Multi-Hazard Mapping using Earth Observation data and Multi-Task Learning

This dataset accompanies the study "Multi-Hazard Mapping using Earth Observation data and Multi-Task Learning". The dataset provides examples of harmonized reference data combining water body and burned area datasets for training and evaluating multi-task deep learning models based on Sentinel-2 and Landsat-8 imagery. Contents 'experiments/` - folder containing example experiments conducted within the study `experiments/mtl/` - experiments conducted in multi-task setup `experiments/stl/` - experiments conducted in single-task setup `reference_data` - example reference patches Source Data All source data used in this study are publicly available: Global S1S2-Water Benchmark Dataset from Wieland et al. (2024) https://zenodo.org/records/11278238 Burned Area Reference Database (BARD) from Copernicus Climate Change Service https://edatos.consorciomadrono.es/dataset.xhtml?persistentId=doi:10.21950/BBQQU7 Copernicus Emergency Management Service (EMS) rapid mapping products https://zenodo.org/records/6597139 HLS burn scars dataset https://huggingface.co/datasets/ibm-nasa-geospatial/hls_burn_scars Sentinel-2 and Landsat-8 Collection 2 Level-2 Surface Reflectance via Microsoft Planetary Computer https://planetarycomputer.microsoft.com/ License Creative Commons Attribution 4.0 International (CC BY 4.0) Citation Please cite the associated publication and this dataset when using the data. [Publication reference to be added upon acceptance]