Monitoring volcanic activity is crucial for ensuring the safety of nearby communities. An important step towards characterising volcanic regimes is the reduction of data into small, yet highly informative, feature vectors. Machine Learning (ML) has the potential to revolutionise volcano monitoring by efficiently processing large amounts of time series data, such as geophysical, geodetic, and geochemical observations, into feature vectors that can be analysed for similarities and differences, recognising patterns indicating unrest and potential precursors of specific eruptive phenomena, providing valuable insights that were previously not possible. However, sometimes ML tools can behave like “black boxes”, providing useful but hard to understand outputs. It is therefore necessary to consider the interpretability of the results provided. This can lead to improved transparency, trust and reliability of the ML tools and in turn enhance decision-making. In this way, the application of sophisticated pattern recognition algorithms at different levels of abstraction, from classifying individual seismic events to recognising distinct levels of unrest and possible outcomes, can really enhance our understanding of volcanic activity and improve our ability to forecast what the volcano will do next.
