height of all buildings within each grid cell:

$$\mathrm{H}\mathrm{M}={\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{N}}}{\mathrm{H}}_{\mathrm{i}}}{\mathrm{N}}}$$

(1)

where H_i is the height of building i and N is the total number of buildings within the grid cell.

BSF was calculated as the ratio between the total building footprint area and the total grid-cell area:

$$\mathrm{B}\mathrm{S}\mathrm{F}={\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{N}}}{\mathrm{A}}_{\mathrm{i}}}{{\mathrm{A}}_{\mathrm{T}}}}$$

(2)

where A_i is the footprint area of building i and A_t is the total area of the grid cell.

Surface-cover indicators were derived from the CLC+ Backbone 2023 dataset. Impervious surfaces were identified using the “Sealed” class (category 1), whereas pervious surfaces were obtained by aggregating all non-sealed classes (categories 2–9), including vegetated, agricultural, natural and semi-natural surfaces. The impervious surface fraction and pervious surface fraction were calculated for each grid cell as:

$$\mathrm{I}\mathrm{S}\mathrm{F},\mathrm{P}\mathrm{S}\mathrm{F}={\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{N}}}{\mathrm{A}}_{\mathrm{i}}}{{\mathrm{A}}_{\mathrm{T}}}}$$

(3)

where A_i represents the area of the pixels classified as sealed when calculating ISF, or the area of the pixels classified as non-sealed when calculating PSF, within each grid cell; A_t is the total grid-cell area.

All spatial operations were performed in QGIS version 3.22.1 (version 3.22.1 https://download.osgeo.org/qgis/windows/, URL accessed on 4 September 2026) using the Zonal Statistics tool and the Field Calculator. Raster-based metrics were aggregated within each 500 m × 500 m grid cell, while building-footprint information was used to compute the geometric indicators [23].

The spatial distribution of the four morphological indicators is shown in Figure 3. The resulting maps describe the main urban–rural morphological gradients within the study domain. HM, BSF and ISF generally reach their highest values in the compact urban fabric, whereas lower values are observed in discontinuous urban areas, industrial or commercial zones, and rural surroundings. Conversely, PSF shows the opposite behavior, with lower values in compact urban areas and higher values in peripheral and rural sectors.

A statistical summary of the morphological indicators for the three urban land-use classes is reported in

Table 2. Mean, maximum and minimum values of the morphological indicators for the urban land-use classes.

	HM (m)	BSF (−)	ISF (−)	PSF (−)
CLC1
Mean	10.50	0.22	0.70	0.30
Max	12.30	0.25	0.85	0.45
Min	8.80	0.19	0.55	0.15
CLC2
Mean	7.10	0.11	0.48	0.52
Max	11.90	0.16	0.90	0.86
Min	4.50	0.03	0.14	0.10
CLC3
Mean	7.20	0.11	0.47	0.52
Max	8.60	0.18	0.65	0.92
Min	6.00	0.04	0.08	0.35

. CLC1 shows the highest mean HM, BSF and ISF values, confirming its role as the most compact and impervious urban class. CLC2 and CLC3 exhibit lower building density and higher proportions of permeable surfaces, although both classes show substantial intra-class variability. These morphological differences were used to support the interpretation of LST and SUHII variability across the different land-use classes. The morphological indicators were therefore used as descriptive class-level characteristics rather than as independent predictors in a statistical model of SUHII.

To complement the land-use class comparison, a bin-based analysis was performed for HM, BSF, and ISF using the 500 m × 500 m grid. Before binning, HM was rescaled to the 0–1 range using min–max normalization, whereas BSF and ISF were already expressed as dimensionless fractions. Within each urban land-use class, grid cells were grouped into predefined intervals of each indicator, resulting in three bins for CLC1 and CLC3 and four bins for CLC2. For each bin, the median morphological value and mean SUHII were calculated. Only bins containing at least five valid grid cells were retained, and error bars represent the standard error of the mean.

The analysis was used to characterize systematic SUHII variations along morphological gradients and was not intended to isolate independent or causal effects of individual morphological parameters.

2.4. Meteorological Data and Heatwave Identification

Meteorological data were obtained from the regional monitoring network managed by the Civil Protection Agency of Calabria and ARPACAL (https://www.cfd.calabria.it/index.php/dati-stazioni/stazioni-monitoraggio accessed on 23 July 2026) and made available through the Allerta Calabria platform. Air temperature and wind observations were collected from two meteorological stations located within or around the Cosenza study domain (Figure 1): Cosenza 118, ID 1017 (39.305297° N, 16.234667° E; 334 m a.s.l.) and Torano Scalo, ID 1130 (39.493831° N, 16.209822° E; 97 m a.s.l.)

