Spatio-temporal analysis of land surface temperature and land cover changes in Prešov city using downscaling approach and machine learning algorithms

Spatio-temporal analysis of land surface temperature and land cover changes in Prešov city using downscaling approach and machine learning algorithms | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Spatio-temporal analysis of land surface temperature and land cover changes in Prešov city using downscaling approach and machine learning algorithms Anton Uhrin, Katarína Onačillová This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5143836/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 03 Jan, 2025 Read the published version in Environmental Monitoring and Assessment → Version 1 posted 4 You are reading this latest preprint version Abstract In recent decades, global climate change and rapid urbanisation have aggravated the urban heat island (UHI) effect, affecting the well-being of urban citizens. Although this significant phenomenon is more pronounced in larger metropolitan areas due to extensive impervious surfaces, small and medium-sized cities also experience UHI effects, yet research on UHI in these cities is rare, emphasizing the importance of Land Surface Temperature (LST) as a key parameter for studying UHI dynamics. Therefore, this paper focusses on the evaluation of LST and land cover (LC) changes in the city of Prešov, Slovakia, a typical medium-sized European city that has recently undergone significant LC changes. In this study, we use the relationship between Landsat-8/-9 derived LST and spectral indices NDBI, NDVI, NDWI derived from Landsat-8/-9 and Sentinel-2 to downscale LST to 10 m. Two machine learning (ML) algorithms, Support Vector Machine (SVM) and Random Forest (RF), are used to assess image classification and identify how different types and LC changes in selected years 2017, 2019, 2023 affect the pattern of LST. The results show that several decisions made during the last decade, such as the construction of new urban fabrics and roads, caused the increase of LST. The LC change evaluation, based on the RF classification algorithm, achieved overall accuracies of 93.2% in 2017, 89.6% in 2019, and 91.5% in 2023, outperforming SVM by 0.8% in 2017 and 4.3% in 2023. This approach identifies UHI-prone areas with higher spatial resolution, helping urban planning mitigate the negative effects of increasing urban LSTs. land surface temperature land cover downscaling urban heat island machine learning classifiers Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Introduction Urbanization and population growth result in an uneven distribution of surfaces that absorb substantial solar radiation (Purio et al., 2022 ), thus altering land cover (LC) dynamics. These changes affect critical processes such as heat transfer (Shamsaei et al., 2022), air circulation (Longley et al., 2004 ), precipitation patterns (Patil & Surawar, 2023 ; Steensen et al., 2022 ), and urban thermal environments (Fadhil et al., 2023 ). This leads to a transformation of the microclimate, resulting in significant temperature differences between urban and rural areas, known as the Urban Heat Island (UHI) effect (Grigoraș & Urițescu, 2019 ). Studies have shown that even in small and medium-sized cities, factors such as urbanization, reduced vegetation, and increased human activity contribute to elevated temperatures compared to surrounding rural areas. Despite the significant impact of UHI on local climates, energy consumption, and public health in these smaller cities, research in this area remains limited, although some studies have addressed this gap (Miles et al., 2023 ; Ivajnšič & Žiberna, 2018 ; Vardoulakis et al., 2013 ; Dobrovolný, 2013 ). Studying the UHI effect relies heavily on analyzing Land Surface Temperature (LST) data, which provides critical insights into spatial and temporal variations in surface temperatures. This understanding is essential to assess urban heat fluxes, identifying the presence and intensity of the UHI phenomenon (Tanoori et al., 2024 ; Han et al., 2023 ), simulate surface energy exchange (Khan et al., 2023 ) and study the energy balance of the surface at global, regional and local scales (Yan et al., 2024 ). LST is also widely used to assess surface moisture and drought levels (Xiong et al., 2022 ; Weiland et al., 2023 ). A robust and efficient method for acquiring LST data involves the use of freely available thermal satellite imagery, facilitating a comprehensive analysis of temperature patterns across broad geographic regions. In urban environments, the spatial and temporal heterogeneity of surface temperature is pronounced due to complex surface structures and materials, reflecting the diversity of urban LC. However, freely available data from thermal sensors typically have coarser spatial resolution and higher temporal resolution, which is inadequate for accurately identifying thermal characteristics of urban areas and conducting detailed analyses of factors influencing the UHI formation (Song and Park, 2020 ). The resolution of acquired thermal data varies, ranging from 1000 m for AVHRR to 100 m for Landsat-8/-9, making it more suitable for global to regional scale studies (Cheval et al., 2011 ). Advancements in satellite technology, improvements in multispectral and thermal sensors, rapid development of GIS and remote sensing methods, along with the availability of open-source platforms such as Google Earth Engine (GEE), are crucial for overcoming these limitations and enabling higher-resolution LST mapping. One of the most widely used approaches to achieve higher spatial resolution in satellite imagery for deriving LST are downscaling methods, transforming a coarser spatial resolution image into a higher spatial resolution image (Hu et al., 2024 ). There are several downscaling approaches or models that play a role in facilitating LST analyses and reducing computational time and storage space (Xu et al., 2021 ; Hutengs & Vohland, 2016 ). This work offers one of the first investigations into a complex spatio-temporal analysis of recent LC and LST changes in the city of Prešov, that has been not subject of UHI studies before (to date) on the local scale for the years 2017, 2019 and 2023 with the use of an open-source cloud platform GEE. We derive LST within the urban environment of the Prešov city at higher spatial resolution using downscaling of Landsat thermal satellite data with multispectral indices derived from Sentinel 2 data. We also use semi-automatic classification of satellite data (Sentinel-2) using selected machine learning algorithms - Support Vector Machine and Random Forest. Information about LST of different urban surfaces and land use types can help to suggest effective strategies in the areas of efficient urbanisation, urban planning, or sustainable development proposals to minimise the negative impacts of climate change and UHIs to optimise or enhance the quality of life of urban residents. Study area The area of interest is the medium-sized city of Prešov with the adjacent municipalities of Haniska, Kendice, Ľubotice, Petrovany and Záborské, located in eastern Slovakia, and characterised by a typical Central European urban environment with significant heterogeneity of LC classes (Fig. 1 ). The total area of the city is approximately 70.44 km 2 with a total population of 84,824 inhabitants, which represents approximately 1204.20 inhabitants per 1 km2 (Statistical Office SR, 2022 ). Agricultural land occupies 35%, and forests account for 31%. Artificial surfaces are also highly represented, mainly by built-up areas that occupy 28% − continuous urban fabric accounts for 4%, discontinuous urban fabric accounts for 8%, and more than 7% are occupied by industrial, commercial, public, military, and private transport units. The road and rail network and associated land account for less than 4%. Sports and leisure facilities comprise 2% and green urban areas account for 1.5% (Copernicus Land Monitoring Service, 2018 ). According to Köppen's climate classification, the territory of Prešov lies in the mild continental climate zone, classification subtype "Dfb", which is characterised by significant seasonality, with warm and humid summers, severe winters and short dry periods (Kotek et al., 2006). The largest watercourse that flows through the territory of the city is the Torysa River and its left-hand tributaries Sekčov and Delňa. Materials and methods The research in this study uses multispectral imagery collected by two satellite missions - the Landsat-8/-9 mission by NASA and Sentinel-2 mission by ESA. Six cloud-free multispectral satellite products obtained over the Prešov area were employed for selected days in 2017, 2019 and 2023: Landsat-8 OLI/TIRS sensor (3 products) and MSI sensor onboard the Sentinel-2 satellite (3 products). These six satellite Landsat-8/9 and Sentinel-2 data products were selected based on the proximity of their acquisition dates within the same year, considering similar atmospheric conditions and the lowest possible percentage of cloud cover. The corresponding air temperature parameters extracted from the OGIMET service developed by the Spanish Meteorological Institute ( www.ogimet.com ) are stated in Table 1 . Table 1 Selected parameters of the satellite products and the corresponding air temperature based on meteorological data from SHMÚ (2024). Satellite, sensor and product Acquisition date Acquisition time [UTC] Cloud cover [%] Air temperature [°C] 09:00 10:00 Landsat-8 OLI/TIRS L2 C2 11.08.2017 09:26:41 0.42 29.7 30.8 Landsat-8 OLI/TIRS L2 C2 30.06.2019 09:26:34 0.02 25.4 27.3 Landsat-9 OLI/TIRS L2 C2 19.07.2023 09:26:16 12.61 25.8 26.9 Sentinel-2A MSI L2A 03.08.2017 09:40:46 6.04 29.6 30.7 Sentinel-2A MSI L2A 01.07.2019 09:30:39 6.59 27.6 28.9 Sentinel-2B MSI L2A 18.07.2023 09:35:49 11.55 20.9 22.3 Three USGS Landsat-8/-9 Level-2 Collection-2 Surface Reflectance products were used to derive three selected spectral indices that are closely related to LST at spatial resolution of 30 m and to calculate atmospherically corrected LST itself using a modified radiative transfer equation inversion methodology: "LC08_L1TP_187026_20170811_20200903_02_T1" (11/08/2017, 09:26:41 UTC) for 2017, "LC08_L1TP_187026_20190630_20200827_02_T1" (30/06/2019, 09:26:34 UTC) for 2019, "LC09_L1TP_187026_20230719_20230719_02_T1" (2023/07/19, 09:26:16 UTC) for 2023. Three Harmonized Sentinel-2 MSI: MultiSpectral Instrument, Level-2A products were used to obtain the same spectral indices as in the case of the Landsat-8 satellite, but in a higher, 10 m spatial resolution, which has been used in the downscaling process. The products were selected based on the proximity of the acquisition date to the selected products of the Landsat-8/-9 satellite: "S2A_MSIL2A_20170803T094031_N0205_R036_T34UEV_20170803T094046" (08/03/2017, 09:40:46 UTC) for 2017, "S2A_MSIL2A_20190701T093041_N0212_R136_T34UEV_20190701T115615" (07/01/2019, 09:30:39 UTC) for 2019, "S2B_MSIL2A_20230718T093549_N0509_R036_T34UEV_20230718T112309" (2023-07-18, 09:35:49 UTC) for 2023. Level-2A products provide users with orthorectified Bottom-Of-Atmosphere surface reflectance, the classification of scene, including cloud cover and cloud shadows, aerosol optical thickness, and water vapour maps (Gascon et al., 2017 ). The workflow is fully coded in JavaScript using the Code Editor Platform of the online cloud application GEE. The multispectral imagery was processed using the pipeline shown in Fig. 2 and the script itself is divided into two main parts: processes associated with LST downscaling approach, described in detail in Subsection 3.1, and processes of semi-automatic image classification using machine learning algorithms, which are described in Subsection 3.2. Validation data To validate the LST downscaling results, field measurements of surface kinetic temperature obtained by temperature probe Pt1000TG7/E with a Comet data logger were used at five selected open-sky locations within the study area to ensure their reliable identification even in satellite images and to represent various types of surfaces, as shown in Fig. 3 . Two natural materials – stone pavement and grass – and three artificial surfaces – tartan, artificial grass, and concrete – were selected. The device allows measurement of the surface temperature in 5-s measurement intervals in the range of -30 to 200°C with a measurement accuracy of 0.15°C (TR Instruments, 2023). Validation measurements were carried out in the area of interest on 19 July 2023 at 7:00 a.m. and 10:00 a.m. to match the acquisition time of the Landsat-9 satellite on the same day, whose thermal data were used to derive LST (acquisition time 09:26:16 UTC). LST downscaling methodology Landsat-8/-9 TIRS allows thermal radiation to be mapped at a native 100 m spatial resolution, which is resampled by United States Geological Survey (USGS) to 30 m for consistency with other spectral bands of these satellites. Thermal radiation can be later converted to LST; however, for detailed analysis of LSTs in heterogeneous urban areas, the coarse spatial resolution of 30 m and longer intervals between acquisitions of these satellites are not sufficient to capture interactions of LST in terms of the heterogeneous character of LC. On the contrary, Sentinel-2 does not have a thermal band, so it is not possible to directly derive LST. However, due to its high spatial resolution (native 10 m resolution in visible and near-infrared (NIR) bands) and temporal resolution (recording every 6 days), this satellite has high potential for calculating various spectral indices, allowing to accurately capture the complexity of urban LC in better detail compared to the Landsat-8/-9 satellites. In this study, the spatial downscaling approach involves establishing a multiple linear model utilizing selected three spectral indices and LST obtained from the thermal band of the Landsat-8/-9 satellite according to the model adopted from Bonafoni et al. ( 2016 ) and Onačillová et al. ( 2022 ). The proposed model is capable to predict LSTs in 10 m spatial resolution by using spectral indices obtained from Sentinel-2 satellite data as input. Calculation of LST predictors - spectral indices The first step involved preparation of LST predictors: three spectral indices ̶ NDVI (Purevdorj et al. 1989; Guo et al., 2021 ), NDBI (Zha et al., 2003 ; Goldblatt et al., 2018 ) and NDWI (McFeeters, 1996 ; Camps et al., 2020 ) ̶ were chosen and derived from Landsat-8/-9 satellite images with a 30 m spatial resolution and Sentinel-2 satellite images with a 10 m spatial resolution. The formulas for calculating the selected spectral indices of the Landsat-8/9 and Sentinel-2 satellites are stated in Table 2 . Table 2 Spectral indices used for LST downscaling. Acronym Description Formulation Modified Formulation for Landsat-8/-9 Bands Modified Formulation for Sentinel-2 Bands NDVI Normalized difference vegetation index (NIR − RED/ (NIR + RED) (B5 − B4)/(B5 + B4) (B8 − B4)/(B8 + B4) NDWI Normalized difference water index (Green − NIR)/ (Green + NIR) (B3 − B5)/(B3 + B5) (B3 − B8)/(B3 + B8) NDBI Normalized difference built-up index (SWIR1 − NIR)/ (SWIR1 + NIR) (B7 − B5)/(B7 + B5) (B12 − B8)/(B12 + B8) To derive NDVI from the Landsat-8/9 satellite, the NIR band with wavelengths of 0.85–0.88 µm and the red (R) band at wavelengths of 0.64–0.67 µm were used. The input spectral bands for the NDWI derivation were the green band with wavelengths of 0.533–0.590 µm and the NIR band. For the NDBI derivation, these were the NIR and shortwave infrared 1 (SWIR1) bands with a wavelength of 1.57–1.65 µm. All these bands had a native 30 m spatial resolution. Equivalent spectral indices were derived from