Applications of Machine Learning in Volcanology

The characterization of volcano state is not a simple task due the complexity of physics processes underway. Understanding their evolution prior to and during eruptions is a critical point for identifying transitions in volcanic state. Permanent monitoring networks are developed for such a purpose. With the increase of the number of monitoring sites, the amount of available continuous data coming from different sources (infrasonic, seismic, GPS, geochemical, etc.) has increased exponentially and extracting the huge amount of information this data brings, represents a non-trivial task for researchers, who are always more often looking at the potentiality of computer algorithms to find correlations among them. Recent developments in the field of Machine Learning (ML) have proven to be very useful and efficient for automatic discrimination, decision, prediction, clustering and information extraction in many fields, including volcanology. In recent times, Deep Learning has seen rapid growth in its popularity along with other supervised strategies, such as Support Vectors Machines and Recurrent neural networks (RNN), which have consistently been applied with success to broader and broader sets of applications and fields. However, supervised machine learning requires labels for training, and obtaining these labels for large volumes of seismic and volcanic data is a very demanding and challenging task. Therefore, semi-supervised and unsupervised methods, such as Self-organized Maps, have been applied with success, to extract relevant information from huge amounts of unlabelled data. In seismic and deformative data processing, these techniques are used for waveform inversion, automatic picking of first arrivals, and interpretation of peculiar characteristics of transients. ML is helpful in the discrimination of magmatic complexes, in distinguishing tectonic settings of volcanic rocks, in the evaluation of correlations between volcanic signals and the chemico-physical composition of erupted materials. Other applications of ML in volcanology include the analysis and classification of geological, geochemical and petrological “static” data to infer for example, the possible source and mechanism of observed deposits, the analysis of satellite imagery to quickly classify vast regions difficult to investigate on the ground or, again, to detect changes that could indicate an unrest. The results obtained with the help of these algorithms would otherwise represent for researchers’ tasks hard to be solved with the usual standard methodologies.

The automatic characterization of signals provided in continuous monitoring data, increasingly encouraging a multiparametric approach, could be used to estimate the probability of an unrest and/or an eruption episode. The use of Machine learning strategies is of rising importance for monitoring aims, for differentiating geochemical and morphological patterns and stratigraphic issues. The aim of this Research Topic is to cover the latest advances of ML in volcanology, promote the method development and applications of ML in data processing, signal detection and classification, development of early warning systems, as well as advancing the data automatic processing and analysis.

Potential contributions include, but are not restricted to, the following topics:

• Development of ML-based signal detection, phase association and signal denoising methods;
• Discrimination of different types of volcano signals and their sources;
• Realtime intelligent processing system and its application at larger scales;
• Application of ML in infrasonic, seismic and deformative networks;
• Application of ML in ground motion forecasting;
• Comparison of the performance of ML models and methods in volcanology.