Station ID 1017 was selected as the urban reference station for the characterization of local wind conditions. This station shows the highest degree of urbanization among the available stations, although its surrounding footprint still includes a substantial proportion of permeable surfaces. Station ID 1130 was used for heatwave identification because long-term daily maximum temperature data were available for the climatological reference period 1961–1990.

To characterize the local environment surrounding each meteorological station, the same morphological indicators used for the gridded analysis were computed within a 500 m radius buffer around each station. The resulting values are reported in

Table 3. Morphological characteristics computed within a 500 m radius buffer around the meteorological stations used in the study.

NAME	ID	HM (m)	BSF (−)	ISF (−)	PSF (−)
Cosenza 118	1017	10.84	0.18	0.20	0.62
Torano Scalo	1130	5.50	0.10	0.14	0.77

.

Wind observations from station ID 1017 were available at 15 min resolution for the period 2018–2025. Wind speed and wind direction were used to characterize local ventilation conditions during the ECOSTRESS acquisition times. Wind data were temporally matched to each ECOSTRESS acquisition by selecting the observation closest to the satellite overpass time. A maximum time difference of ±30 min was allowed. Scenes without an available wind observation within this interval were assigned as a missing value. For the SUHII–wind analysis, wind speed was classified into two empirical wind-intensity categories: WI1, corresponding to wind speed ≤3 m s⁻¹, and WI2, corresponding to wind speed >3 m s⁻¹. This threshold was selected to distinguish weak wind from more ventilated conditions within the observed wind-speed distribution.

Heatwaves were identified using a Warm Spell Duration Index (WSDI)-based approach. For each calendar day of the reference period 1961–1990, the 90th percentile of daily maximum air temperature (Tmax) was calculated using data from station ID 1130. Daily Tmax values for the study period 2018–2025 were then compared with the corresponding calendar-day threshold. Days exceeding the 90th percentile were first classified as TX90p exceedance days, namely warm-threshold exceedance days [24]. Heatwave events were then defined as sequences of at least six consecutive TX90p exceedance days, following the WSDI criterion adopted in heatwave climatology [24,25].

3. Results

This section presents the main results of the study, focusing on the spatial and temporal variability in surface temperature in Cosenza. First, the annual and diurnal variability in ECOSTRESS-derived LST is analyzed across the selected land-use classes. The SUHII is then examined in relation to the diurnal phase, TX90p conditions, neighborhood-scale morphology, and wind-speed conditions. Please note that the boxplot analyses presented in this section describe pooled pixel-level distributions of LST and SUHII.

3.1. Annual Variability in LST

Figure 4 shows the annual distribution of ECOSTRESS-derived LST across the four land-use classes (CLC1–CLC4) during the summer months (JJA) over the period 2018–2025. The number of valid ECOSTRESS observations varies among years, with a limited number of acquisitions available in 2018 and a larger number of scenes in subsequent years. Therefore, interannual comparisons should be interpreted considering differences in sampling frequency and satellite overpass timing.

Across the analyzed period, urban classes generally exhibit higher LST values than the rural reference class (CLC4), indicating higher LST values in the urban classes, although part of the observed contrast may also reflect topographic and vegetation differences between urban and rural pixels. Among the urban categories, CLC1 usually shows the highest LST values, consistent with its compact built-up structure and higher impervious surface fraction. CLC2 tends to show lower LST values within the urban classes, likely reflecting its more heterogeneous surface composition and higher proportion of permeable areas. CLC3 displays intermediate thermal behavior, although its response varies depending on year, acquisition time and local surface composition.

Overall, the relative thermal ranking among land-use classes remains broadly consistent across the study period, with the compact urban fabric showing the strongest surface warming and the rural reference class showing lower LST values. However, some interannual differences are observed, which may reflect variations in meteorological conditions, heatwave occurrence, satellite sampling time and the number of valid ECOSTRESS acquisitions available for each year. High-end LST values occasionally approach or exceed 50 °C, particularly during daytime observations over densely urbanized areas, indicating the occurrence of localized surface hotspots.