Sentinel-2 satellite data: the NIR band with a central wavelength of 0.84 µm and the red band with a central wavelength of 0.66 µm were used to calculate the NDVI. The green band with a central wavelength of 0.56 µm and the NIR band were used to obtain the NDWI. NDBI was calculated based on the NIR and SWIR1 bands with a central wavelength of 1.61 µm (resampled from native spatial resolution of 20 to 10 m to ensure consistency with the resolution of NDVI and NDWI) (Fig. 4 ). Multiple linear regression model between LC indices and LST In this study, the spatial downscaling approach relies on establishing a linear relation utilizing three spectral indices derived from Landsat-8/9 imagery (NDVI, NDWI, NDBI) as predictors, with LST serving as the predicted variable. Then the established model is applied to the study areas with finer resolution. The LST with spatial resolution of 30 m was derived from the surface temperature (ST_B10) of the Landsat-8/9 TIRS sensor. ST_B10 is derived from Landsat-8 Collection 2 Level 1 Thermal Infrared Sensor (TIRS) Band 10 data (USGS, 2024 ). Then, the multiple linear regression statistical model between LST and spectral indices was constructed as predictors at coarser (30 m) resolution. The model for the calculation of LST at the coarser level from Landsat-8/-9 imagery using the output regression coefficients to calculate the LST in 30 m spatial resolution can be expressed as follows: $$\:{LST}_{c}={a}_{0}+\:{a}_{1}\times\:{NDVI}_{c}+\:{a}_{2}\times\:{NDBI}_{c}+\:{a}_{3}\times\:{NDWI}_{c}\:$$ 1 where the subscript c means the variable with coarser 30 m spatial resolution derived from Landsat-8/-9 satellite data and the variables a 0 − a 3 represent the output regression coefficients. Then, these derived regression coefficients are applied to obtain the downscaled LST using selected Sentinel-2 NDVI, NDBI, and NDWI spectral indices with finer 10 m spatial resolution. The fundamental premise of this method is that the thermal bands have a direct relation with the spectral bands, and the relationship is scale invariable (Gao et al., 2012). The subsequent model was employed to downscale LST at finer resolution: $$\:{LST}_{f}={a}_{0}+\:{a}_{1}\times\:{NDVI}_{f}+\:{a}_{2}\times\:{NDBI}_{f}+\:{a}_{3}\times\:{NDWI}_{f}\:$$ 2 where the coefficients a 0 − a 3 of Eq. ( 1 ) are applied and the subscript f means the variable (spectral index) with a finer 10 m spatial resolution derived from Sentinel-2 satellite imagery. Image classification using ML algorithms SVM and RF ML algorithms were used and compared for the image classification of multispectral imagery using the "Harmonized Sentinel-2 MSI level 2A" data product. Firstly, three cloud-free median compositions were generated for the respective years 2017, 2019, and 2023. This process involved utilizing Sentinel-2 satellite scenes that met specific criteria: captured between June 1st and August 8th of each selected year with cloud cover below 20%. Subsequently, these composits were displayed as true color RGB compositions used to create five datasets of training samples, representing the five dominant classes/types of classified LC – water bodies, forests, grasslands, agricultural lands, and built-up areas. which were systematically distributed across our study area. These training samples were generated separately for the years 2017, 2019, and 2023 to accurately capture the respective types of LC for each year. Subsequently, these training samples were used for image classification using the SVM and RF algorithms. In GEE, the SVM algorithm implemented under the "ee.Classifier.libsvm()" tool was applied. An overview of the parameters and their respective values used to train the SVM model is presented in Table 3 . RF is implemented using the tool "ee.Classifier.smileRandomForest()" and uses 6 input parameters (Table 3 ). For both algorithms, to achieve higher classification accuracy, we also used the NDVI and NDBI spectral indices as input, similarly to the study of Svoboda et al. ( 2022 ). These indices were used to train the models to better distinguish between vegetation and built-up areas. To evaluate classification accuracy and determine a more reliable classifier, the accuracy assessment was performed separately for the 2017, 2019 and 2023 classification results and for the two classification algorithms used (SVM and RF). Training and testing data sets were established at an 80:20 ratio. The training data sets were used for model training, whereas the testing datasets were employed for validating the accuracy of the classifications. Table 3 Parameters used for image classification by SVM and RF algorithms in the GEE environment. Source: Google Earth Engine ( 2024 ), author's processing. Algorithm Parameter Value SVM Decision Procedure Default: Voting SVM type Default: C_SVC (for non-linear data) Kernel type Linear Cost 10 RF Number of decision trees 1500 Number of variables per split Default: the square root of the number of variables Minimum leaf population Default: 1 Randomization seed Default: 0 Fraction of input to bag per tree Default: 0.5 Maximum number of leaf nodes Default: no limit Results and discussion Downscaled LST using the relationship between spectral indices and LST The LST prediction model was established using ordinary least squares linear regression and three predictors − NDVI, NDWI, and NDBI. Scatter plots between the pairs of variables, Landsat-8 LST, and each predictor revealed that there is a significant correlation between the spectral indices and LST in Prešov city: a negative correlation with NDVI and positive correlations with NDBI and NDWI. The results show that there is a direct correlation between LST and NDBI (r 2017 = 0.99, r 2019 = 1.00, r 2023 = 0.50) which means that highly built-up areas have high recorded temperature values. This observation agrees with the multivariate analysis. An indirect relationship is observed between LST and NDVI (r 2017 = − 0.73, r 2019 = − 0.61, r 2023 = − 0.85) and LST and NDWI (r 2017 = − 0.72, r 2019 = − 0.67, r 2023 = − 0.60). Based on these results, it can be inferred that the correlation values suggest that NDBI is a good indicator for LST, which is aligned with the findings of Purio et al., ( 2022 ). Figure 5 shows a comparison of downscaling results for one of those years (2023) - LST c in the original 30 m spatial resolution (before downscaling) and LST f after downscaling in 10 m spatial resolution with clearly improved (sharpened) LST pattern. The coefficient of determination (R 2 ) is 0.92. Differences in the distribution of LST can be clearly linked to specific LC categories. From the LST patterns in Fig. 5 it can be seen that the presence of the UHI phenomenon in the city is significant. The lowest LST values in the study area are spread across a wide forest park in the western part of the city of Prešov, with a significant percentage of vegetation cover and low human intervention in nature. Low temperature values are also found in the southern part of the area with agricultural fields and along the Torysa River valley. In contrast, the highest values, that reach sometimes reaching up to 50°C are visible in the historical city core, areas of other urban fabrics on the outskirts of the city core and recently built road communications and industrial zones in the south of the city core. Differences in LST patterns between selected images in 2017, 2019, and 2023 were predominantly observed in suburban areas affected by the construction of a highway bypass and the development of new buildings in the industrial parks of Haniska and Petrovany. These selected locations were further analysed using LST f . The location map is depicted in Fig. 6 , with numbers 1–4 marking 4 locations where significant changes in LST were detected within the area of interest. The corresponding changes in LSTs for these four locations are presented in Table 4 . Based on the maps in Fig. 6 and the data presented in the Table 4 , we observe that in areas where LC has changed due to the expansion of built-up areas and roads, there is a corresponding increase in LST values. Localities 1 and 2 indicate sites located on newly constructed sections of the highway and expressway, where LST increased approximately 6–7°C in 2023 (after construction) compared to 2017 (before construction). An even greater difference in LSTs can be observed at sites 3 and 4, where new buildings have been constructed in the industrial areas of Haniska and Petrovany. At site 3, the LST value was 38°C in 2017 (before the construction of the production building), and by 2023, this value had increased by approximately 10°C to 48°C (after the construction of the production building). Similarly, at location 4, the LST increased by approximately 17°C after the construction of the logistics centre. The increase in LST values in these cases is attributed both to the alteration in LC and the nature of the surfaces themselves. Previously, these areas had relatively natural landscapes or were used for agriculture. By 2023, these areas had undergone development characterized by impermeable artificial materials, including asphalt roads (sites 1–2) and buildings predominantly featuring metal roofs (sites 3–4). Table 4 LST values for selected 4 locations in 2017, 2019 and 2023. Locality Change in LC LST (°C) 2017 2019 2023 1 road communication - expressway 36.7 40.3 42.5 2 road communication - highway 35.3 38.8 42.9 3 industrial zone - production building 38.1 47.7 48.1 4 industrial zone - logistics center 35.0 36.9 51.9 Image classification using ML algorithms – RF and SVM The second part of our research was dedicated to semi-automatic supervised classification of satellite imagery using two machine learning algorithms SVM and RF. Classification outputs using the SVM and RF algorithms are shown in Fig. 7 . The overall accuracies of both algorithms for respective years 2017, 2019 and 2023 are displayed in Table 5 . Table 5 Overall accuracy of image classifications using SVM and RF algorithms for 2017, 2019 and 2023. Year Overall accuracy (in %) SVM RF 2017 92.4 93.2 2019 92.2 89.6 2023 87.2 91.5 Based on the visual assessment in Fig. 7 , semiautomatic classification using SVM and RF machine learning algorithms demonstrated considerable success, although some instances of pixel misclassification were observed (e.g., built-up areas classified as water bodies, or agricultural land as built-up areas). According to Table 5 , the overall accuracy of the six classifications evaluated using the SVM and RF algorithms ranged from 87.2–93.2%. Satellite imagery from 2017 was classified with an accuracy of 92.4% by the SVM model and 93.2% accuracy by the RF model. In 2019, the SVM model had a reliability of 92.2%, slightly less than 3% more than the RF model. By 2023, the SVM model's success rate had dropped to 87.2%, while the RF model's success rate was approximately 91.5%. Several factors impacted the accuracy of image classification. Primarily, the spatial resolution of Sentinel-2 images played a significant role. The class of water bodies was frequently misclassified due to the difficulty in identifying the Torysa River and its tributaries in parts of the images. This challenge arose because their width is less than 10 metres, which is the native resolution of the Sentinel-2 bands. The second factor concerns the radiometric characteristics of satellite scenes, particularly when they are median composites generated from satellite images over several months. This can result in minor inaccuracies in classification, such as with agricultural soils, because the varying stages of vegetation of individual crops can make them harder to distinguish from grasslands. Based on the results achieved, we evaluated that the RF algorithm appeared to be more reliable and therefore, in the further analysis, we exclusively used the classification results of this particular more reliable classifier (RF). Higher reliability of RF classification was also confirmed, for example, by studies by Chowdhury ( 2024 ), Zafar et al. ( 2024 ). Spatio-temporal evaluation of changes in LC and LST Spatio-temporal analysis was conducted using derived downscaled LST outputs in 10 m spatial resolution and the results of supervised classification by the RF classifier for the years 2017, 2019 and 2023. Figure 8 shows four localities within the area of interest, represented by image classification by the RF algorithm and the corresponding LST, highlighting the most significant changes in LC. The LST values for these locations are presented in Table 4 . Locations 1 and 2 depict areas where the highway bypass and tunnel were constructed between 2017 and 2019, leading to an increase in LST. In 2017, location 1 consisted of grassed areas and part of a garden area, with an LST value of approximately 37°C. By 2019, during road construction, the LST increased to 40°C and by 2023, it had risen to nearly 43°C. A similar trend is observed at location 2, where the LST value increased from 35°C in 2017 to nearly 43°C in 2023. Locations 3 and 4, which were previously agricultural land, have been transformed into built-up areas. At location 3, LSTs were approximately 38°C in 2017. By 2019, the SPINEA, s.r.o. production building was constructed on this site, featuring a metal roof, metal facade, and concrete parking lot, resulting in an LST exceeding 48°C by 2023. A similar transformation occurred at location 4, where agricultural land was replaced by an industrial zone – CTPark Prešov South, housing multiple companies. The entire area is paved with concrete surfaces and the buildings feature metal roofs and facades. This has led to an increase in LST values from 35°C in 2017 to nearly 52°C by 2023. The areas surrounding both sites are exclusively agricultural lands without trees. Vegetation consists mainly of shrubs near local streams, although these were not clearly detected by our image classifications due to the spatial resolution. However, pixels depicting water bodies in the actual landscape exhibit noticeably lower LSTs compared to the adjacent banks, which are lined with stones. Data validation To assess LSTs in the urban environment, validation data was collected using a data logger at five selected locations (A-E) within the urban area of Prešov. The locations and corresponding LST data are recorded in Fig. 9 and Table 6 . Table 6 Selected localities in the Prešov city, listing their kinetic surface temperatures measured by the Comet data logger and comparing them to the corresponding LST pixel value − 30 m pixel from Landsat-8/-9-derived LST c and downscaled 10 m pixel of LST f . Locality Characteristic LST - data logger (°C) LST c (°C) 09:26:16 LST f (°C) 09:26:16 07:00 10:00 13:00 A tartan 27.8 32.3 - 35.9 34.1 B artificial grass 29.0 35.8 30.1 37.3 38.4 C stone paving 30.9 40.1 - 40.3 42.7 D natural grass 29.6 35.6 - 43.0 36.9 E concrete 31.3 39.5 35.4 37.1 45.0 Figure 9 shows that areas with higher LST values correspond to specific forms of urban fabric (continuous urban fabric in the city centre, discontinuous urban fabric such as blocks of flats, commercial buildings, roads, parking lots, etc.). On the contrary, pixels displaying lower LSTs indicate areas with present vegetation, which contributes to cooling the surroundings. Location A is a tartan-covered athletic track located on the grounds of the elementary school grounds. Minimum LST values here hover around 30°C, occurring in areas with dense tree vegetation. The maximum LST, reaching approximately 47°C, is observed in the southwest part of the area, attributed to a light sheet metal roof of a department store. Location B was located near a football field with artificial grass, adjacent to the Torysa River. Typically, water