HLR-MUSTANG: A dataset of High and Low Resolution Multimodal Urban land Surface Temperature for Artificial Neural network traininG

Urban areas are experiencing increasing heat exposure as a result of climate change and ongoing urbanization, creating a strong demand for accurate, high-resolution monitoring of Land Surface Temperature (LST). Upcoming satellite missions such as TRISHNA and LSTM will provide global LST observations at approximately 60 m spatial resolution; however, many urban-scale applications require substantially finer detail.<br><br>

This level of spatial detail can be achieved using super-resolution approaches that combine coarse-resolution LST observations with complementary high-resolution information describing the observed urban environment. Deep learning methods are particularly well suited for this task but require large, diverse, and well-curated training datasets.

This repository presents HLR-MUSTANG, a dataset of High- and Low-Resolution Multimodal Urban Land Surface Temperature for Artificial Neural Network Training. The dataset is designed to support LST super-resolution from 60 m to 10 m spatial resolution by integrating thermal, spectral, morphological, and three-dimensional information. HLR-MUSTANG combines airborne thermal infrared observations from multiple campaigns (DESIREX, AI4GEO/CAMCATT, HyTES) with Sentinel-2 multispectral imagery, land-cover maps, and digital height models across a range of European urban environments.