Date of Final Oral Examination (Defense)
Type of Culminating Activity
Master of Science in Geophysics
Jeffrey B. Johnson, Ph.D.
John R. (Jack) Pelton, Ph.D.
Jodi Mead, Ph.D.
Volcanic eruptions are powerful natural phenomena that often occur unpredictably in time and magnitude. Nearby communities are put at risk during volcanic unrest; however, when hazards are well understood and clearly defined risk can be mitigated. This thesis addresses the problem of forecasting the likelihood of future explosive volcanic behavior by monitoring ongoing eruptive history with infrasound. I parameterize inter-event temporal behavior to determine the eruption controlling processes is material failure opposed to changes in magma and volatile supply.
I analyze data from Sakurajima, a type-example open volcano, using two local (4 km from the vent) microphone arrays, which recorded infrasound continuously from 18-25 July 2013. Both explosive and emergent degassing events are classified using the infrasound data, along with the inter-eruptive quiescent periods. I use the Fisher statistic to quantitatively measure acoustic signal coherency originating from Sakurajima’s active vent, Showa crater. This allows me to determine the statistics associated with vent activity prior to 366 detected degassing events. All observed 366 repose intervals form a distribution that I compare with known exponential, gamma, and Weibull probability distribution models. The entire set of repose interval lengths is best fit by a gamma distribution model representative of a stationary Poisson process, suggesting that events are controlled by material failure phenomenon rather than a dynamic process such as changes in magma or volatile flux.
Detected volcano infrasound is categorized based on recorded pressure amplitude as either explosive (> 3.5 Pa) or passive degassing (< 3.5 Pa). By observing the separate distributions of repose interval lengths that precede the two eruptive modes, I develop a forecasting variable, the Relative Squared Median Residual Sum (RSMRS) that describes which mode is more likely to occur during an observed period of quiescence, based on past behavior. The forecasting reliability depends on the separation and the peakedness of RSMRS distributions for each mode. A RSMRS threshold value is used to anticipate either passive degassing or explosive degassing. Results may differ for other volcanoes with different styles of eruption or for Sakurajima activity during different periods.
The RSMRS forecaster is run coincident with signal detection and is capable of operation in near real-time with the availability of telemetered data. The forecasting algorithm is trained with enough data such that repose interval distribution for each mode of activity begins to take shape. Consistent with Poisson process (gamma distribution) assumptions, each eruptive mode converges on its respective arrival rate. Explosion forecasting results in 76% true positive (anticipated explosion resulted in explosion) rate at an RSMRS cutoff of 0.2998. The true negative rate was 97.5% and is defined as correctly anticipating passive degassing following periods of quiescence.
VonLintig, Matthew R., "Volcano Infrasound Monitoring with Applications for Statistical Forecasting of Explosions at Sakurajima (Japan)" (2018). Boise State University Theses and Dissertations. 1497.