Figure 5 provides a complementary view of LST variability by showing the distribution of valid ECOSTRESS observations according to the year and acquisition time interval. Only scenes that passed the quality-control procedures were included. Although temporal sampling was not uniform across years or time slots, the dataset covers different phases of the diurnal cycle and allows the main evolution of summer surface temperatures to be characterized. Daytime acquisitions, particularly those collected between 07:00 and 18:00 local time, were more frequent than evening and nighttime observations, consistently with the temporal distribution of ECOSTRESS acquisitions reported by Esposito et al. [20]. Nighttime coverage was more limited and was absent for some time intervals and years, whereas 2025 provided a more complete representation of the diurnal cycle. Consequently, part of the apparent interannual variability may reflect differences in acquisition timing rather than actual year-to-year changes in thermal conditions.

Despite this irregular sampling, a clear diurnal pattern emerges. The highest LST values were generally observed during the central daytime and afternoon intervals, particularly between 11:00–14:00 and 15:00–18:00, when strong incoming solar radiation produces maximum surface heating. Conversely, the lowest temperatures occurred during nighttime and early-morning acquisitions. Daytime distributions were also generally wider, with larger interquartile ranges and more extended whiskers, indicating greater spatial and temporal variability associated with solar forcing, meteorological conditions and differences in surface properties. Nighttime distributions were narrower, suggesting more homogeneous and statistically stable surface thermal conditions after sunset. Similar differences between daytime and nighttime LST variability have been reported by Esposito et al. [20,26] who found that nighttime thermal patterns tend to be less variable than daytime conditions.

The non-uniform temporal distribution of the observations is partly related to the orbital characteristics of the International Space Station, which result in irregular ECOSTRESS overpass times and the absence of some time-of-day classes in individual years. This limitation should therefore be considered when interpreting differences among annual LST distributions. Nevertheless, the consistency of the observed diurnal cycle indicates that solar forcing remains the primary driver of summer LST variability, while interannual meteorological conditions and acquisition timing mainly modulate its magnitude. These annual and diurnal LST patterns provide the basis for the subsequent SUHII analysis, in which urban surface temperatures are compared with the rural reference class to quantify urban–rural thermal contrasts.

3.2. Surface Urban Heat Island Intensity and Diurnal Variability

As described in the Methodology section, pixel-level SUHII was calculated for each ECOSTRESS scene and urban class by subtracting the median LST of valid rural CLC4 pixels with PSF ≥ 0.95 from the LST of each valid urban pixel.

Figure 6 shows the diurnal variability in pixel-level SUHII distributions across the three urban land-use classes. Median SUHII values remain positive across all time intervals, indicating that valid urban pixels were generally warmer than the scene-specific rural reference during summer. This behavior differs from some Mediterranean coastal cities, where daytime inverse SUHI conditions may occur when rural or peri-urban surfaces become warmer than the urban core. The predominantly positive SUHII values may partly reflect the combined effects of the selected rural reference, the inland topographic setting, and urban surface characteristics.

A marked diurnal cycle is observed. SUHII increases from the early morning hours and reaches its maximum during the central daytime interval, particularly between 11:00 and 14:00. During this period, CLC1 shows the highest SUHII, with median values reaching approximately 6.5 °C, followed by CLC3 at about 5.6 °C and CLC2 at about 4.7 °C. These peak values suggest the presence of strong daytime surface thermal contrasts in the pooled pixel-level distributions, especially in compact and highly urbanized areas.

After midday, SUHII gradually decreases during the afternoon and evening. During the 15:00–18:00 interval, median values remain relatively high, generally ranging between approximately 3.2 and 4.3 °C, before further decreasing during the evening hours. At night, SUHII reaches its lowest values, generally between approximately 1.0 and 1.9 °C, but remains positive. This indicates that the urban–rural surface thermal contrast weakens after sunset, although urban pixels remain warmer than the rural reference after sunset, a pattern consistent with slower nocturnal surface cooling.

Differences among land-use classes are most evident during daytime. CLC1 generally shows the highest pixel-level SUHII distributions, particularly during the central daytime interval. CLC3 generally shows intermediate-to-high SUHII values and can approach CLC1 during the hottest hours, probably due to the presence of large roofs, paved areas and exposed artificial surfaces typical of industrial and commercial zones. CLC2 exhibits lower SUHII values, likely reflecting its more discontinuous urban structure and higher proportion of permeable surfaces.

Overall, the pixel-level distributions indicate a persistent positive Surface Urban Heat Island pattern in Cosenza, with maximum intensity during the central daytime hours and reduced but still detectable urban–rural surface thermal contrasts at night. The different behavior of CLC1, CLC2 and CLC3 highlights the importance of distinguishing among urban land-use classes when describing surface thermal patterns.