surfaces exhibit lower LST values, but here the river's pixels show higher values likely due to lower spatial resolution. Along the nearby banks of the Torysa River, minimum LST values of around 27°C were identified. This cooling effect is attributed to the widespread vegetation cover that includes grassy areas, meadows, and dense tree stands. In contrast, the maximum value of LST in this area was recorded near the swimming pool, where the surfaces are made of brick and stone pavements, reaching over 56°C. Location C, in the center of Prešov, shows notably high LST values above 50°C. The minimum LST here is 34°C, the highest among the studied locations, due to dense urban fabric with red tin roofs and impermeable surfaces like stone streets and squares. Vegetation, such as trees in parks and along the main square, is less distinguishable in LST pixel values due to the satellite's lower resolution, which mainly reflects built-up areas. Location D is located within the southern park area near the University of Prešov, not far from the Torysa River. In particular, the central and eastern parts are densely built areas, with maximum LST values exceeding 46°C. On the contrary, the western part of this section comprises forested areas where the minimum LST values hover around 31°C. Location E encompasses Prešov’s largest housing estate, Sekčov, featuring shopping centres with metal roofs and extensive concrete parking lots. These surface characteristics contribute to maximum LST values near 50°C. Contrasting this, the western area adjacent to the built-up zone shows a minimum LST of approximately 31°C, characterized by long-term grassy areas interspersed with bushes. Table 6 shows that the LST values recorded by the handheld data logger (at 10:00 UTC) are consistently lower than the corresponding pixel values derived from satellite images (acquisition time: 09:26:16 UTC). This is primarily due to the higher accuracy and pointwise measurement capability of the data logger. A strong positive correlation of 0.98 was found between the data from the datalogger and the LST f and a correlation of 0.97 was found between the data from the datalogger and the LST c . In the morning at 7:00 a.m. under clear weather conditions, tartan recorded the lowest temperature at 27.8°C, followed by artificial grass at 29.6°C, and natural grass at 29.0°C. The stone paving reached 30.9°C, while the concrete had an LST of 31.3°C. At 10:00 a.m., with no change in cloud cover, these temperatures rose as follows: tartan increased by less than 5°C to 32.3°C, both natural and artificial grasses increased by 6.8°C to approximately 35.6°C and 35.8°C respectively. Stone paving and concrete showed the most significant increases in surface temperature, rising by almost 10°C to around 40°C. The surface LST values obtained from the data logger were compared with those derived from satellites at 30 m spatial resolution (before downscaling) and 10 m resolution (after downscaling) from Landsat-8 data acquired at 9:26:16 UTC. Even in the satellite-derived LST data, tartan showed the lowest value at 34.1°C, followed by natural grass at approximately 36.9°C, artificial grass at 38.4°C, stone paving at 42.7°C, and concrete that recorded the highest temperature at 45.0°C. Despite differences between the LST values obtained from the data logger and satellite data, there are similarities in the overall LST patterns observed with both methods. Both data sets show a consistent trend from the coldest to the warmest surface types, although the order of the LST data for concrete and stone paving changed after downscaling. Our findings indicate that man-made surfaces, such as concrete and stone pavements, experience rapid heating, up to 10°C in 3 hours under sunny conditions. These surfaces are prevalent in urban areas, contributing to elevated LSTs and thus heating their surroundings. On the contrary, natural surfaces, such as vegetation, exhibit lower LSTs, which help cool their environment. However, certain parts of natural grassy areas occasionally show slightly higher LST values, often due to damage such as exposed soils, which do not fully represent healthy grass samples (see Fig. 3 D, Fig. 9 ). Considering the prevalence of impermeable surfaces like concrete, asphalt, and stone in urban areas, the potential for the UHI phenomenon in the city is evident. Conclusions The effect of UHI and the increase in LST are concerns not only for large cities but also for smaller cities like Prešov in Slovakia. The main objective of this research was to analyze the spatial distribution and temporal changes of LSTs in the city of Prešov. This analysis used a downscaling technique, which integrated satellite observations from the Landsat-8, Landsat-9, and Sentinel-2 missions. Despite not having the thermal band, Sentinel-2 provides higher sptial resolution for estimating the thermal emissivity of LC for LST calculation from Landsat-8/-9 thermal data. A significant correlation was discovered between spectral indices (NDBI, NDVI, NDWI) and LST that was leveraged to downscale the native LST c with a 30 m spatial resolution to a finer LST f with a 10 m resolution, using a multiple linear regression model. This improved resolution allowed a better identification of changes in LST during the years 2017, 2019, and 2023. Furthermore, two machine learning algorithms, SVM and RF, were employed to classify images and assess the influence of varying LC types and changes throughout the years 2017, 2019, and 2023 on the city's LST patterns. Evaluation of LC changes using the RF classification algorithm revealed higher accuracy compared to SVM - RF outperformed SVM by 0.8% in 2017 and 4.3% in 2023. The results show that over recent years, a previously natural landscape has been extensively altered, primarily in suburban areas that cause the increase of LSTs, thus influencing the emergence and intensification of the UHI phenomenon. These changes were mainly driven by the construction of a highway bypass and the expansion of industrial zones. It was observed that impervious man-made surfaces, such as concrete and stone pavements, experience rapid heating, up to 10°C in 3 hours, subsequently warming the surrounding areas. Therefore, strategic urban planning and city development are essential to promote sustainable expansion and protect urban greenery, ultimately improving the quality of life of city residents. Declarations Acknowledgements The authors are very grateful for the detailed and constructive comments to improve the manuscript provided by the editors and anonymous reviewers. Author contribution Anton Uhrin: Conceptualization, Methodology, Investigation, Validation, Formal analysis, Resources, Writing—original draft preparation. Katarína Onačillová: Conceptualization, Methodology, Visualization, Review and Editing, Funding. Data availability Some or all data, models, or codes that support the findings of this study are available from the corresponding author upon reasonable request. Funding This study was funded by the Slovak Research and Development Agency (APVV) under the contract No. APVV-23-0210 and by the Scientific Grant Agency of the Ministry of Education, Science, Research and Sport of the Slovak Republic and the Slovak Academy of Sciences (VEGA) under the contract No. VEGA 1/0085/23. Ethical Approval All authors have read, understood, and have complied as applicable with the statement on “Ethical responsibilities of Authors” as found in the Instructions for Authors. Competing interests The authors declare no competing interests. References BONAFONI, S., ANNIBALLE, R., GIOLI, B. & TOSCANO, P. (2016). Downscaling Landsat Land Surface Temperature over the urban area of Florence. European Journal of Remote Sensing , 49, 553–569. https://doi.org/10.5721/EuJRS20164929. CAMPS, A., PARK, H., CASTELLVI, J., CORBERA, J. & ASCASO, E. (2020). Single-Pass Soil Moisture Retrievals Using GNSS-R: Lessons Learned. Remote Sensing , 12, 2064. https://doi.org/10.3390/rs12122064. CHEVAL, S., DUMITRESCU, A. & PETRIŞOR, A.-I. (2011). The July Surface Temperature Lapse In The Romanian Carpathians. Carpathian Journal of Earth and Environmental Sciences , 6(1), 189–198. CHOWDHURY, M. S. 2024. Comparison of accuracy and reliability of random forest, support vector machine, artificial neural network and maximum likelihood method in land use/cover classification of urban setting. Environmental Challenges , 14, 100800. https://doi.org/10.1016/j.envc.2023.100800. COPERNICUS LAND MONITORING SERVICE (2018). Urban Atlas 2018. Retrieved June 10, 2017, from http://land.copernicus.eu/local/urban-atlas/urban-atlas-2012. DOBROVOLNÝ, P. (2013). The surface urban heat island in the city of Brno (Czech Republic) derived from land surface temperatures and selected reasons for its spatial variability. Theoretical Applied Climatology , 112, 89–98. https://doi.org/10.1007/s00704-012-0717-8. FADHIL, M., HAMOODI, M. N. & ZIBOON, A. R. T. (2023). Mitigating urban heat island effects in urban environments: strategies and tools. IOP Conference Series: Earth and Environmental Science , 1129, 012025. https://doi.org/10.1088/1755-1315/1129/1/012025. GAO, F., KUSTAS, W. P. & ANDERSON, M. C. (2012). A data mining approach for sharpening thermal satellite imagery over land. Remote Sensing , 4, 3287–3319. https://doi.org/10.3390/rs4113287. GASCON, F., BOUZINAC, C., THÉPAUT, O., JUNG, M., FRANCESCONI, B., LOUIS, J., LONJOU, V., LAFRANCE, B., MASSERA, S. & GAUDEL-VACARESSE, A. (2017). Copernicus Sentinel-2A Calibration and Products Validation Status. Remote Sensing , 9(6), 584. https://doi.org/10.3390/rs9060584. GOLDBLATT, R., STUHLMACHER, M. F., TELLMAN, B., CLINTON, N., HANSON, G., GEORGESCU, M., WANG, CH., SERRANO-CANDELA, F., KHANDELWAL, A. K., CHENG, W. H. & BALLING, R. C. (2018). Using Landsat and nighttime lights for supervised pixel-based image classification of urban land cover. Remote Sensing of Environment , 205, 253–275. https://doi.org/10.1016/j.rse.2017.11.026. GOOGLE EARTH ENGINE (2024). RF Classifier. Retrieved March, 2024, from https://developers.google.com/earth-engine/apidocs/ee-classifier-smilerandomforest. Last access: 13 March 2024. GRIGORAȘ, G. & URIȚESCU, B. (2019). Land Use/Land Cover changes dynamics and their effects on Surface Urban Heat Island in Bucharest, Romania. International Journal of Applied Earth Observation and Geoinformation , 80, 115–126. https://doi.org/10.1016/j.jag.2019.03.009. GUO, X., WANG, M., JIA, M. & WANG, W. (2021). Estimating mangrove leaf area index based on red-edge vegetation indices: A comparison among UAV, WorldView-2 and Sentinel-2 imagery. International Journal of Applied Earth Observation and Geoinformation , 103, 102493. https://doi.org/10.1016/j.jag.2021.102493. HAN, L., LU, L., FU, P., REN, C., CAI, M. & LI, Q. (2023). Exploring the seasonality of surface urban heat islands using enhanced land surface temperature in a semi-arid city. Urban Climate , 49, 101455. DOI: https://doi.org/10.1016/j.uclim.2023.101455. HU, D., GUO, F., MENG, Q., SCHLINK, U., WANG, S., HERTEL, D. & GAO, J. (2024). A novel dual-layer composite framework for downscaling urban land surface temperature coupled with spatial autocorrelation and spatial heterogeneity . International Journal of Applied Earth Observation and Geoinformation , 130, 103900. https://doi.org/10.1016/j.jag.2024.103900. HUTENGS, C., VOHLAND, M. (2016). Downscaling land surface temperatures at regional scales with random forest regression. Remote Sensing of Environment , 178, 127–141. https://doi.org/10.1016/j.rse.2016.03.006. IVAJNŠIČ, D. & ŽIBERNA, I. (2018). The effect of weather patterns on winter small city urban heat islands. Meteorological Applications , 2018, 1–9. https://doi.org/10.1002/met.1752. KHAN, M., QASIM, M., TAHIR, A. A. & FAROOQI, A. (2023). Machine learning-based assessment and simulation of land use modification effects on seasonal and annual land surface temperature variations. Heliyon , 9(12), e23043. https://doi.org/10.1016/j.heliyon.2023.e23043. KOTTEK, M., GRIESER, J., BECK, C., RUDOLF, B. & RUBEL, F. (2006). World Map of the Köppen-Geiger Climate Classification Updated. Meteorologishe Zeitschrift , 15, 259–263. https://doi.org/10.1127/0941-2948/2006/0130. LONGLEY, I. D. GALLAGHER, M.W., DORSEY, J. R., FLYNN, M. & BARLOW, J. F. (2004). Short-term measurements of airflowand turbulence in two street canyons in Manchester. Atmospheric Environment , 38, 69–79. https://doi.org/10.1016/j.atmosenv.2003.09.060. MCFEETERS, S. K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International Journal of Remote Sensing , 17, 1425–1432. http://dx.doi.org/10.1080/01431169608948714. MILES, V., ESAU, I. & MILES, M. W. (2023). The urban climate of the largest cities of the European Arctic. Urban Climate , 48, 1–12. https://doi.org/10.1016/j.uclim.2023.101423. ONAČILLOVÁ, K., GALLAY, M., PALUBA, D., PÉLIOVÁ, A., TOKARČÍK, O. & LAUBERTOVÁ, D. (2022). Combining Landsat 8 and Sentinel-2 data in Google Earth Engine to derive higher resolution land surface temperature maps in urban environment. Remote Sensing , 14(16), 4076. https://doi.org/10.3390/rs14164076. PATIL, R. & SURAWAR, M. (2023). Impact of Urban Heat Island on Formation of Precipitation in Indian Western Coastal Cities. Journal of Contemporary Urban Affairs , 7(2), 37–55. https://doi.org/10.25034/ijcua.2023.v7n2-3. PUREVDORJ, T., TATEISHI, R., ISHIYAMA, T. & HONDA, Y. (1998). Relationships between percent vegetation cover and vegetation indices. International Journal of Remote Sensing , 19, 3519–3535. https://doi.org/10.1080/014311698213795. PURIO, M. A. C., YOSHITAKE, T. & CHO, M. (2022). Assessment of Intra-Urban heat Island in a densely populated city using Remote sensing: A case study for Manila City. Remote Sensing , 14(21), 5573. https://doi.org/10.3390/rs14215573. SHAMSAEI, M., CARTER, A. & VAILLANCOURT, M. (2022). A review on the heat transfer in asphalt pavements and urban heat island mitigation methods. Construction and Building Materials , 359, 129350. https://doi.org/10.1016/j.conbuildmat.2022.129350. SONG, B. & PARK, K. (2020). Verification of Accuracy of Unmanned Aerial Vehicle (UAV) Land Surface Temperature Images Using In-Situ Data. Remote Sensing , 12(2), 288. https://doi.org/10.3390/s12020288. STATISTICAL OFFICE SR (2022). Public database of the Slovak Statistical Office of the Slovak Republic . Retrieved February 8, 2024, from https://slovak.statistics.sk. Last access: 8 February 2024. STEENSEN, B. M., MARELLE, L., HODNEBROG, Ø. ET AL. (2022). Future urban heat island influence on precipitation. Climate Dynamics , 58, 3393–3403. https://doi.org/10.1007/s00382-021-06105-z. SVOBODA, J., ŠTYCH, P., LAŠTOVIČKA, J., PALUBA, D. & KOBLIUK, N. (2022). Random Forest Classification of Land Use, Land-Use Change and Forestry (LULUCF) UsingSentinel-2 Data – A Case Study of Czechia. Remote Sensing , 14, 1189. https://doi.org/10.3390/rs14051189. TANOORI, G., SOLTANI, A. & MODIRI, A. (2024). Machine Learning for Urban Heat Island (UHI) analysis: Predicting Land Surface Temperature (LST) in Urban Environments. Urban Climate , 55, 101962. https://doi.org/10.1016/j.uclim.2024.101962. USGS (2024). Landsat 8-9 Collection 2 (C2) Level 2 Science Product (L2SP) Guide. Retrieved May 9, 2024, from https://www.usgs.gov/media/files/landsat-8-9-coll-ection-2-level-2-science-product-guide. VARDOULAKIS, E., KARAMANIS, D., FOTIADI, A. & MIHALAKAKOU, G.,2013. The urban heat island effect in a small Mediterranean city of high summer temperatures and cooling energy demands. Solar Energy , 94, 128–144. https://doi.org/ 10.1016/j.solener.2013.04.016. WEILAND, L., ROGERS, C. A., SOTHE, C., ARAIN, M. A. & GONSAMO, A. (2023). Satellite-based land surface temperature and soil moisture observations accurately predict soil respiration in temperate deciduous and coniferous forests. Agricultural and Forest Meteorology , 340, 109618. https://doi.org/10.1016/j.agrformet.2023.109618. XIONG, Y., CHEN, X., TANG, L. & WANG, H. (2022). Comparison of surface renewal and Bowen ratio derived evapotranspiration measurements in an arid vineyard. Journal of Hydrology , 613, 128474. https://doi.org/10.1016/j.jhydrol.2022.128474. XU, S., ZHAO, Q., YIN, K., HE, G., ZHANG, Z., WANG, G., WEN, M. & ZHANG, N. (2021). Spatial downscaling of land surface temperature based on a Multi-Factor geographically weighted machine learning model. Remote Sensing , 13(6), 1186. https://doi.org/10.3390/rs13061186. YAN, W., JIANG, J., HE, L., ZHAO, W., NAIR, R., WANG, X. & XIONG, Y. (2024). Correcting land surface temperature from thermal imager by considering heterogeneous emissivity. International Journal of Applied Earth Observation and Geoinformation , 129, 103824. https://doi.org/10.1016/j.jag.2024.103824. ZAFAR, Z., ZUBAIR, M., ZHA, Y., FAHD, S. & NADEEM, A. A. (2024). Performance assessment of machine learning algorithms for mapping of land use/land cover using remote sensing data. The Egyptian Journal of Remote Sensing and Space Sciences , 27(2), 216–226. https://doi.org/10.1016/j.ejrs.2024.03.003. ZHA, Y., GAO, J. & NI, S. (2003). Use of normalized difference built-up index in automatically mapping urban areas from TM imagery. International Journal of Remote Sensing , 24, 583–594. https://doi.org/10.1080/01431160304987. