Company or Institution
National Observatory of Athens
Sustainable Development Goals (SDGs)
SDG 11: Sustainable Cities and Communities
General description of the AI solution
Volcanic eruptions are typically preceded by volcanic unrest events, expressed via ground deformation uplift. Such uplift can be captured via geodetic measuring systems. However, installing and maintaining ground based sensing and geodetic measurement systems in all active volcanoes is not feasible or cost-effective.
The Pluto project will put into production the first global volcanic activity early warning system. It builds on top of our previous research prototype system that exploits freely available global coverage Sentinel-1 Interferometric Synthetic Aperture Radar (InSAR) satellite data, coupled with artificial intelligence and computer vision algorithmic tools, to identify with high accuracy ground deformation patterns caused by volcanic unrest activity, providing warnings even months before an eruption.
Our proposed system offers a consistent, free-of-charge, volcano monitoring solution for the entire globe. Pluto is a natural successor of our Hephaestus dataset, the first large-scale InSAR dataset revolving around volcanic activity. It contains detailed annotations for the most active volcanoes of the world, including a ground deformation segmentation mask, the type of the deformation and the state of the volcano (i.e unrest, rest or rebound). As with all problems revolving around natural hazards, our dataset suffers from major class imbalance. To overcome this problem we exploit the vast amount of freely available, unlabeled, InSAR data to train our model in a self-supervised fashion, as these are more robust to class imbalance when compared to supervised learning. Even more, by utilizing a vast and diverse unlabeled dataset to extract knowledge, we are able to exploit areas and periods of time that do not exist in Hephaestus for representation learning. This leads to the creation of generic, task-agnostic encoders with high robustness to spatio-temporal distribution shifts. We then fine-tune these encoders to solve the downstream task of volcanic activity detection.
Github, open data repository, prototype or working demo
 Bountos et al. "Hephaestus: A large scale multitask dataset towards InSAR understanding." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022.
 Bountos et al. "Self-supervised contrastive learning for volcanic unrest detection." IEEE Geoscience and Remote Sensing Letters 19 (2021): 1-5.
 Bountos et al. "Learning from Synthetic InSAR with Vision Transformers: The case of volcanic unrest detection." IEEE Transactions on Geoscience and Remote Sensing (2022).
HPC resources and/or Cloud Computing Services