Urban heat islands, characterized by localized areas with significantly elevated temperatures compared to their surroundings, present critical challenges to both the environment and public health. In this research, an innovative machine-learning model is proposed, integrating data from the Sentinel-3 satellite, weather forecasts, and other remote sensing sources. The primary objective is to predict high-resolution spatiotem-poral maps of maximum temperatures during a day in Turin. Experiments demonstrate the model's efficacy in forecasting temperature levels, reaching a MAE over 2023 of 2.09°C at a resolution of 20 meters/pixel, thus contributing valuable insights to the understanding of urban climate dynamics. This research advances comprehension of urban microclimates, underscores the significance of interdisciplinary data fusion, and provides a foundation for evidence-based policy decisions aimed at mitigating the adverse effects of extreme temperatures in urban environments.