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 03 Jan, 2025 Read the published version in Environmental Monitoring and Assessment → Version 1 posted Editorial decision: Revision requested 01 Oct, 2024 Editor assigned by journal 30 Sep, 2024 Submission checks completed at journal 30 Sep, 2024 First submitted to journal 24 Sep, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5143836","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":361309911,"identity":"f8ad02ed-cf0f-4766-9212-59372e360f21","order_by":0,"name":"Anton Uhrin","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABFUlEQVRIiWNgGAWjYBCDBAZmBsYHQAZjA0SADYqxAWa4FmYDIAXUwkysFqAKCSQtDDi18M/uP/i4gOFwnnk777HKnz8Oy/ZL9x9g5t3BZ8/AfywBmxaJO4eZjWcwHC6WOcyXdpsn4bDxzDmHGZh5z7AlNkikHcBqzY1kNmkehsOJM5h5zG4zJBxO3HAjGailjS2BQYK9AZsOeWQthT+AWvZDtQAddhyrFgNkLQw8IFskIFqA4YDdYYY3ko2NeQzSQVqMpXnS0o1n3Eg2ODgX6Jc2iTSs3pe7kfjwMU+FdeIM/jOGH3/YWMv2z0h8+ODtjmP2/PzHDLB6H+I8NP4BxoZjOCMSB2BsqCFNwygYBaNgFAxnAAATuViXKdh6HAAAAABJRU5ErkJggg==","orcid":"","institution":"Pavol Jozef Šafárik University","correspondingAuthor":true,"prefix":"","firstName":"Anton","middleName":"","lastName":"Uhrin","suffix":""},{"id":361309912,"identity":"6988c60d-196f-4702-b655-29ecaaa5b1a9","order_by":1,"name":"Katarína Onačillová","email":"","orcid":"","institution":"Pavol Jozef Šafárik University","correspondingAuthor":false,"prefix":"","firstName":"Katarína","middleName":"","lastName":"Onačillová","suffix":""}],"badges":[],"createdAt":"2024-09-24 09:38:45","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5143836/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5143836/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s10661-024-13598-8","type":"published","date":"2025-01-03T15:57:14+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":70949236,"identity":"7781d7da-b2ec-4abb-aa51-415f2df49365","added_by":"auto","created_at":"2024-12-09 13:24:02","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":4627102,"visible":true,"origin":"","legend":"\u003cp\u003eLocation of the study area of the Prešov city. The background maps are: © ESRI. \"World Imagery\", \"Terrain with labels\", acquired on March 7, 2024.\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-5143836/v1/1edcb697edbebf16813b09b6.png"},{"id":70948044,"identity":"e1fbfb8e-4ec5-4ea3-ab60-9bb24cb2f4cf","added_by":"auto","created_at":"2024-12-09 13:16:02","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1268014,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart summarizing the basic steps of LST downscaling and image classification using ML algorithms.\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-5143836/v1/8e90da59a7a8711bf6f7e48a.png"},{"id":70949551,"identity":"f532d3b3-fd39-4efa-97a5-48f0f66fb43f","added_by":"auto","created_at":"2024-12-09 13:32:02","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":5432950,"visible":true,"origin":"","legend":"\u003cp\u003eFive locations, which were the subject of field validation measurements within the area of interest: A: tartan, B: artificial grass, C: stone paving, D: grass, and E: concrete; Source: photos A – E: author´s archive, Background maps are: © ESRI. \"World Imagery\", acquired on August 23, 2023.\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-5143836/v1/752a84b2b1dd3c3ae51870da.png"},{"id":70948047,"identity":"98b1c81f-7a4f-4220-bf4f-d2a3e3c38ef5","added_by":"auto","created_at":"2024-12-09 13:16:02","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":12496871,"visible":true,"origin":"","legend":"\u003cp\u003eNDVI, NDWI and NDBI spectral indices derived from (A) Landsat-9 OLI satellite data in 30-meter spatial resolution for 19 July 2023, (B) Sentinel-2 MSI satellite data in 10-meter spatial resolution for 18 July 2023, for the area of interest encompassing the city of Prešov and surrounding municipalities.\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-5143836/v1/77123f19da7de7edeca621dc.png"},{"id":70948048,"identity":"672367f0-a396-49f4-862a-6d1b70f139f8","added_by":"auto","created_at":"2024-12-09 13:16:02","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":4039450,"visible":true,"origin":"","legend":"\u003cp\u003e(\u003cstrong\u003eA\u003c/strong\u003e) Coarse LST\u003csub\u003ec\u003c/sub\u003e derived from Landsat-8 in spatial resolution of 30 m, (\u003cstrong\u003eB\u003c/strong\u003e) Downscaled LST\u003csub\u003ef\u003c/sub\u003e in spatial resolution of 10 m, for the study area of the city of Prešov.\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-5143836/v1/7f02f3fce72c6b50e4eb72a9.png"},{"id":70948049,"identity":"d9864e9b-fe26-4809-afc8-7c64530c72df","added_by":"auto","created_at":"2024-12-09 13:16:02","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":23518490,"visible":true,"origin":"","legend":"\u003cp\u003eLocations 1 – 4 where significant changes in LST between the years 2017 and 2023 were detected. Source: LST maps: author's processing, the background true color map (on the right) is: © ESRI. \"World Imagery\".\u003c/p\u003e","description":"","filename":"Fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-5143836/v1/590543fbc6ab0b6156bbada8.png"},{"id":70948053,"identity":"46a55e94-8736-4863-aa2d-6ee46a2aae0e","added_by":"auto","created_at":"2024-12-09 13:16:02","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":44552141,"visible":true,"origin":"","legend":"\u003cp\u003eSupervised classification using \u003cstrong\u003eA\u003c/strong\u003e) SVM and \u003cstrong\u003eB\u003c/strong\u003e) RF classification algorithms.\u003c/p\u003e","description":"","filename":"Fig7.png","url":"https://assets-eu.researchsquare.com/files/rs-5143836/v1/3d714beca3a9771b7d52c98a.png"},{"id":70948046,"identity":"4cd32222-a74d-4a0a-84c6-3385a1026a6d","added_by":"auto","created_at":"2024-12-09 13:16:02","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":4442914,"visible":true,"origin":"","legend":"\u003cp\u003eLC changes evaluated using supervised classification employing the Random Forest algorithm, along with the respective downscaled LST for the years 2017, 2019, and 2023 at locations 1-4.\u003c/p\u003e","description":"","filename":"Fig8.png","url":"https://assets-eu.researchsquare.com/files/rs-5143836/v1/a5461ab3a71ff0fefeb4bdfa.png"},{"id":70949237,"identity":"a951445a-68b3-49cf-9760-9bf4ecc258b0","added_by":"auto","created_at":"2024-12-09 13:24:02","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":18112060,"visible":true,"origin":"","legend":"\u003cp\u003eLocations A-E where the validation data were collected, the background images - true color compositions and their corresponding LSTs. The background true color map (on the left) is: © ESRI. \"World Imagery\".\u003c/p\u003e","description":"","filename":"Fig9.png","url":"https://assets-eu.researchsquare.com/files/rs-5143836/v1/e7c22b3e5deafdc5fe1e4656.png"},{"id":73093225,"identity":"fb39086b-bb37-48da-83ba-30e2dfaef1a8","added_by":"auto","created_at":"2025-01-06 16:11:09","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":108092161,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5143836/v1/4fd8420c-de53-47a5-b8c4-cf79217df650.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Spatio-temporal analysis of land surface temperature and land cover changes in Prešov city using downscaling approach and machine learning algorithms","fulltext":[{"header":"Introduction","content":"\u003cp\u003eUrbanization and population growth result in an uneven distribution of surfaces that absorb substantial solar radiation (Purio et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), thus altering land cover (LC) dynamics. These changes affect critical processes such as heat transfer (Shamsaei et al., 2022), air circulation (Longley et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2004\u003c/span\u003e), precipitation patterns (Patil \u0026amp; Surawar, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Steensen et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), and urban thermal environments (Fadhil et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). This leads to a transformation of the microclimate, resulting in significant temperature differences between urban and rural areas, known as the Urban Heat Island (UHI) effect (Grigoraș \u0026amp; Urițescu, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Studies have shown that even in small and medium-sized cities, factors such as urbanization, reduced vegetation, and increased human activity contribute to elevated temperatures compared to surrounding rural areas. Despite the significant impact of UHI on local climates, energy consumption, and public health in these smaller cities, research in this area remains limited, although some studies have addressed this gap (Miles et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Ivajnšič \u0026amp; Žiberna, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Vardoulakis et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Dobrovoln\u0026yacute;, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2013\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eStudying the UHI effect relies heavily on analyzing Land Surface Temperature (LST) data, which provides critical insights into spatial and temporal variations in surface temperatures. This understanding is essential to assess urban heat fluxes, identifying the presence and intensity of the UHI phenomenon (Tanoori et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Han et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), simulate surface energy exchange (Khan et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and study the energy balance of the surface at global, regional and local scales (Yan et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). LST is also widely used to assess surface moisture and drought levels (Xiong et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Weiland et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eA robust and efficient method for acquiring LST data involves the use of freely available thermal satellite imagery, facilitating a comprehensive analysis of temperature patterns across broad geographic regions. In urban environments, the spatial and temporal heterogeneity of surface temperature is pronounced due to complex surface structures and materials, reflecting the diversity of urban LC. However, freely available data from thermal sensors typically have coarser spatial resolution and higher temporal resolution, which is inadequate for accurately identifying thermal characteristics of urban areas and conducting detailed analyses of factors influencing the UHI formation (Song and Park, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). The resolution of acquired thermal data varies, ranging from 1000 m for AVHRR to 100 m for Landsat-8/-9, making it more suitable for global to regional scale studies (Cheval et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2011\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAdvancements in satellite technology, improvements in multispectral and thermal sensors, rapid development of GIS and remote sensing methods, along with the availability of open-source platforms such as Google Earth Engine (GEE), are crucial for overcoming these limitations and enabling higher-resolution LST mapping. One of the most widely used approaches to achieve higher spatial resolution in satellite imagery for deriving LST are downscaling methods, transforming a coarser spatial resolution image into a higher spatial resolution image (Hu et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). There are several downscaling approaches or models that play a role in facilitating LST analyses and reducing computational time and storage space (Xu et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Hutengs \u0026amp; Vohland, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2016\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThis work offers one of the first investigations into a complex spatio-temporal analysis of recent LC and LST changes in the city of Prešov, that has been not subject of UHI studies before (to date) on the local scale for the years 2017, 2019 and 2023 with the use of an open-source cloud platform GEE. We derive LST within the urban environment of the Prešov city at higher spatial resolution using downscaling of Landsat thermal satellite data with multispectral indices derived from Sentinel 2 data. We also use semi-automatic classification of satellite data (Sentinel-2) using selected machine learning algorithms - Support Vector Machine and Random Forest. Information about LST of different urban surfaces and land use types can help to suggest effective strategies in the areas of efficient urbanisation, urban planning, or sustainable development proposals to minimise the negative impacts of climate change and UHIs to optimise or enhance the quality of life of urban residents.\u003c/p\u003e\n\u003ch3\u003eStudy area\u003c/h3\u003e\n\u003cp\u003eThe area of interest is the medium-sized city of Prešov with the adjacent municipalities of Haniska, Kendice, Ľubotice, Petrovany and Z\u0026aacute;borsk\u0026eacute;, located in eastern Slovakia, and characterised by a typical Central European urban environment with significant heterogeneity of LC classes (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The total area of the city is approximately 70.44 km\u003csup\u003e2\u003c/sup\u003e with a total population of 84,824 inhabitants, which represents approximately 1204.20 inhabitants per 1 km2 (Statistical Office SR, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAgricultural land occupies 35%, and forests account for 31%. Artificial surfaces are also highly represented, mainly by built-up areas that occupy 28% \u0026minus; continuous urban fabric accounts for 4%, discontinuous urban fabric accounts for 8%, and more than 7% are occupied by industrial, commercial, public, military, and private transport units. The road and rail network and associated land account for less than 4%. Sports and leisure facilities comprise 2% and green urban areas account for 1.5% (Copernicus Land Monitoring Service, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAccording to K\u0026ouml;ppen's climate classification, the territory of Prešov lies in the mild continental climate zone, classification subtype \"Dfb\", which is characterised by significant seasonality, with warm and humid summers, severe winters and short dry periods (Kotek et al., 2006). The largest watercourse that flows through the territory of the city is the Torysa River and its left-hand tributaries Sekčov and Delňa.