Introduction: The increasing speed of urbanization has led to significant changes in land cover and land use patterns, resulting in alterations in the quantity and spatial distribution of land surface temperature and the formation of heat islands in urban environments. For this reason, monitoring the spatiotemporal changes of heat islands in urban areas can be important for controlling and reducing the effects of this phenomenon in cities, especially large ones. In order to determine the location and changes of heat islands, calculating the land surface temperature is essential. Considering the diversity of land cover in urban areas and the sensitivity of land surface temperature to changes in land cover, there is a need for high spatial resolution data. However, thermal images usually do not have high spatial resolution. Additionally, some thermal images, such as Landsat images, which have suitable spatial resolution, lack the appropriate temporal resolution for continuous monitoring of surface temperature changes. Therefore, it is not possible to regularly monitor temporal changes in surface temperature using these images. In this study, the spatiotemporal trend of this phenomenon in areas 3, 6, and 11 of Tehran Municipality was examined in all seasons of the years 1393 and 1401. The low spatial resolution of the thermal bands of the available and accessible sensors, due to the complexity of the impact of various urban factors on land surface temperature, limits the ability to examine the temperature distinctions in complex and heterogeneous urban areas. Since a better examination of land surface temperature in cities requires thermal data with precise spatial information, in this study, the spatial resolution of the land surface temperature map retrieved from the Landsat 8 sensor has been improved from 100 meters to 30 meters using a statistical method.

Materials and methods: In this study, to improve the spatial resolution of the land surface temperature map retrieved from the Landsat 8 data from 100 meters to 30 meters, land surface characteristic indices retrieved from images of the same sensor, as well as auxiliary data including a digital elevation map and road maps, were used as predictor variables. In this context, and to find the best method with the highest accuracy, machine learning methods, including algorithms such as Random Forest, Support Vector Regression, and Multi-Layer Perceptron Neural Network, were used, and the results were compared with the TsHARP method as a conventional algorithm.

Results and discussion: The results in all images indicate the superiority of the random forest algorithm over all other methods. The best model accuracy was related to the winter season of 2014, with a root mean square error of 0.38 and a coefficient of determination of 0.94. The order of importance of the input predictor variables for this algorithm has varied in different seasons of each year. It has also shown differences in fixed seasons over the years. The examination of the spatiotemporal trends in the intensity and extent of heat islands indicates an increasing trend in temperature classes corresponding to this phenomenon in all seasons of the studied years. The extent of these increasing changes in terms of area was the highest in spring, measuring 68.3 square kilometers, and the lowest in winter, measuring 1.99 square kilometers. Also, in terms of intensity, it was highest in the summer and lowest in the winter.

Conclusion: The trend of changes in urban heat island intensity has always been increasing. The level of this index has been highest in the summer and lowest in the winter throughout all the years. The trend of changes in the area of temperature layers corresponding to heat islands in the region over the years studied has always been increasing, with varying degrees in different seasons. The spatial location of heat islands in these areas has remained constant throughout all seasons and years, and the heat islands with greater extent and intensity correspond to land uses such as educational, recreational, and medical areas; government offices; and heavily trafficked highways.