3.3. Relationship Between SUHII and Urban Morphological Indicators

The bin-based analysis provides a more detailed characterization of the relationship between neighborhood-scale morphology and SUHII within the three urban land-use classes (Figure 7). The clearest morphological gradients are observed for CLC2 and CLC3, where increasing HM, BSF and ISF are consistently associated with increasing SUHII across the different phases of the diurnal cycle. In CLC2, daytime and afternoon/evening mean SUHII values span approximately 1.8–4.8 °C across the morphological bins, with the largest gradient generally associated with ISF. At night, the overall magnitude decreases substantially, but the positive morphological gradient remains evident, with values ranging approximately between 0.7 and 2.3 °C.

A similarly clear, and in some cases stronger, gradient is observed for CLC3. During daytime, mean SUHII ranges from approximately 2.1 to 5.8 °C across the morphological bins, while afternoon/evening values vary between approximately 2.4 and 5.1 °C. The strongest variation is associated with BSF, although increasing HM and ISF also show consistent positive gradients. At night, SUHII decreases to approximately 0.7–1.8 °C, but the ordering among morphological bins is largely preserved. CLC1 exhibits a less systematic response. SUHII values vary within a narrower range than in CLC2 and CLC3, approximately 4.0–5.1 °C during daytime, 3.7–5.0 °C during the afternoon/evening, and 2.0–2.7 °C at night. A positive gradient is visible for HM and ISF during daytime, whereas the relationships become weaker or non-monotonic during the other periods. BSF, in particular, does not show a consistent positive gradient. This behavior may partly reflect the relatively restricted morphological variability in the compact urban class and the smaller number of grid cells available within individual bins.

Overall, the bin-based analysis shows that the differences among the three land-use classes are accompanied by clear within-class morphological gradients, particularly for CLC2 and CLC3. Higher building height, building surface fraction and impervious surface fraction are generally associated with stronger urban–rural surface thermal contrasts, whereas these relationships are less evident within the already compact and highly urbanized CLC1 class. These results complement the categorical CLC analysis by showing that SUHII variability also occurs along continuous neighborhood-scale morphological gradients.

3.4. Characterization of TX90p Conditions and Warm-Spell Persistence

After characterizing the diurnal variability in SUHII under summer conditions, extreme-temperature conditions were characterized using the TX90p classification described in Section 2.4 (Figure 8). This analysis was performed to quantify the occurrence, intensity, and persistence of warm-threshold exceedance conditions during the 2018–2025 ECOSTRESS observation period and to provide the climatic context for the subsequent comparison of SUHII between TX90p and No-TX90p days.

Figure 8a compares the distributions of daily maximum air temperature (Tmax) during TX90p exceedance and No-TX90p days for each year of the study period. As expected from the percentile-based classification, TX90p exceedance days exhibit systematically higher Tmax values than No-TX90p days. During TX90p exceedance days, Tmax frequently reaches or exceeds 40 °C, whereas No-TX90p days generally exhibit lower and more variable maximum temperatures. The figure therefore characterizes the magnitude and interannual distribution of the air-temperature conditions associated with the two categories rather than providing an independent validation of the TX90p classification.

Figure 8b shows the duration and intensity of consecutive TX90p exceedance sequences. No sequences reaching the six-day threshold occurred in 2018 and 2020, whereas persistent warm spells were identified in all other years. In 2019 and 2021, multiple sequences were observed, with durations of up to approximately 12 consecutive days. Longer events occurred in 2022 and 2025, reaching about 14 and 13 days, respectively, while 2023 was characterized by fewer persistent sequences, including one lasting approximately 9 days and showing relatively high thermal intensity. The longest event of the study period occurred in 2024, when one sequence persisted for about 15 consecutive days.

Overall, the analyzed period was characterized by recurrent TX90p exceedance days and, in most years, by persistent sequences of extreme-temperature conditions. Given the relatively short 2018–2025 record, these results are intended only to characterize the thermal conditions occurring during the ECOSTRESS observation period and should not be interpreted as evidence of long-term climatic variability or trends. Importantly, the six-day sequence criterion is used here only to describe the persistence of TX90p conditions. The subsequent SUHII analysis is based on the classification of individual ECOSTRESS acquisition days as TX90p or No-TX90p and does not distinguish between heatwave and non-heatwave events.

