Skip to content

Probabilistic Failure Forecast Method (FFM) of geochemical observables to forecast the time of geological events in volcanic systems

Nemesio M. Pérez1,2, Luca D'Auria1,2, Pedro A. Hernández1,2, Gladys V. Melián1,2, Eleazar Padrón1,2, Germán D. Padilla1,2 María Asensio-Ramos1 and  Berverley Coldwell1,2

  • Affiliations: 1Instituto Volcanológico de Canarias (INVOLCAN), Puerto de la Cruz, Tenerife, Canary Islands 2Instituto Tecnológico y de Energías Renovables (ITER), Granadilla de Abona, Tenerife, Canary Islands   

  • Presentation type: Poster

  • Presentation time: Monday 16:30 - 18:30, Room Poster Hall

  • Poster Board Number: 279

  • Programme No: 2.4.48

  • Theme 2 > Session 4

Abstract

Voight (1988, 1989), Voight and Cornelius (1991) and Cornelius and Voight (1995) proposed their material Failure Forecast Method (FFM), starting from the study of the Mt. St. Helens eruption. Since then, most of the FFM implementations have been applied to geophysical observables such as deformation and seismicity. The energy released by the diffuse volcanic degassing process is considerable, and it cannot be neglected in the energy balance of a volcanic systems. Since gas emission rates and/or chemical variations may be influenced by several factors associated with brittle fracture associated to volcanic unrest or eruptions, the rapid increase in these geochemical observations, prior to geological events in volcanic systems, can be considered as a geochemical precursor. After selecting an appropriate time window in the geochemical time series, we can apply the Failure Forecast Method (FFM) to forecast the time of a significant geological event like a volcanic unrest or eruption. In this work, we propose the application of the FFM in a probabilistic fashion (PFFM) by using the bootstrap method. For each geochemical parameter we obtained a probability distribution of the intercept time. Specifically, we use the 5th and 95th percentiles to define the 90% confidence range. Since most of the obtained probability distribution are highly multimodal, and with numerous outliers, we considered using the median instead of the maximum likelihood of the mean value to estimate the most reasonable forecast date.  (B. Voight, Nature, 1998; B. Voight, Science, 1999; R. Cornelius and B. Voight, JVGR, 1995)