Short term forecasting of explosions at Ubinas volcano, Perú

• Autores: Traversa, Paola, Lengliné, Olivier, Macedo Sánchez, Orlando, Métaxian, Jean-Philippe, Grasso, Jean-Robert, Inza Callupe, Lamberto Adolfo, Taipe Maquerhua, Edu Luis
• Formato: artículo
• Fecha de Publicación: 2011
• Institución: Instituto Geológico, Minero y Metalúrgico
• Repositorio: INGEMMET-Institucional
• Lenguaje: español
• OAI Identifier: oai:repositorio.ingemmet.gob.pe:20.500.12544/2305
• Enlace del recurso: https://hdl.handle.net/20.500.12544/2305; https://doi.org/10.1029/2010JB008180
• Nivel de acceso: acceso abierto
• Materia: Erupciones explosivas; Explosiones; Sismicidad; Explosion; Forecasting; Seismicity; Volcano

Most seismic eruption forerunners are described using Volcano-Tectonic earthquakes, seismic energy release, deformation rates or seismic noise analyses. Using the seismic data recorded at Ubinas volcano (Peru) between 2006 and 2008, we explore the time evolution of the Long Period (LP) seismicity rate prior to 143 explosions. We resolve an average acceleration of the LP rate above the background level during the 2-3 hours preceding the explosion onset. Such an average pattern, which emerges when stacking over LP time series, is robust and stable over all the 2006-2008 period, for which data is available. This accelerating pattern is also recovered when conditioning the LP rate on the occurrence of an other LP event, rather than on the explosion time. It supports a common mechanism for the generation of explosions and LP events, the magma conduit pressure increase being the most probable candidate. The average LP rate acceleration toward an explosion is highly significant prior to the higher energy explosions, supposedly the ones associated with the larger pressure increases. The dramatic decay of the LP activity following explosions, still reinforce the strong relationship between these two processes. We test and we quantify the retrospective forecasting power of these LP rate patterns to predict Ubinas explosions. The prediction quality of the forecasts (e.g. for 17% of alarm time, we predict 63% of Ubinas explosions, with 58% of false alarms) is evaluated using error diagrams. The prediction results are stable and the prediction algorithm validated, i.e. its performance is better than the random guess.