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003eThe research in this study uses multispectral imagery collected by two satellite missions - the Landsat-8/-9 mission by NASA and Sentinel-2 mission by ESA. Six cloud-free multispectral satellite products obtained over the Pre\u0026scaron;ov area were employed for selected days in 2017, 2019 and 2023: Landsat-8 OLI/TIRS sensor (3 products) and MSI sensor onboard the Sentinel-2 satellite (3 products). These six satellite Landsat-8/9 and Sentinel-2 data products were selected based on the proximity of their acquisition dates within the same year, considering similar atmospheric conditions and the lowest possible percentage of cloud cover. The corresponding air temperature parameters extracted from the OGIMET service developed by the Spanish Meteorological Institute (\u003cspan\u003e\u003cspan\u003ewww.ogimet.com\u003c/span\u003e\u003c/span\u003e) are stated in Table \u003cspan\u003e1\u003c/span\u003e.\u003c/p\u003e \u0026nbsp;\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 1\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eSelected parameters of the satellite products and the corresponding air temperature based on meteorological data from SHM\u0026Uacute; (2024).\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eSatellite, sensor and product\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eAcquisition\u003c/p\u003e\n \u003cp\u003edate\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eAcquisition time [UTC]\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eCloud cover\u003c/p\u003e\n \u003cp\u003e[%]\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eAir temperature [\u0026deg;C]\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e09:00\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e10:00\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLandsat-8 OLI/TIRS L2 C2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11.08.2017\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e09:26:41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e29.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e30.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLandsat-8 OLI/TIRS L2 C2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e30.06.2019\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e09:26:34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e25.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e27.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLandsat-9 OLI/TIRS L2 C2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e19.07.2023\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e09:26:16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e12.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e25.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e26.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSentinel-2A MSI L2A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e03.08.2017\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e09:40:46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e29.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e30.7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSentinel-2A MSI L2A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e01.07.2019\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e09:30:39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e27.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e28.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSentinel-2B MSI L2A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e18.07.2023\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e09:35:49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e20.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e22.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThree USGS Landsat-8/-9 Level-2 Collection-2 Surface Reflectance products were used to derive three selected spectral indices that are closely related to LST at spatial resolution of 30 m and to calculate atmospherically corrected LST itself using a modified radiative transfer equation inversion methodology:\u003c/p\u003e\n\u003cp\u003e\u0026quot;LC08_L1TP_187026_20170811_20200903_02_T1\u0026quot; (11/08/2017, 09:26:41 UTC) for 2017,\u003c/p\u003e\n\u003cp\u003e\u0026quot;LC08_L1TP_187026_20190630_20200827_02_T1\u0026quot; (30/06/2019, 09:26:34 UTC) for 2019,\u003c/p\u003e\n\u003cp\u003e\u0026quot;LC09_L1TP_187026_20230719_20230719_02_T1\u0026quot; (2023/07/19, 09:26:16 UTC) for 2023.\u003c/p\u003e\n\u003cp\u003eThree Harmonized Sentinel-2 MSI: MultiSpectral Instrument, Level-2A products were used to obtain the same spectral indices as in the case of the Landsat-8 satellite, but in a higher, 10 m spatial resolution, which has been used in the downscaling process. The products were selected based on the proximity of the acquisition date to the selected products of the Landsat-8/-9 satellite:\u003c/p\u003e\n\u003cp\u003e\u0026quot;S2A_MSIL2A_20170803T094031_N0205_R036_T34UEV_20170803T094046\u0026quot; (08/03/2017, 09:40:46 UTC) for 2017,\u003c/p\u003e\n\u003cp\u003e\u0026quot;S2A_MSIL2A_20190701T093041_N0212_R136_T34UEV_20190701T115615\u0026quot; (07/01/2019, 09:30:39 UTC) for 2019,\u003c/p\u003e\n\u003cp\u003e\u0026quot;S2B_MSIL2A_20230718T093549_N0509_R036_T34UEV_20230718T112309\u0026quot; (2023-07-18, 09:35:49 UTC) for 2023.\u003c/p\u003e\n\u003cp\u003eLevel-2A products provide users with orthorectified Bottom-Of-Atmosphere surface reflectance, the classification of scene, including cloud cover and cloud shadows, aerosol optical thickness, and water vapour maps (Gascon et al., \u003cspan\u003e2017\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eThe workflow is fully coded in JavaScript using the Code Editor Platform of the online cloud application GEE. The multispectral imagery was processed using the pipeline shown in Fig.\u0026nbsp;\u003cspan\u003e2\u003c/span\u003e and the script itself is divided into two main parts: processes associated with LST downscaling approach, described in detail in Subsection 3.1, and processes of semi-automatic image classification using machine learning algorithms, which are described in Subsection 3.2.\u003c/p\u003e\n\u003ch3\u003eValidation data\u003c/h3\u003e\n\u003cp\u003eTo validate the LST downscaling results, field measurements of surface kinetic temperature obtained by temperature probe Pt1000TG7/E with a Comet data logger were used at five selected open-sky locations within the study area to ensure their reliable identification even in satellite images and to represent various types of surfaces, as shown in Fig.\u0026nbsp;\u003cspan\u003e3\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eTwo natural materials \u0026ndash; stone pavement and grass \u0026ndash; and three artificial surfaces \u0026ndash; tartan, artificial grass, and concrete \u0026ndash; were selected. The device allows measurement of the surface temperature in 5-s measurement intervals in the range of -30 to 200\u0026deg;C with a measurement accuracy of 0.15\u0026deg;C (TR Instruments, 2023). Validation measurements were carried out in the area of interest on 19 July 2023 at 7:00 a.m. and 10:00 a.m. to match the acquisition time of the Landsat-9 satellite on the same day, whose thermal data were used to derive LST (acquisition time 09:26:16 UTC).\u003c/p\u003e\n\u003cp\u003eLST downscaling methodology\u003c/p\u003e\n\u003cp\u003eLandsat-8/-9 TIRS allows thermal radiation to be mapped at a native 100 m spatial resolution, which is resampled by United States Geological Survey (USGS) to 30 m for consistency with other spectral bands of these satellites. Thermal radiation can be later converted to LST; however, for detailed analysis of LSTs in heterogeneous urban areas, the coarse spatial resolution of 30 m and longer intervals between acquisitions of these satellites are not sufficient to capture interactions of LST in terms of the heterogeneous character of LC. On the contrary, Sentinel-2 does not have a thermal band, so it is not possible to directly derive LST. However, due to its high spatial resolution (native 10 m resolution in visible and near-infrared (NIR) bands) and temporal resolution (recording every 6 days), this satellite has high potential for calculating various spectral indices, allowing to accurately capture the complexity of urban LC in better detail compared to the Landsat-8/-9 satellites. In this study, the spatial downscaling approach involves establishing a multiple linear model utilizing selected three spectral indices and LST obtained from the thermal band of the Landsat-8/-9 satellite according to the model adopted from Bonafoni et al. (\u003cspan\u003e2016\u003c/span\u003e) and Onačillov\u0026aacute; et al. (\u003cspan\u003e2022\u003c/span\u003e). The proposed model is capable to predict LSTs in 10 m spatial resolution by using spectral indices obtained from Sentinel-2 satellite data as input.\u003c/p\u003e\n\u003ch3\u003eCalculation of LST predictors - spectral indices\u003c/h3\u003e\n\u003cp\u003eThe first step involved preparation of LST predictors: three spectral indices ̶ NDVI (Purevdorj et al. 1989; Guo et al., \u003cspan\u003e2021\u003c/span\u003e), NDBI (Zha et al., \u003cspan\u003e2003\u003c/span\u003e; Goldblatt et al., \u003cspan\u003e2018\u003c/span\u003e) and NDWI (McFeeters, \u003cspan\u003e1996\u003c/span\u003e; Camps et al., \u003cspan\u003e2020\u003c/span\u003e) ̶ were chosen and derived from Landsat-8/-9 satellite images with a 30 m spatial resolution and Sentinel-2 satellite images with a 10 m spatial resolution. The formulas for calculating the selected spectral indices of the Landsat-8/9 and Sentinel-2 satellites are stated in Table\u0026nbsp;\u003cspan\u003e2\u003c/span\u003e.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 2\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eSpectral indices used for LST downscaling.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"5\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAcronym\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDescription\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFormulation\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModified Formulation for Landsat-8/-9 Bands\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModified Formulation for Sentinel-2 Bands\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNDVI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNormalized difference vegetation index\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(NIR\u0026thinsp;\u0026minus;\u0026thinsp;RED/\u003c/p\u003e\n \u003cp\u003e(NIR\u0026thinsp;+\u0026thinsp;RED)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(B5\u0026thinsp;\u0026minus;\u0026thinsp;B4)/(B5\u0026thinsp;+\u0026thinsp;B4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(B8\u0026thinsp;\u0026minus;\u0026thinsp;B4)/(B8\u0026thinsp;+\u0026thinsp;B4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNDWI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNormalized difference water index\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(Green\u0026thinsp;\u0026minus;\u0026thinsp;NIR)/\u003c/p\u003e\n \u003cp\u003e(Green\u0026thinsp;+\u0026thinsp;NIR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(B3\u0026thinsp;\u0026minus;\u0026thinsp;B5)/(B3\u0026thinsp;+\u0026thinsp;B5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(B3\u0026thinsp;\u0026minus;\u0026thinsp;B8)/(B3\u0026thinsp;+\u0026thinsp;B8)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNDBI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNormalized difference built-up index\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(SWIR1\u0026thinsp;\u0026minus;\u0026thinsp;NIR)/\u003c/p\u003e\n \u003cp\u003e(SWIR1\u0026thinsp;+\u0026thinsp;NIR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(B7\u0026thinsp;\u0026minus;\u0026thinsp;B5)/(B7\u0026thinsp;+\u0026thinsp;B5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(B12\u0026thinsp;\u0026minus;\u0026thinsp;B8)/(B12\u0026thinsp;+\u0026thinsp;B8)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eTo derive NDVI from the Landsat-8/9 satellite, the NIR band with wavelengths of 0.85\u0026ndash;0.88 \u0026micro;m and the red (R) band at wavelengths of 0.64\u0026ndash;0.67 \u0026micro;m were used. The input spectral bands for the NDWI derivation were the green band with wavelengths of 0.533\u0026ndash;0.590 \u0026micro;m and the NIR band. For the NDBI derivation, these were the NIR and shortwave infrared 1 (SWIR1) bands with a wavelength of 1.57\u0026ndash;1.65 \u0026micro;m. All these bands had a native 30 m spatial resolution.\u003c/p\u003e\n\u003cp\u003eEquivalent spectral indices were derived from Sentinel-2 satellite data: the NIR band with a central wavelength of 0.84 \u0026micro;m and the red band with a central wavelength of 0.66 \u0026micro;m were used to calculate the NDVI. The green band with a central wavelength of 0.56 \u0026micro;m and the NIR band were used to obtain the NDWI. NDBI was calculated based on the NIR and SWIR1 bands with a central wavelength of 1.61 \u0026micro;m (resampled from native spatial resolution of 20 to 10 m to ensure consistency with the resolution of NDVI and NDWI) (Fig.\u0026nbsp;\u003cspan\u003e4\u003c/span\u003e).\u003c/p\u003e\n\u003ch3\u003eMultiple linear regression model between LC indices and LST\u003c/h3\u003e\n\u003cp\u003eIn this study, the spatial downscaling approach relies on establishing a linear relation utilizing three spectral indices derived from Landsat-8/9 imagery (NDVI, NDWI, NDBI) as predictors, with LST serving as the predicted variable. Then the established model is applied to the study areas with finer resolution. The LST with spatial resolution of 30 m was derived from the surface temperature (ST_B10) of the Landsat-8/9 TIRS sensor. ST_B10 is derived from Landsat-8 Collection 2 Level 1 Thermal Infrared Sensor (TIRS) Band 10 data (USGS, \u003cspan\u003e2024\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eThen, the multiple linear regression statistical model between LST and spectral indices was constructed as predictors at coarser (30 m) resolution. The model for the calculation of LST at the coarser level from Landsat-8/-9 imagery using the output regression coefficients to calculate the LST in 30 m spatial resolution can be expressed as follows:\u003c/p\u003e\n\u003cdiv id=\"Equ1\"\u003e\n \u003cdiv id=\"FileID_Equ1\" name=\"EquationSource\"\u003e$$\\:{LST}_{c}={a}_{0}+\\:{a}_{1}\\times\\:{NDVI}_{c}+\\:{a}_{2}\\times\\:{NDBI}_{c}+\\:{a}_{3}\\times\\:{NDWI}_{c}\\:$$\u003c/div\u003e\n \u003cdiv\u003e1\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003ewhere the subscript c means the variable with coarser 30 m spatial resolution derived from Landsat-8/-9 satellite data and the variables a\u003csub\u003e0\u003c/sub\u003e\u0026thinsp;\u0026minus;\u0026thinsp;a\u003csub\u003e3\u003c/sub\u003e represent the output regression coefficients.