Tremor clustering reveals precursors and evolution of the 2021 Geldingadalir eruption

Abstract Timely manner seismic data processing and analyses are essential for potential eruption prediction and early warning in volcanology. However, the complexity of eruption processes and precursory activities makes the analysis challenging. Here, we show that advanced machine learning techniques can provide an effective and efficient tool for extracting overlooked information from continuous seismic signal recorded during the 2021 Geldingadalir eruption in Iceland and reveal the temporal evolution of the eruptive activity. We identify the major phases of the eruption based on observed seismic signals throughout the eruptive activity. We distinguish unrest activities, continuous lava extrusion, and different levels of lava fountaining. We discover a precursory volcanic tremor sequence starting three days prior to the eruption, which could be used as an indicator of imminent eruptive activity. Based on the extracted patterns of seismicity and their temporal variations we provide an explanation for the transition mechanism from vigorous outflow to lava fountaining. Our observation suggests that the transition to episodic tremors in the seismic signal in early May, could be a result of an increase in the discharge rate in late April.