How Artificial Intelligence and Earth Observation Satellites are re-shaping volcano monitoring

A growing number of satellite missions offer data at various spatial resolutions and revisit times.There are plans for new missions to maintain near-continuous monitoring of volcanic activity globally. Artificial Intelligence (AI) techniques, particularly Machine Learning (ML) and Deep Learning (DL) models, offer distinct advantages in extracting information and knowledge from these vast datasets. Here, the potential of Artificial Intelligence techniques, satellite data, and cloud computing in addressing some of the most critical challenges and questions related to volcanic hazards is shown. A description of how the data-driven science paradigm is used to solve volcanic hazard problems is provided highlighting how earth observation satellite data are able to drive the entire estimation process through advanced AI techniques. An overview of the most important concepts and techniques to assist in interpreting satellite data for volcano monitoring is provided. From feature engineering methods enhancing the input signal for AI models, to convolution filters that can strategically be used in Convolutional Neural Network (CNN) architectures to find patterns, to the concept of “attention” in neural networks and the powerful abilities it brings to briefly discuss strategies from unsupervised, self-supervised and transfer learning to reduce the need for large labeled datasets. The objective of this work is dual, providing the basic concepts of EarthObservation (EO) and AI and showing how they have re-shaped volcano monitoring.