\u003c/p\u003e\n\u003cp\u003eThen, these derived regression coefficients are applied to obtain the downscaled LST using selected Sentinel-2 NDVI, NDBI, and NDWI spectral indices with finer 10 m spatial resolution. The fundamental premise of this method is that the thermal bands have a direct relation with the spectral bands, and the relationship is scale invariable (Gao et al., 2012). The subsequent model was employed to downscale LST at finer resolution:\u003c/p\u003e\n\u003cdiv id=\"Equ2\"\u003e\n \u003cdiv id=\"FileID_Equ2\" name=\"EquationSource\"\u003e$$\\:{LST}_{f}={a}_{0}+\\:{a}_{1}\\times\\:{NDVI}_{f}+\\:{a}_{2}\\times\\:{NDBI}_{f}+\\:{a}_{3}\\times\\:{NDWI}_{f}\\:$$\u003c/div\u003e\n \u003cdiv\u003e2\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003ewhere the coefficients a\u003csub\u003e0\u003c/sub\u003e\u0026thinsp;\u0026minus;\u0026thinsp;a\u003csub\u003e3\u003c/sub\u003e of Eq.\u0026nbsp;(\u003cspan\u003e1\u003c/span\u003e) are applied and the subscript f means the variable (spectral index) with a finer 10 m spatial resolution derived from Sentinel-2 satellite imagery.\u003c/p\u003e\n\u003cp\u003eImage classification using ML algorithms\u003c/p\u003e\n\u003cp\u003eSVM and RF ML algorithms were used and compared for the image classification of multispectral imagery using the \u0026quot;Harmonized Sentinel-2 MSI level 2A\u0026quot; data product.\u003c/p\u003e\n\u003cp\u003eFirstly, three cloud-free median compositions were generated for the respective years 2017, 2019, and 2023. This process involved utilizing Sentinel-2 satellite scenes that met specific criteria: captured between June 1st and August 8th of each selected year with cloud cover below 20%. Subsequently, these composits were displayed as true color RGB compositions used to create five datasets of training samples, representing the five dominant classes/types of classified LC \u0026ndash; water bodies, forests, grasslands, agricultural lands, and built-up areas. which were systematically distributed across our study area. These training samples were generated separately for the years 2017, 2019, and 2023 to accurately capture the respective types of LC for each year. Subsequently, these training samples were used for image classification using the SVM and RF algorithms.\u003c/p\u003e\n\u003cp\u003eIn GEE, the SVM algorithm implemented under the \u0026quot;ee.Classifier.libsvm()\u0026quot; tool was applied. An overview of the parameters and their respective values used to train the SVM model is presented in Table\u0026nbsp;\u003cspan\u003e3\u003c/span\u003e. RF is implemented using the tool \u0026quot;ee.Classifier.smileRandomForest()\u0026quot; and uses 6 input parameters (Table\u0026nbsp;\u003cspan\u003e3\u003c/span\u003e). For both algorithms, to achieve higher classification accuracy, we also used the NDVI and NDBI spectral indices as input, similarly to the study of Svoboda et al. (\u003cspan\u003e2022\u003c/span\u003e). These indices were used to train the models to better distinguish between vegetation and built-up areas.\u003c/p\u003e\n\u003cp\u003eTo evaluate classification accuracy and determine a more reliable classifier, the accuracy assessment was performed separately for the 2017, 2019 and 2023 classification results and for the two classification algorithms used (SVM and RF). Training and testing data sets were established at an 80:20 ratio. The training data sets were used for model training, whereas the testing datasets were employed for validating the accuracy of the classifications.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 3\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eParameters used for image classification by SVM and RF algorithms in the GEE environment.\u003c/p\u003e\n \u003cdiv\u003e\n \u003cp\u003eSource: Google Earth Engine (\u003cspan\u003e2024\u003c/span\u003e), author\u0026apos;s processing.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"3\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAlgorithm\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eParameter\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eValue\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"4\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDecision Procedure\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDefault: Voting\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSVM type\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDefault: C_SVC (for non-linear data)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eKernel type\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLinear\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"6\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNumber of decision trees\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1500\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNumber of variables per split\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDefault: the square root of the number of variables\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMinimum leaf population\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDefault: 1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRandomization seed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDefault: 0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFraction of input to bag per tree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDefault: 0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMaximum number of leaf nodes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDefault: no limit\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e"},{"header":"Results and discussion","content":"\u003cp\u003eDownscaled LST using the relationship between spectral indices and LST\u003c/p\u003e \u003cp\u003eThe LST prediction model was established using ordinary least squares linear regression and three predictors\u0026thinsp;\u0026minus;\u0026thinsp;NDVI, NDWI, and NDBI. Scatter plots between the pairs of variables, Landsat-8 LST, and each predictor revealed that there is a significant correlation between the spectral indices and LST in Prešov city: a negative correlation with NDVI and positive correlations with NDBI and NDWI.\u003c/p\u003e \u003cp\u003eThe results show that there is a direct correlation between LST and NDBI (r\u003csub\u003e2017\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.99, r\u003csub\u003e2019\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;1.00, r\u003csub\u003e2023\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.50) which means that highly built-up areas have high recorded temperature values. This observation agrees with the multivariate analysis. An indirect relationship is observed between LST and NDVI (r\u003csub\u003e2017\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.73, r\u003csub\u003e2019\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.61, r\u003csub\u003e2023\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.85) and LST and NDWI (r\u003csub\u003e2017\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.72, r\u003csub\u003e2019\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.67, r\u003csub\u003e2023\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.60). Based on these results, it can be inferred that the correlation values suggest that NDBI is a good indicator for LST, which is aligned with the findings of Purio et al., (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e shows a comparison of downscaling results for one of those years (2023) - LST\u003csub\u003ec\u003c/sub\u003e in the original 30 m spatial resolution (before downscaling) and LST\u003csub\u003ef\u003c/sub\u003e after downscaling in 10 m spatial resolution with clearly improved (sharpened) LST pattern. The coefficient of determination (R\u003csup\u003e2\u003c/sup\u003e) is 0.92.\u003c/p\u003e \u003cp\u003eDifferences in the distribution of LST can be clearly linked to specific LC categories. From the LST patterns in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e it can be seen that the presence of the UHI phenomenon in the city is significant. The lowest LST values in the study area are spread across a wide forest park in the western part of the city of Prešov, with a significant percentage of vegetation cover and low human intervention in nature. Low temperature values are also found in the southern part of the area with agricultural fields and along the Torysa River valley. In contrast, the highest values, that reach sometimes reaching up to 50\u0026deg;C are visible in the historical city core, areas of other urban fabrics on the outskirts of the city core and recently built road communications and industrial zones in the south of the city core. Differences in LST patterns between selected images in 2017, 2019, and 2023 were predominantly observed in suburban areas affected by the construction of a highway bypass and the development of new buildings in the industrial parks of Haniska and Petrovany.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThese selected locations were further analysed using LST\u003csub\u003ef\u003c/sub\u003e. The location map is depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, with numbers 1\u0026ndash;4 marking 4 locations where significant changes in LST were detected within the area of interest.\u003c/p\u003e \u003cp\u003eThe corresponding changes in LSTs for these four locations are presented in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. Based on the maps in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e and the data presented in the Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, we observe that in areas where LC has changed due to the expansion of built-up areas and roads, there is a corresponding increase in LST values. Localities 1 and 2 indicate sites located on newly constructed sections of the highway and expressway, where LST increased approximately 6\u0026ndash;7\u0026deg;C in 2023 (after construction) compared to 2017 (before construction).\u003c/p\u003e \u003cp\u003eAn even greater difference in LSTs can be observed at sites 3 and 4, where new buildings have been constructed in the industrial areas of Haniska and Petrovany. At site 3, the LST value was 38\u0026deg;C in 2017 (before the construction of the production building), and by 2023, this value had increased by approximately 10\u0026deg;C to 48\u0026deg;C (after the construction of the production building). Similarly, at location 4, the LST increased by approximately 17\u0026deg;C after the construction of the logistics centre.\u003c/p\u003e \u003cp\u003eThe increase in LST values in these cases is attributed both to the alteration in LC and the nature of the surfaces themselves. Previously, these areas had relatively natural landscapes or were used for agriculture. By 2023, these areas had undergone development characterized by impermeable artificial materials, including asphalt roads (sites 1\u0026ndash;2) and buildings predominantly featuring metal roofs (sites 3\u0026ndash;4).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eLST values for selected 4 locations in 2017, 2019 and 2023.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLocality\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eChange in LC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c5\" namest=\"c3\"\u003e \u003cp\u003eLST (\u0026deg;C)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2017\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2019\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2023\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eroad communication - expressway\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e36.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e40.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e42.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eroad communication - highway\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e35.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e38.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e42.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eindustrial zone - production building\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e38.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e47.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e48.1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eindustrial zone - logistics center\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e35.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e36.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e51.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eImage classification using ML algorithms \u0026ndash; RF and SVM\u003c/p\u003e \u003cp\u003eThe second part of our research was dedicated to semi-automatic supervised classification of satellite imagery using two machine learning algorithms SVM and RF. Classification outputs using the SVM and RF algorithms are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e. The overall accuracies of both algorithms for respective years 2017, 2019 and 2023 are displayed in Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eOverall accuracy of image classifications using SVM and RF algorithms for 2017, 2019 and 2023.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eYear\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eOverall accuracy (in %)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e92.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e93.2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e92.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e89.6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2023\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e87.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e91.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eBased on the visual assessment in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, semiautomatic classification using SVM and RF machine learning algorithms demonstrated considerable success, although some instances of pixel misclassification were observed (e.g., built-up areas classified as water bodies, or agricultural land as built-up areas).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAccording to Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, the overall accuracy of the six classifications evaluated using the SVM and RF algorithms ranged from 87.2\u0026ndash;93.2%. Satellite imagery from 2017 was classified with an accuracy of 92.4% by the SVM model and 93.2% accuracy by the RF model. In 2019, the SVM model had a reliability of 92.2%, slightly less than 3% more than the RF model. By 2023, the SVM model's success rate had dropped to 87.2%, while the RF model's success rate was approximately 91.5%.\u003c/p\u003e \u003cp\u003eSeveral factors impacted the accuracy of image classification. Primarily, the spatial resolution of Sentinel-2 images played a significant role. The class of water bodies was frequently misclassified due to the difficulty in identifying the Torysa River and its tributaries in parts of the images. This challenge arose because their width is less than 10 metres, which is the native resolution of the Sentinel-2 bands. The second factor concerns the radiometric characteristics of satellite scenes, particularly when they are median composites generated from satellite images over several months. This can result in minor inaccuracies in classification, such as with agricultural soils, because the varying stages of vegetation of individual crops can make them harder to distinguish from grasslands.