3.5. SUHII Under TX90p and No-TX90p Conditions

This section compares SUHII variability under TX90p and No-TX90p conditions across the three urban land-use classes and different phases of the diurnal cycle. ECOSTRESS observations were classified according to the TX90p-based classification described in Section 2.4. To account for diurnal variability, the dataset was grouped into three broad time intervals: daytime (06:00–15:00), afternoon/evening (15:00–20:00), and nighttime (20:00–06:00). The TX90p subset comprised 50 ECOSTRESS scenes, including 29 daytime, 13 afternoon/evening, and eight nighttime acquisitions, whereas the No-TX90p subset comprised 77 scenes, including 48 daytime, 15 afternoon/evening, and 14 nighttime acquisitions.

Figure 9 shows that pixel-level SUHII distributions remain generally positive under both TX90p and No-TX90p conditions, confirming that urban surfaces are typically warmer than the rural reference areas during summer. However, median SUHII values are generally lower during TX90p conditions than during No-TX90p conditions. During the daytime, median SUHII reaches approximately 3.9 °C for CLC1, 3.1 °C for CLC2, and 3.6 °C for CLC3 under TX90p conditions, compared with approximately 4.5 °C, 3.3 °C, and 3.7 °C, respectively, during No-TX90p conditions.

A similar pattern is observed during the nighttime. Under TX90p conditions, median nighttime SUHII values are approximately 1.9 °C for CLC1, 1.0 °C for CLC2, and 0.8 °C for CLC3, whereas under No-TX90p conditions they increase to approximately 2.7 °C, 1.5 °C, and 1.6 °C, respectively. These pixel-level distributions suggest that TX90p conditions are associated with a reduced urban–rural surface thermal gradient within the analyzed scenes.

The lower pixel-level SUHII observed during TX90p conditions may be related to the stronger warming of rural reference surfaces under extreme synoptic heat. During TX90p conditions, rural and peri-rural areas may experience intense solar heating, reduced soil moisture availability, lower evaporative cooling, and vegetation drying, particularly under Mediterranean summer conditions. Consequently, the temperature difference between urban and rural surfaces decreases, even though absolute LST values remain very high across the entire study domain.

Differences among land-use classes remain evident under both TX90p and No-TX90p conditions. CLC1 consistently exhibits the highest SUHII values, reflecting the thermal behavior of compact and highly impervious urban fabric. CLC2 generally shows the lowest SUHII values, consistent with its more discontinuous structure and higher proportion of permeable surfaces, whereas CLC3 displays intermediate values. Overall, the analysis indicates that TX90p conditions were associated with lower SUHII distributions while preserving the relative differences among urban land-use classes.

Because ECOSTRESS acquisitions are not evenly distributed between TX90p and No-TX90p conditions, these results should be interpreted with caution, considering the potential influence of differences in sample size and satellite overpass timing between the two groups.

3.6. SUHII and Wind Conditions

Wind conditions were analyzed to evaluate their potential role in modulating SUHII under both TX90p and No-TX90p conditions. Wind observations from station ID 1017 were used because this station provides the available local measurements of wind speed and direction used to characterize wind conditions at the ECOSTRESS acquisition times.

3.6.1. Typical Wind Pattern

A preliminary characterization of the local wind regime was performed using wind speed and wind direction observations from station ID 1017 over the period 2018–2025 (Figure 10). This information provides the background for interpreting the subsequent SUHII–wind analysis, particularly in relation to the possible role of ventilation and valley-channeled flows in modulating urban–rural thermal contrasts.

The wind rose (Figure 10a) shows a clear prevalence of southerly winds, particularly from the SSE–S sectors, while westerly and south-westerly directions represent a secondary component. Most observations fall within the range of 2–5 m s⁻¹, suggesting generally light-to-moderate wind conditions, while stronger winds occur only occasionally.

Figure 10b shows a clear diurnal wind cycle. During the night and early morning, winds blow predominantly from the SSE at relatively stable speeds of about 3 m s⁻¹, likely reflecting channeling along the Crati Valley. After sunrise, wind speed decreases to approximately 1.5–2 m s⁻¹ and weak northerly winds occur around 09:00–10:00, possibly marking the transition between nocturnal and daytime valley circulations. From late morning to the afternoon, winds rotate toward WSW–SW and strengthen, reaching their maximum intensity in mid-afternoon. This pattern may be associated with the development of a daytime up-valley flow and local channeling through the Busento Valley. During the evening, the circulation gradually returns to the prevailing SSE regime. These conditions may promote greater turbulent mixing and improved urban ventilation in the afternoon, potentially helping to reduce temperature contrasts between urban and rural areas. In contrast, the morning circulation (which is weaker and less persistent) may be associated with reduced ventilation. These possible effects are examined more directly in the subsequent SUHII–wind analysis.

