Using a process-based model of pre-eruptive seismic patterns to forecast evolving eruptive styles at Sinabung Volcano, Indonesia
Most volcanoes worldwide are not monitored in real-time; for those that are, patterns of pre-eruptive earthquakes coupled with conceptual models of magma ascent enable short-term forecasting of eruption onset. Basic event locations, characterization of background seismicity, and recognition of changes in earthquake types and energy release are most important to successful eruption forecasting. During renewed activity at Sinabung volcano, Indonesia, this approach was used by the Center for Volcanology and Geological Hazards Mitigation (CVGHM) and the USGS Volcano Disaster Assistance Program to forecast eruption onset, identify changes in eruptive styles and raise or lower alert levels and extend or contract evacuation zones. After > 400 years of quiescence, Sinabung began erupting in August 2010. The volcano was unmonitored at the onset of these eruptions, which were phreatic, but soon after a monitoring network was installed by CVGHM. Increasing swarms of high-frequency volcano tectonic (VT) earthquakes were used to forecast continuing phreatic eruptions. Volcanic activity decreased in mid-September 2010, while additional intrusions at depth (inferred from continued distal VT swarms) continued through September 2013, when explosive phreatic eruptions recurred. Explosive eruptions were forecast based on increases in real-time seismic amplitude measurement (RSAM) and VT seismicity. Seismicity changed markedly in late November and early December 2013 with the occurrence of deep earthquakes and an overall transition from low-frequency (LF) dominated and irregular (in time and magnitude) earthquakes to more regular LF and hybrid seismicity – a transition that accompanied the continued rise, eventual emergence and growth of a lava dome in the summit crater. This lava dome was first observed on 18 December. In late December 2013 to early January 2014, the eruptive style changed again as additional ascending magma deformed the summit and the dome grew beyond the capacity of the summit crater, resulting in the en masse collapse of the lava dome (2 Mm3) on 11 January and the largest pyroclastic flow to date. The collapse was forecast on the basis of a several order of magnitude increase in RSAM, continued strong distal VT seismicity, an increase in proximal seismicity, and large-scale observed deformation of the summit area. Similarly, a later collapse of a second summit lava dome on 1 February 2014 was forecast on the basis of increased distal seismicity. Here, we demonstrate how a process-based volcano seismicity model was used in combination with real-time data to forecast the time and magnitude of eruptions, as well as changes in eruption style.