You are here

Long-term (probabilistic) forecasting of volcanic eruption onsets

Event Type: 
Event Date: 
31 July 2018 - 12:00pm to 1:00pm
ASE 3D Viz Laboratory Room (N2-B1c-16c)
Prof. Mark Bebbington
About the speaker: 

Mark Bebbington received a BSc in geophysics and an MSc in probability theory from Victoria University of Wellington (New Zealand), followed by a PhD in applied probability from the University of Cambridge. After a post-doc at the University of Queensland he took up a lectureship in stochastic operations research at Massey University, where he is now Professor in Geostatistics. Originally a statistical seismologist, Mark gravitated to volcanology when found collaborators at Massey. He has focussed on developing new probabilistic models for volcanic hazard forecasting, while continually attempting to lessen his ignorance of things volcanological. A past-chair of the IAVCEI Commission on Statistics in Volcanology, Mark has published more than 120 refereed papers, about half of them in volcanology and seismology, the rest in other areas of probability and statistics.

About the event: 

Volcanic risk mitigation strategies require a scientific assessment of the future evolution of the volcanic system concerned. Such assessments are subject to considerable uncertainty, which can only be quantified and expressed by means of probabilistic forecasting. This is divided into two branches, short-term (during a period of unrest or eruption) and long-term (during quiescence). This distinction is not arbitrary, but motivated by the availability of monitoring data in the former instance. In this seminar I will explore the latter case, which is based on the assumption that the past history of the volcano is the best guide to its future behavior. As a statistician in good standing, I will begin with data (historical, geological and sedimentary) proceeding to models beginning with renewal models, in which the only controlling factor is elapsed repose length. Examining the limitations and volcanological interpretation of these models will motivate non-stationary models which allow the volcano behavior to evolve. These require additional ancillary data on such as eruption size, geochemistry, etc., and new statistical techniques to accommodate them. Time permitting, we will take a brief look at the complicating effects of record incompleteness.