Effective Communicator Award

Congratulations to our group member, Carolina Natel, who - together with her teammates -received the Effective Communicator Award at the ELLIS Summer School: AI for Earth and Climate Sciences for their project, “Self-supervised Learning for Flood Mapping.”

Carolina and four other participants collaborated on a research challenge at the intersection of deep learning and environmental science. They proposed the JENA-Autoencoder, an adaptation of a state-of-the-art self-supervised learning method, which incorporates both spatial and temporal context from a triplet of Sentinel-1 images during pre-training.

Self-supervised learning is a powerful approach for training DL models when large amounts of unlabelled data (e.g., remote sensing imagery) are available, but labelled data are scarce. In this framework, a model is first trained on a pretext task, such as image reconstruction, and is then fine-tuned on labelled data for a downstream task like flood mapping segmentation.