Could seismo-volcanic catalogs be improved or created using weakly supervised approaches with pre-trained systems?

Abstract. Real-time monitoring of volcano-seismic signals is complex. Typically, automatic systems are built by learning from large seismic catalogs, where each instance has a label indicating its source mechanism. However, building complete catalogs is difficult owing to the high cost of data labeling. Current machine learning techniques have achieved great success in constructing predictive monitoring tools; nevertheless, catalog-based learning can introduce bias into the system. Here, we show that while monitoring systems trained on annotated data from seismic catalogs achieve performance of up to 90 % in event recognition, other information describing volcanic behavior is not considered but is either discarded. We found that weakly supervised learning approaches have the remarkable capability of simultaneously identifying unannotated seismic traces in the catalog and correcting misannotated seismic traces. When a system trained on a master dataset and catalog from Deception Island Volcano (Antarctica) is used as a pseudo-labeler in other volcanic contexts, such as Popocatépetl (Mexico) and Tajogaite (Canary Islands) volcanoes, within the framework of weakly supervised learning, it can uncover and update valuable information related to volcanic dynamics. Our results offer the potential for developing more sophisticated semi-supervised models to increase the reliability of monitoring tools. For example, the use of more sophisticated pseudo-labeling techniques involving data from several catalogs could be tested. Ultimately, there is potential to develop universal monitoring tools that are able to consider unforeseen temporal changes in monitored signals at any volcano.