\u003c/p\u003e \u003cp\u003eBased on the results achieved, we evaluated that the RF algorithm appeared to be more reliable and therefore, in the further analysis, we exclusively used the classification results of this particular more reliable classifier (RF). Higher reliability of RF classification was also confirmed, for example, by studies by Chowdhury (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), Zafar et al. (\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSpatio-temporal evaluation of changes in LC and LST\u003c/p\u003e \u003cp\u003eSpatio-temporal analysis was conducted using derived downscaled LST outputs in 10 m spatial resolution and the results of supervised classification by the RF classifier for the years 2017, 2019 and 2023. Figure\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e shows four localities within the area of interest, represented by image classification by the RF algorithm and the corresponding LST, highlighting the most significant changes in LC. The LST values for these locations are presented in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eLocations 1 and 2 depict areas where the highway bypass and tunnel were constructed between 2017 and 2019, leading to an increase in LST. In 2017, location 1 consisted of grassed areas and part of a garden area, with an LST value of approximately 37\u0026deg;C. By 2019, during road construction, the LST increased to 40\u0026deg;C and by 2023, it had risen to nearly 43\u0026deg;C. A similar trend is observed at location 2, where the LST value increased from 35\u0026deg;C in 2017 to nearly 43\u0026deg;C in 2023. Locations 3 and 4, which were previously agricultural land, have been transformed into built-up areas. At location 3, LSTs were approximately 38\u0026deg;C in 2017. By 2019, the SPINEA, s.r.o. production building was constructed on this site, featuring a metal roof, metal facade, and concrete parking lot, resulting in an LST exceeding 48\u0026deg;C by 2023. A similar transformation occurred at location 4, where agricultural land was replaced by an industrial zone \u0026ndash; CTPark Prešov South, housing multiple companies. The entire area is paved with concrete surfaces and the buildings feature metal roofs and facades. This has led to an increase in LST values from 35\u0026deg;C in 2017 to nearly 52\u0026deg;C by 2023.\u003c/p\u003e \u003cp\u003eThe areas surrounding both sites are exclusively agricultural lands without trees. Vegetation consists mainly of shrubs near local streams, although these were not clearly detected by our image classifications due to the spatial resolution. However, pixels depicting water bodies in the actual landscape exhibit noticeably lower LSTs compared to the adjacent banks, which are lined with stones.\u003c/p\u003e \u003cp\u003eData validation\u003c/p\u003e \u003cp\u003eTo assess LSTs in the urban environment, validation data was collected using a data logger at five selected locations (A-E) within the urban area of Prešov. The locations and corresponding LST data are recorded in Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e and Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSelected localities in the Prešov city, listing their kinetic surface temperatures measured by the Comet data logger and comparing them to the corresponding LST pixel value \u0026minus;\u0026thinsp;30 m pixel from Landsat-8/-9-derived LST\u003csub\u003ec\u003c/sub\u003e and downscaled 10 m pixel of LST\u003csub\u003ef\u003c/sub\u003e.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLocality\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eCharacteristic\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c5\" namest=\"c3\"\u003e \u003cp\u003eLST -\u0026nbsp;data logger (\u0026deg;C)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLST\u003csub\u003ec\u003c/sub\u003e (\u0026deg;C)\u003c/p\u003e \u003cp\u003e09:26:16\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLST\u003csub\u003ef\u003c/sub\u003e (\u0026deg;C)\u003c/p\u003e \u003cp\u003e09:26:16\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e07:00\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10:00\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e13:00\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003etartan\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e27.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e32.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e35.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e34.1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eartificial grass\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e29.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e35.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e30.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e37.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e38.4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003estone paving\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e30.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e40.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e40.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e42.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003enatural grass\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e29.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e35.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e43.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e36.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eE\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003econcrete\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e31.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e39.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e35.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e37.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e45.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e shows that areas with higher LST values correspond to specific forms of urban fabric (continuous urban fabric in the city centre, discontinuous urban fabric such as blocks of flats, commercial buildings, roads, parking lots, etc.). On the contrary, pixels displaying lower LSTs indicate areas with present vegetation, which contributes to cooling the surroundings.\u003c/p\u003e \u003cp\u003eLocation A is a tartan-covered athletic track located on the grounds of the elementary school grounds. Minimum LST values here hover around 30\u0026deg;C, occurring in areas with dense tree vegetation. The maximum LST, reaching approximately 47\u0026deg;C, is observed in the southwest part of the area, attributed to a light sheet metal roof of a department store. Location B was located near a football field with artificial grass, adjacent to the Torysa River. Typically, water surfaces exhibit lower LST values, but here the river's pixels show higher values likely due to lower spatial resolution. Along the nearby banks of the Torysa River, minimum LST values of around 27\u0026deg;C were identified. This cooling effect is attributed to the widespread vegetation cover that includes grassy areas, meadows, and dense tree stands. In contrast, the maximum value of LST in this area was recorded near the swimming pool, where the surfaces are made of brick and stone pavements, reaching over 56\u0026deg;C.\u003c/p\u003e \u003cp\u003eLocation C, in the center of Prešov, shows notably high LST values above 50\u0026deg;C. The minimum LST here is 34\u0026deg;C, the highest among the studied locations, due to dense urban fabric with red tin roofs and impermeable surfaces like stone streets and squares. Vegetation, such as trees in parks and along the main square, is less distinguishable in LST pixel values due to the satellite's lower resolution, which mainly reflects built-up areas. Location D is located within the southern park area near the University of Prešov, not far from the Torysa River. In particular, the central and eastern parts are densely built areas, with maximum LST values exceeding 46\u0026deg;C. On the contrary, the western part of this section comprises forested areas where the minimum LST values hover around 31\u0026deg;C.\u003c/p\u003e \u003cp\u003eLocation E encompasses Prešov\u0026rsquo;s largest housing estate, Sekčov, featuring shopping centres with metal roofs and extensive concrete parking lots. These surface characteristics contribute to maximum LST values near 50\u0026deg;C. Contrasting this, the western area adjacent to the built-up zone shows a minimum LST of approximately 31\u0026deg;C, characterized by long-term grassy areas interspersed with bushes.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e shows that the LST values recorded by the handheld data logger (at 10:00 UTC) are consistently lower than the corresponding pixel values derived from satellite images (acquisition time: 09:26:16 UTC). This is primarily due to the higher accuracy and pointwise measurement capability of the data logger. A strong positive correlation of 0.98 was found between the data from the datalogger and the LST\u003csub\u003ef\u003c/sub\u003e and a correlation of 0.97 was found between the data from the datalogger and the LST\u003csub\u003ec\u003c/sub\u003e. In the morning at 7:00 a.m. under clear weather conditions, tartan recorded the lowest temperature at 27.8\u0026deg;C, followed by artificial grass at 29.6\u0026deg;C, and natural grass at 29.0\u0026deg;C. The stone paving reached 30.9\u0026deg;C, while the concrete had an LST of 31.3\u0026deg;C. At 10:00 a.m., with no change in cloud cover, these temperatures rose as follows: tartan increased by less than 5\u0026deg;C to 32.3\u0026deg;C, both natural and artificial grasses increased by 6.8\u0026deg;C to approximately 35.6\u0026deg;C and 35.8\u0026deg;C respectively. Stone paving and concrete showed the most significant increases in surface temperature, rising by almost 10\u0026deg;C to around 40\u0026deg;C.\u003c/p\u003e \u003cp\u003eThe surface LST values obtained from the data logger were compared with those derived from satellites at 30 m spatial resolution (before downscaling) and 10 m resolution (after downscaling) from Landsat-8 data acquired at 9:26:16 UTC. Even in the satellite-derived LST data, tartan showed the lowest value at 34.1\u0026deg;C, followed by natural grass at approximately 36.9\u0026deg;C, artificial grass at 38.4\u0026deg;C, stone paving at 42.7\u0026deg;C, and concrete that recorded the highest temperature at 45.0\u0026deg;C.\u003c/p\u003e \u003cp\u003eDespite differences between the LST values obtained from the data logger and satellite data, there are similarities in the overall LST patterns observed with both methods. Both data sets show a consistent trend from the coldest to the warmest surface types, although the order of the LST data for concrete and stone paving changed after downscaling.\u003c/p\u003e \u003cp\u003eOur findings indicate that man-made surfaces, such as concrete and stone pavements, experience rapid heating, up to 10\u0026deg;C in 3 hours under sunny conditions. These surfaces are prevalent in urban areas, contributing to elevated LSTs and thus heating their surroundings. On the contrary, natural surfaces, such as vegetation, exhibit lower LSTs, which help cool their environment. However, certain parts of natural grassy areas occasionally show slightly higher LST values, often due to damage such as exposed soils, which do not fully represent healthy grass samples (see Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD, Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e). Considering the prevalence of impermeable surfaces like concrete, asphalt, and stone in urban areas, the potential for the UHI phenomenon in the city is evident.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThe effect of UHI and the increase in LST are concerns not only for large cities but also for smaller cities like Prešov in Slovakia. The main objective of this research was to analyze the spatial distribution and temporal changes of LSTs in the city of Prešov.\u003c/p\u003e \u003cp\u003eThis analysis used a downscaling technique, which integrated satellite observations from the Landsat-8, Landsat-9, and Sentinel-2 missions. Despite not having the thermal band, Sentinel-2 provides higher sptial resolution for estimating the thermal emissivity of LC for LST calculation from Landsat-8/-9 thermal data. A significant correlation was discovered between spectral indices (NDBI, NDVI, NDWI) and LST that was leveraged to downscale the native LST\u003csub\u003ec\u003c/sub\u003e with a 30 m spatial resolution to a finer LST\u003csub\u003ef\u003c/sub\u003e with a 10 m resolution, using a multiple linear regression model. This improved resolution allowed a better identification of changes in LST during the years 2017, 2019, and 2023.\u003c/p\u003e \u003cp\u003eFurthermore, two machine learning algorithms, SVM and RF, were employed to classify images and assess the influence of varying LC types and changes throughout the years 2017, 2019, and 2023 on the city's LST patterns. Evaluation of LC changes using the RF classification algorithm revealed higher accuracy compared to SVM - RF outperformed SVM by 0.8% in 2017 and 4.3% in 2023. The results show that over recent years, a previously natural landscape has been extensively altered, primarily in suburban areas that cause the increase of LSTs, thus influencing the emergence and intensification of the UHI phenomenon. These changes were mainly driven by the construction of a highway bypass and the expansion of industrial zones. It was observed that impervious man-made surfaces, such as concrete and stone pavements, experience rapid heating, up to 10\u0026deg;C in 3 hours, subsequently warming the surrounding areas.\u003c/p\u003e \u003cp\u003eTherefore, strategic urban planning and city development are essential to promote sustainable expansion and protect urban greenery, ultimately improving the quality of life of city residents.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors are very grateful for the detailed and constructive comments to improve the manuscript provided by the editors and anonymous reviewers.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contribution\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAnton Uhrin: Conceptualization, Methodology, Investigation, Validation, Formal analysis, Resources, Writing\u0026mdash;original draft preparation. Katar\u0026iacute;na Onačillov\u0026aacute;: Conceptualization, Methodology, Visualization, Review and Editing, Funding.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSome or all data, models, or codes that support the findings of this study are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was funded by the Slovak Research and Development Agency (APVV) under the contract No. APVV-23-0210 and by the Scientific Grant Agency of the Ministry of Education, Science, Research and Sport of the Slovak Republic and the Slovak Academy of Sciences (VEGA) under the contract No. VEGA 1/0085/23.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical Approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors have read, understood, and have complied as applicable with the statement on \u0026ldquo;Ethical responsibilities of Authors\u0026rdquo; as found in the Instructions for Authors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBONAFONI, S., ANNIBALLE, R., GIOLI, B. \u0026amp; TOSCANO, P. (2016). Downscaling Landsat Land Surface Temperature over the urban area of Florence. \u003cem\u003eEuropean Journal of Remote Sensing\u003c/em\u003e, 49, 553\u0026ndash;569. https://doi.org/10.5721/EuJRS20164929.\u003c/li\u003e\n\u003cli\u003eCAMPS, A., PARK, H., CASTELLVI, J., CORBERA, J. \u0026amp; ASCASO, E. (2020). Single-Pass Soil Moisture Retrievals Using GNSS-R: Lessons Learned. \u003cem\u003eRemote Sensing\u003c/em\u003e, 12, 2064. https://doi.org/10.3390/rs12122064.\u003c/li\u003e\n\u003cli\u003eCHEVAL, S., DUMITRESCU, A. \u0026amp; PETRIŞOR, A.-I. (2011). The July Surface Temperature Lapse In The Romanian Carpathians. \u003cem\u003eCarpathian Journal of Earth and Environmental Sciences\u003c/em\u003e, 6(1), 189\u0026ndash;198.\u003c/li\u003e\n\u003cli\u003eCHOWDHURY, M. S. 2024. Comparison of accuracy and reliability of random forest, support vector machine, artificial neural network and maximum likelihood method in land use/cover classification of urban setting. \u003cem\u003eEnvironmental Challenges\u003c/em\u003e, 14, 100800. https://doi.org/10.1016/j.envc.2023.100800.