3.6.2. SUHII Under Different Wind-Speed Conditions and TX90p Categories

Following the workflow proposed by Esposito et al. [27], to assess the possible modulation of SUHII by wind intensity, wind observations from station ID 1017 were classified into two empirical wind-intensity categories: WI1, corresponding to wind speeds ≤3 m s⁻¹, and WI2, corresponding to wind speeds >3 m s⁻¹. Pixel-level SUHII distributions were then analyzed separately for each land-use class, wind-intensity category, TX90p condition, and diurnal period. To simplify the comparison, the analysis was performed for daytime (06:00–20:00) and nighttime (20:00–06:00). Of the 127 ECOSTRESS scenes included in the analysis, 80 were classified as WI1 and 47 as WI2. During daytime, 65 scenes occurred under WI1 conditions and 38 under WI2 conditions, whereas the nighttime sample comprised 15 and 9 scenes, respectively. At the aggregate level, mean daytime SUHII was 3.72 °C under WI1 and 3.49 °C under WI2, while nighttime values were 1.57 °C and 1.66 °C, respectively. These relatively small and non-systematic differences indicate that stronger wind conditions were not consistently associated with lower SUHII across the analyzed observations.

Figure 11 shows that SUHII remains generally positive under both WI1 and WI2 conditions, indicating that urban pixels were typically warmer than the scene-specific rural reference. Daytime SUHII distributions are generally higher than nighttime distributions, reflecting the dominant role of solar forcing in enhancing urban–rural surface thermal contrasts during the day.

Differences between WI1 and WI2 are generally modest and vary across land-use classes, TX90p conditions, and diurnal periods. In some cases, WI2 is associated with lower pixel-level SUHII values, suggesting a possible ventilation-related modulation of surface thermal contrasts. However, this pattern is not consistent across all panels. In several cases, WI1 and WI2 exhibit similar distributions, and their relative behavior depends on the specific combination of land-use class, TX90p condition, and time of day.

During the daytime, differences between WI1 and WI2 are generally limited, suggesting that strong solar heating may partly mask or offset any potential ventilation-related effect. At night, wind-related differences become more evident in some panels, although the response remains variable among land-use classes and TX90p conditions. Therefore, these results should be interpreted as descriptive evidence of a possible wind-related modulation rather than as a robust demonstration of a systematic reduction in SUHII under stronger wind conditions.

The interpretation of the wind effect is further constrained by the use of wind observations from a single urban station. Given the complex topography of the Cosenza valley, local ventilation may vary substantially across the study domain because of topographic channeling, valley-axis flows, local surface roughness, and urban morphology. Moreover, the two wind-intensity classes may not be evenly distributed throughout the day, potentially confounding the effects of wind intensity with the diurnal variability in SUHII.

Overall, the pixel-level distributions suggest that wind intensity may contribute to modulating SUHII under specific conditions, although this effect appears secondary and is neither spatially nor temporally uniform. Urban land-use class, surface characteristics, and diurnal forcing remain the dominant factors controlling the observed SUHII variability.

4. Discussion

The results describe how surface overheating in Cosenza varies across land-use classes characterized by different morphological and surface-cover properties and along neighborhood-scale morphological gradients, and how these pixel-level SUHII distributions change under different thermal (TX90p/No-TX90p) and wind-intensity conditions. The use of ECOSTRESS LST observations allowed clear intra-urban differences in surface thermal response to be identified, while ground-based meteorological measurements provided information on TX90p occurrence and local wind conditions.

Continuous urban fabric (CLC1) generally showed the highest pixel-level SUHII distributions, especially during the central daytime hours, when median values reached approximately 6.5 °C. This behavior is consistent with the morphological characteristics of this class, which exhibits higher building surface fraction, higher impervious surface fraction and lower availability of permeable surfaces. Discontinuous urban fabric (CLC2) showed lower SUHII values, likely reflecting its more heterogeneous structure and higher proportion of permeable surfaces. Industrial and commercial areas (CLC3) displayed intermediate-to-high SUHII values, confirming that these zones can contribute substantially to urban surface overheating depending on the specific combination of roofs, paved surfaces, open exposed areas, vegetation and local morphology.

