Super Resolution of Satellite-Based Land Surface Temperature Through Airborne Thermal Imaging

Urban heat island pose a significant threat to public health and urban livability. UHI maps are created using satellite thermal data, a crucial source for earth monitoring and for delivering mitigation strategies. Nowadays there is still a resolution gap between high-resolution optical data and low-resolution satellite thermal imagery. This study introduces a novel deep learning approach—named Dilated Spatio-Temporal U-Net (DST-UNet)—to bridge this gap. DST-UNET is a modified U-Net architecture which incorporates dilated convolutions to address the multiscale nature of urban thermal patterns. The model is trained to generate high-resolution, airborne-like thermal maps from available, low-resolution satellite imagery and ancillary data. Our results demonstrate that the DST-UNet can effectively generalise across different urban environments, enabling municipalities to generate detailed thermal maps with a frequency far exceeding that of traditional airborne campaigns. This framework leverages open-source data from missions like Landsat to provide a cost-effective and scalable solution for continuous, high-resolution urban thermal monitoring, empowering more effective climate resilience and public health initiatives.