\u003c/li\u003e\n\u003cli\u003eCOPERNICUS LAND MONITORING SERVICE (2018). Urban Atlas 2018. Retrieved June 10, 2017, from http://land.copernicus.eu/local/urban-atlas/urban-atlas-2012.\u003c/li\u003e\n\u003cli\u003eDOBROVOLN\u0026Yacute;, P. (2013). The surface urban heat island in the city of Brno (Czech Republic) derived from land surface temperatures and selected reasons for its spatial variability. \u003cem\u003eTheoretical Applied Climatology\u003c/em\u003e, 112, 89\u0026ndash;98. https://doi.org/10.1007/s00704-012-0717-8.\u003c/li\u003e\n\u003cli\u003eFADHIL, M., HAMOODI, M. N. \u0026amp; ZIBOON, A. R. T. (2023). Mitigating urban heat island effects in urban environments: strategies and tools. \u003cem\u003eIOP Conference Series: Earth and Environmental Science\u003c/em\u003e, 1129, 012025. https://doi.org/10.1088/1755-1315/1129/1/012025.\u003c/li\u003e\n\u003cli\u003eGAO, F., KUSTAS, W. P. \u0026amp; ANDERSON, M. C. (2012). A data mining approach for sharpening thermal satellite imagery over land. \u003cem\u003eRemote Sensing\u003c/em\u003e, 4, 3287\u0026ndash;3319. https://doi.org/10.3390/rs4113287.\u003c/li\u003e\n\u003cli\u003eGASCON, F., BOUZINAC, C., TH\u0026Eacute;PAUT, O., JUNG, M., FRANCESCONI, B., LOUIS, J., LONJOU, V., LAFRANCE, B., MASSERA, S. \u0026amp; GAUDEL-VACARESSE, A. (2017). Copernicus Sentinel-2A Calibration and Products Validation Status. \u003cem\u003eRemote Sensing\u003c/em\u003e, 9(6), 584. https://doi.org/10.3390/rs9060584.\u003c/li\u003e\n\u003cli\u003eGOLDBLATT, R., STUHLMACHER, M. F., TELLMAN, B., CLINTON, N., HANSON, G., GEORGESCU, M., WANG, CH., SERRANO-CANDELA, F., KHANDELWAL, A. K., CHENG, W. H. \u0026amp; BALLING, R. C. (2018). Using Landsat and nighttime lights for supervised pixel-based image classification of urban land cover. \u003cem\u003eRemote Sensing of Environment\u003c/em\u003e, 205, 253\u0026ndash;275. https://doi.org/10.1016/j.rse.2017.11.026.\u003c/li\u003e\n\u003cli\u003eGOOGLE EARTH ENGINE (2024). \u003cem\u003eRF Classifier.\u003c/em\u003e Retrieved March, 2024, from https://developers.google.com/earth-engine/apidocs/ee-classifier-smilerandomforest. Last access: 13 March 2024.\u003c/li\u003e\n\u003cli\u003eGRIGORAȘ, G. \u0026amp; URIȚESCU, B. (2019). Land Use/Land Cover changes dynamics and their effects on Surface Urban Heat Island in Bucharest, Romania. \u003cem\u003eInternational Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e, 80, 115\u0026ndash;126. https://doi.org/10.1016/j.jag.2019.03.009.\u003c/li\u003e\n\u003cli\u003eGUO, X., WANG, M., JIA, M. \u0026amp; WANG, W. (2021). Estimating mangrove leaf area index based on red-edge vegetation indices: A comparison among UAV, WorldView-2 and Sentinel-2 imagery. \u003cem\u003eInternational Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e, 103, 102493. https://doi.org/10.1016/j.jag.2021.102493.\u003c/li\u003e\n\u003cli\u003eHAN, L., LU, L., FU, P., REN, C., CAI, M. \u0026amp; LI, Q. (2023). Exploring the seasonality of surface urban heat islands using enhanced land surface temperature in a semi-arid city. \u003cem\u003eUrban Climate\u003c/em\u003e, 49, 101455. DOI: https://doi.org/10.1016/j.uclim.2023.101455.\u003c/li\u003e\n\u003cli\u003eHU, D., GUO, F., MENG, Q., SCHLINK, U., WANG, S., HERTEL, D. \u0026amp; GAO, J. (2024). A novel dual-layer composite framework for downscaling urban land surface temperature coupled with spatial autocorrelation and spatial heterogeneity\u003cem\u003e. International Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e, 130, 103900. https://doi.org/10.1016/j.jag.2024.103900.\u003c/li\u003e\n\u003cli\u003eHUTENGS, C., VOHLAND, M. (2016). Downscaling land surface temperatures at regional scales with random forest regression. \u003cem\u003eRemote Sensing of Environment\u003c/em\u003e, 178, 127\u0026ndash;141. https://doi.org/10.1016/j.rse.2016.03.006.\u003c/li\u003e\n\u003cli\u003eIVAJN\u0026Scaron;IČ, D. \u0026amp; ŽIBERNA, I. (2018). The effect of weather patterns on winter small city urban heat islands. \u003cem\u003eMeteorological Applications\u003c/em\u003e, 2018, 1\u0026ndash;9. https://doi.org/10.1002/met.1752.\u003c/li\u003e\n\u003cli\u003eKHAN, M., QASIM, M., TAHIR, A. A. \u0026amp; FAROOQI, A. (2023). Machine learning-based assessment and simulation of land use modification effects on seasonal and annual land surface temperature variations. \u003cem\u003eHeliyon\u003c/em\u003e, 9(12), e23043. https://doi.org/10.1016/j.heliyon.2023.e23043.\u003c/li\u003e\n\u003cli\u003eKOTTEK, M., GRIESER, J., BECK, C., RUDOLF, B. \u0026amp; RUBEL, F. (2006). World Map of the K\u0026ouml;ppen-Geiger Climate Classification Updated. \u003cem\u003eMeteorologishe Zeitschrift\u003c/em\u003e, 15, 259\u0026ndash;263. https://doi.org/10.1127/0941-2948/2006/0130.\u003c/li\u003e\n\u003cli\u003eLONGLEY, I. D. GALLAGHER, M.W., DORSEY, J. R., FLYNN, M. \u0026amp; BARLOW, J. F. (2004). Short-term measurements of airflowand turbulence in two street canyons in Manchester. \u003cem\u003eAtmospheric Environment\u003c/em\u003e, 38, 69\u0026ndash;79. https://doi.org/10.1016/j.atmosenv.2003.09.060.\u003c/li\u003e\n\u003cli\u003eMCFEETERS, S. K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. \u003cem\u003eInternational Journal of Remote Sensing\u003c/em\u003e, 17, 1425\u0026ndash;1432. http://dx.doi.org/10.1080/01431169608948714.\u003c/li\u003e\n\u003cli\u003eMILES, V., ESAU, I. \u0026amp; MILES, M. W. (2023). The urban climate of the largest cities of the European Arctic. \u003cem\u003eUrban Climate\u003c/em\u003e, 48, 1\u0026ndash;12. https://doi.org/10.1016/j.uclim.2023.101423.\u003c/li\u003e\n\u003cli\u003eONAČILLOV\u0026Aacute;, K., GALLAY, M., PALUBA, D., P\u0026Eacute;LIOV\u0026Aacute;, A., TOKARČ\u0026Iacute;K, O. \u0026amp; LAUBERTOV\u0026Aacute;, D. (2022). Combining Landsat 8 and Sentinel-2 data in Google Earth Engine to derive higher resolution land surface temperature maps in urban environment. \u003cem\u003eRemote Sensing\u003c/em\u003e, 14(16), 4076. https://doi.org/10.3390/rs14164076.\u003c/li\u003e\n\u003cli\u003ePATIL, R. \u0026amp; SURAWAR, M. (2023). Impact of Urban Heat Island on Formation of Precipitation in Indian Western Coastal Cities. \u003cem\u003eJournal of Contemporary Urban Affairs\u003c/em\u003e, 7(2), 37\u0026ndash;55. https://doi.org/10.25034/ijcua.2023.v7n2-3.\u003c/li\u003e\n\u003cli\u003ePUREVDORJ, T., TATEISHI, R., ISHIYAMA, T. \u0026amp; HONDA, Y. (1998). Relationships between percent vegetation cover and vegetation indices. \u003cem\u003eInternational Journal of Remote Sensing\u003c/em\u003e, 19, 3519\u0026ndash;3535. https://doi.org/10.1080/014311698213795.\u003c/li\u003e\n\u003cli\u003ePURIO, M. A. C., YOSHITAKE, T. \u0026amp; CHO, M. (2022). Assessment of Intra-Urban heat Island in a densely populated city using Remote sensing: A case study for Manila City. \u003cem\u003eRemote Sensing\u003c/em\u003e, 14(21), 5573. https://doi.org/10.3390/rs14215573.\u003c/li\u003e\n\u003cli\u003eSHAMSAEI, M., CARTER, A. \u0026amp; VAILLANCOURT, M. (2022). A review on the heat transfer in asphalt pavements and urban heat island mitigation methods. \u003cem\u003eConstruction and Building Materials\u003c/em\u003e, 359, 129350. https://doi.org/10.1016/j.conbuildmat.2022.129350.\u003c/li\u003e\n\u003cli\u003eSONG, B. \u0026amp; PARK, K. (2020). Verification of Accuracy of Unmanned Aerial Vehicle (UAV) Land Surface Temperature Images Using In-Situ Data. \u003cem\u003eRemote Sensing\u003c/em\u003e, 12(2), 288. https://doi.org/10.3390/s12020288.\u003c/li\u003e\n\u003cli\u003eSTATISTICAL OFFICE SR (2022). \u003cem\u003ePublic database of the Slovak Statistical Office of the Slovak Republic\u003c/em\u003e. Retrieved February 8, 2024, from https://slovak.statistics.sk. Last access: 8 February 2024.\u003c/li\u003e\n\u003cli\u003eSTEENSEN, B. M., MARELLE, L., HODNEBROG, \u0026Oslash;. ET AL. (2022). Future urban heat island influence on precipitation. \u003cem\u003eClimate Dynamics\u003c/em\u003e, 58, 3393\u0026ndash;3403. https://doi.org/10.1007/s00382-021-06105-z.\u003c/li\u003e\n\u003cli\u003eSVOBODA, J., \u0026Scaron;TYCH, P., LA\u0026Scaron;TOVIČKA, J., PALUBA, D. \u0026amp; KOBLIUK, N. (2022). Random Forest Classification of Land Use, Land-Use Change and Forestry (LULUCF) UsingSentinel-2 Data \u0026ndash; A Case Study of Czechia. \u003cem\u003eRemote Sensing\u003c/em\u003e, 14, 1189. https://doi.org/10.3390/rs14051189.\u003c/li\u003e\n\u003cli\u003eTANOORI, G., SOLTANI, A. \u0026amp; MODIRI, A. (2024). Machine Learning for Urban Heat Island (UHI) analysis: Predicting Land Surface Temperature (LST) in Urban Environments. \u003cem\u003eUrban Climate\u003c/em\u003e, 55, 101962. https://doi.org/10.1016/j.uclim.2024.101962.\u003c/li\u003e\n\u003cli\u003eUSGS (2024). \u003cem\u003eLandsat 8-9 Collection 2 (C2) Level 2 Science Product (L2SP) Guide. \u003c/em\u003eRetrieved May 9, 2024, from\u003cem\u003e \u003c/em\u003ehttps://www.usgs.gov/media/files/landsat-8-9-coll-ection-2-level-2-science-product-guide.\u003c/li\u003e\n\u003cli\u003eVARDOULAKIS, E., KARAMANIS, D., FOTIADI, A. \u0026amp; MIHALAKAKOU, G.,2013. The urban heat island effect in a small Mediterranean city of high summer temperatures and cooling energy demands. \u003cem\u003eSolar Energy\u003c/em\u003e, 94, 128\u0026ndash;144. https://doi.org/ 10.1016/j.solener.2013.04.016.\u003c/li\u003e\n\u003cli\u003eWEILAND, L., ROGERS, C. A., SOTHE, C., ARAIN, M. A. \u0026amp; GONSAMO, A. (2023). Satellite-based land surface temperature and soil moisture observations accurately predict soil respiration in temperate deciduous and coniferous forests. \u003cem\u003eAgricultural and Forest Meteorology\u003c/em\u003e, 340, 109618. https://doi.org/10.1016/j.agrformet.2023.109618.\u003c/li\u003e\n\u003cli\u003eXIONG, Y., CHEN, X., TANG, L. \u0026amp; WANG, H. (2022). Comparison of surface renewal and Bowen ratio derived evapotranspiration measurements in an arid vineyard. \u003cem\u003eJournal of Hydrology\u003c/em\u003e, 613, 128474. https://doi.org/10.1016/j.jhydrol.2022.128474.\u003c/li\u003e\n\u003cli\u003eXU, S., ZHAO, Q., YIN, K., HE, G., ZHANG, Z., WANG, G., WEN, M. \u0026amp; ZHANG, N. (2021). Spatial downscaling of land surface temperature based on a Multi-Factor geographically weighted machine learning model. \u003cem\u003eRemote Sensing\u003c/em\u003e, 13(6), 1186. https://doi.org/10.3390/rs13061186.\u003c/li\u003e\n\u003cli\u003eYAN, W., JIANG, J., HE, L., ZHAO, W., NAIR, R., WANG, X. \u0026amp; XIONG, Y. (2024). Correcting land surface temperature from thermal imager by considering heterogeneous emissivity. \u003cem\u003eInternational Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e, 129, 103824. https://doi.org/10.1016/j.jag.2024.103824.\u003c/li\u003e\n\u003cli\u003eZAFAR, Z., ZUBAIR, M., ZHA, Y., FAHD, S. \u0026amp; NADEEM, A. A. (2024). Performance assessment of machine learning algorithms for mapping of land use/land cover using remote sensing data. \u003cem\u003eThe Egyptian Journal of Remote Sensing and Space Sciences\u003c/em\u003e, 27(2), 216\u0026ndash;226. https://doi.org/10.1016/j.ejrs.2024.03.003.\u003c/li\u003e\n\u003cli\u003eZHA, Y., GAO, J. \u0026amp; NI, S. (2003). Use of normalized difference built-up index in automatically mapping urban areas from TM imagery. \u003cem\u003eInternational Journal of Remote Sensing\u003c/em\u003e, 24, 583\u0026ndash;594. https://doi.org/10.1080/01431160304987.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"environmental-monitoring-and-assessment","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"emas","sideBox":"Learn more about [Environmental Monitoring and Assessment](http://link.springer.com/journal/10661)","snPcode":"10661","submissionUrl":"https://submission.nature.com/new-submission/10661/3","title":"Environmental Monitoring and Assessment","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"land surface temperature, land cover, downscaling, urban heat island, machine learning classifiers","lastPublishedDoi":"10.21203/rs.3.rs-5143836/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5143836/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIn recent decades, global climate change and rapid urbanisation have aggravated the urban heat island (UHI) effect, affecting the well-being of urban citizens. Although this significant phenomenon is more pronounced in larger metropolitan areas due to extensive impervious surfaces, small and medium-sized cities also experience UHI effects, yet research on UHI in these cities is rare, emphasizing the importance of Land Surface Temperature (LST) as a key parameter for studying UHI dynamics. Therefore, this paper focusses on the evaluation of LST and land cover (LC) changes in the city of Prešov, Slovakia, a typical medium-sized European city that has recently undergone significant LC changes. In this study, we use the relationship between Landsat-8/-9 derived LST and spectral indices NDBI, NDVI, NDWI derived from Landsat-8/-9 and Sentinel-2 to downscale LST to 10 m. Two machine learning (ML) algorithms, Support Vector Machine (SVM) and Random Forest (RF), are used to assess image classification and identify how different types and LC changes in selected years 2017, 2019, 2023 affect the pattern of LST. The results show that several decisions made during the last decade, such as the construction of new urban fabrics and roads, caused the increase of LST. The LC change evaluation, based on the RF classification algorithm, achieved overall accuracies of 93.2% in 2017, 89.6% in 2019, and 91.5% in 2023, outperforming SVM by 0.8% in 2017 and 4.3% in 2023. This approach identifies UHI-prone areas with higher spatial resolution, helping urban planning mitigate the negative effects of increasing urban LSTs.\u003c/p\u003e","manuscriptTitle":"Spatio-temporal analysis of land surface temperature and land cover changes in Prešov city using downscaling approach and machine learning algorithms","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-09 13:15:57","doi":"10.21203/rs.3.rs-5143836/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-10-01T23:34:11+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-09-30T06:14:15+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-09-30T06:12:33+00:00","index":"","fulltext":""},{"type":"submitted","content":"Environmental Monitoring and Assessment","date":"2024-09-24T09:25:18+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"environmental-monitoring-and-assessment","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"emas","sideBox":"Learn more about [Environmental Monitoring and Assessment](http://link.springer.com/journal/10661)","snPcode":"10661","submissionUrl":"https://submission.nature.com/new-submission/10661/3","title":"Environmental Monitoring and Assessment","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"798c14bb-1db3-4c99-9277-2f2f7ee538aa","owner":[],"postedDate":"December 9th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-01-06T16:00:29+00:00","versionOfRecord":{"articleIdentity":"rs-5143836","link":"https://doi.org/10.1007/s10661-024-13598-8","journal":{"identity":"environmental-monitoring-and-assessment","isVorOnly":false,"title":"Environmental Monitoring and Assessment"},"publishedOn":"2025-01-03 15:57:14","publishedOnDateReadable":"January 3rd, 2025"},"versionCreatedAt":"2024-12-09 13:15:57","video":"","vorDoi":"10.1007/s10661-024-13598-8","vorDoiUrl":"https://doi.org/10.1007/s10661-024-13598-8","workflowStages":[]},"version":"v1","identity":"rs-5143836","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5143836","identity":"rs-5143836","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}