The complementary bin-based analysis further showed that the class-level differences were accompanied by distinct within-class morphological gradients. In CLC2 and CLC3, increasing HM, BSF and ISF were generally associated with progressively higher SUHII across the analyzed diurnal periods, indicating that thermal variability within the same land-use class is also related to continuous variations in urban form and surface imperviousness. In contrast, CLC1 showed weaker and less systematic relationships, particularly for BSF, which may partly reflect the narrower morphological range of this already compact and highly urbanized class. These results strengthen the evidence that neighborhood-scale morphology is associated with SUHII variability beyond the categorical CLC comparison. However, the observed gradients should be interpreted as descriptive associations rather than as independent effects of individual morphological parameters, because HM, BSF and ISF are not mutually independent and are also related to land-use configuration.

The positive daytime SUHII observed in Cosenza is partly consistent with the findings reported by Albini et al. [28] for Catanzaro, another Calabrian city characterized by complex terrain. In that study, SUHII varied according to topographic setting, with positive values in hilly sectors and inverse SUHII patterns in lowland areas. These findings suggest that the sign and magnitude of SUHII in Mediterranean cities depend not only on urban density and imperviousness, but also on elevation, rural reference conditions, local morphology and broader geographic setting. Compared with Mediterranean coastal cities such as Cagliari and Bari, where weak or inverse SUHII patterns have been reported, Cosenza shows a clearer positive daytime SUHII, probably linked to its inland valley setting and the thermal behavior of surrounding rural surfaces.

A relevant finding is that TX90p conditions were associated with lower SUHII values than No-TX90p conditions. This does not imply reduced thermal stress during TX90p conditions. Rather, it suggests that the urban–rural surface temperature gradient decreases, possibly because rural reference areas also warm strongly under extreme synoptic heat conditions. During TX90p conditions, rural and peri-rural areas are likely to experience intense solar heating, reduced soil moisture, limited evapotranspiration and vegetation drying, particularly under Mediterranean summer conditions [1,2,4,18]. These processes can increase rural LST and narrow the thermal contrast with urban surfaces. As reported by Esposito et al. [20], absolute surface temperatures may remain very high even when SUHII values decrease. This finding indicates that changes in SUHII do not necessarily reflect changes in absolute land surface temperatures during TX90p conditions.

Wind intensity exhibited a more complex relationship with SUHII. The comparison of WI1 and WI2 indicates that wind speed may modulate pixel-level SUHII distributions under some conditions, but the pattern was not uniform across land-use classes, TX90p/No-TX90p categories and diurnal periods. Therefore, the results do not support a simple or systematic reduction in SUHII under stronger wind conditions. Rather, they indicate that ventilation may act as a secondary modulating factor whose influence likely depends on the timing of the ECOSTRESS acquisition, the TX90p classification, and the local land-use context.

This interpretation is broadly consistent with previous studies showing that weak winds can favor urban heat accumulation, whereas stronger atmospheric circulation can enhance turbulent exchange and heat transport. Esposito et al. [27], for example, reported an approximately 1.5 °C reduction in nighttime SUHII under stronger wind conditions across Italian cities, together with a contraction of the areas characterized by the highest SUHII values. Similarly, Donateo et al. [29] found that UHI intensity in Lecce decreased when wind speed exceeded 3 m s⁻¹, particularly under northerly winds that enhanced urban ventilation. In the present study, differences between WI1 and WI2 were generally modest during daytime and became more evident in some nighttime comparisons, particularly for the more urbanized classes. Nevertheless, these differences were not sufficiently uniform to demonstrate a generalized wind-induced mitigation of SUHII.

The influence of wind should also be interpreted in relation to the complex topographic setting of Cosenza. Valley geometry, topographic channeling, slope flows and spatial variations in urban roughness may produce substantial local differences in wind speed and direction, which cannot be fully represented by observations from a single meteorological station. Moreover, the use of an empirical threshold of 3 m s⁻¹ simplifies a continuously varying atmospheric process. Overall, wind appears to be a secondary but non-negligible factor associated with SUHII variability, while the persistence of positive urban–rural thermal contrasts under both wind regimes indicates that land-cover characteristics and urban morphology show more consistent relationships with the observed SUHII patterns.

Some limitations should be acknowledged. First, although ECOSTRESS provides high spatial resolution and variable acquisition times, the number of valid scenes is not uniform across years, time intervals, TX90p categories, and wind conditions. Therefore, interannual and diurnal comparisons should be interpreted considering differences in sampling frequency and overpass timing. In particular, the interpretation of lower SUHII during TX90p conditions should be considered cautiously. Although this pattern may be related to stronger warming of rural reference surfaces during extreme heat, the present analysis focuses mainly on urban–rural contrasts and does not fully separate the absolute LST response of urban and rural surfaces. Differences in overpass time, month and year between TX90p and No-TX90p observations may therefore partly influence the observed contrast.

Second, the LST and SUHII boxplots are based on pooled pixel-level distributions. These distributions are useful for describing spatial variability within land-use classes, but they should not be interpreted as formal statistical comparisons based on independent observations, because pixels belonging to the same ECOSTRESS scene may be spatially autocorrelated and share the same meteorological forcing. Future analyses should use scene-level class medians or hierarchical models with scene/date as a grouping factor.

Third, the wind analysis is based on one urban anemometric station and on an empirical wind-speed threshold. Although wind direction was described through the wind rose and mean diurnal wind-cycle analysis, it was not explicitly used to stratify SUHII distributions. Therefore, the role of wind direction, valley-axis channeling and spatial variability in ventilation remains only partially assessed. Future work should combine scene-level SUHII estimates with directional wind-sector stratification and spatially distributed meteorological observations to better isolate the role of ventilation.

Despite these limitations, the integrated use of satellite-derived LST, land-use information, morphological indicators, TX90p classification, and wind observations provides a useful framework for assessing urban overheating in Mediterranean valley environments. The identification of within-class morphological gradients further shows that SUHII variability cannot be described solely through categorical land-use classes, but also reflects continuous differences in neighborhood-scale urban form and surface cover.

The results suggest that climate-sensitive planning in Cosenza and similar cities should consider not only compact urban fabrics and impervious surfaces, but also the behavior of rural surfaces under TX90p conditions, local ventilation conditions, and the role of topography in shaping urban–rural thermal contrasts.

5. Conclusions

This study assessed SUHII in Cosenza, a Mediterranean city characterized by complex terrain in southern Italy, by integrating ECOSTRESS-derived LST, land-use classification, neighborhood-scale morphological indicators, TX90p classification, and wind observations. The analysis provides an integrated framework for investigating the associations of urban form, surface cover, extreme-temperature conditions, and local ventilation with urban–rural surface thermal contrasts in a complex topographic setting.

The results show that surface overheating differs clearly among land-use classes. Continuous urban fabric (CLC1) exhibited the highest SUHII values, with median intensities reaching approximately 6.5 °C during midday, while discontinuous urban fabric (CLC2) showed lower values and industrial or commercial areas (CLC3) displayed intermediate-to-high thermal contrasts. These differences highlight the importance of considering land-use and morphological heterogeneity when describing surface heat patterns.

The bin-based analysis further showed that the class-level differences were accompanied by distinct within-class morphological gradients. In CLC2 and CLC3, increasing building height, building surface fraction, and impervious surface fraction were generally associated with progressively higher SUHII across the diurnal periods considered. In contrast, CLC1 showed weaker and less systematic gradients, particularly for building surface fraction, likely reflecting the more restricted morphological variability in this already compact urban class. These relationships should be interpreted as descriptive associations rather than as independent or causal effects of individual morphological parameters.

TX90p conditions were associated with lower SUHII values than No-TX90p conditions. This indicates that, during TX90p conditions, rural reference areas may also experience substantial warming, reducing the urban–rural surface thermal gradient. However, this does not necessarily imply lower surface temperatures, since the present analysis is based on urban–rural thermal contrasts rather than on direct comparisons of absolute land surface temperatures. Thus, the lower SUHII observed under TX90p conditions primarily reflects changes in the relative thermal contrast between urban and rural surfaces rather than a reduction in urban surface temperature itself.

Wind-related differences in SUHII were observed, but they were modest and not uniform across land-use classes, TX90p/No-TX90p categories, and diurnal periods. The available observations therefore do not support a systematic reduction in SUHII under stronger wind-speed conditions, although local ventilation may contribute to modulating surface thermal contrasts under specific conditions. The present pixel-level analysis and the use of wind observations from a single urban station do not allow a robust assessment of its spatially and temporally variable influence on SUHII.

Overall, the findings suggest that urban overheating in Mediterranean cities with complex terrain is associated with spatial differences in land-use and neighborhood-scale morphology, together with the temporal variability in extreme-temperature and local ventilation conditions. Climate-sensitive planning strategies may therefore benefit from measures aimed at limiting impervious surface cover, increasing permeable and vegetated areas, and preserving potential ventilation pathways, while accounting for the strong spatial heterogeneity of urban form. Future research should integrate directional wind analysis, additional meteorological observations, vegetation and soil-moisture indicators, scene-level or spatially explicit statistical approaches, and urban climate modeling to improve the understanding of urban overheating processes and support more effective heat-mitigation strategies in Mediterranean cities with complex terrain.