Analysis of Seismic Similarity in Strombolian Eruptions from Mount Erebus, Antarctica By

II ABSTRACT Mount Erebus has long been recognized to produce very similar seismic signatures associated with its Strombolian eruptions. decorrelation of the coda, accompanied by relative stability in the early seismogram. Using coda wave interferometry, this decorrelation was interpreted as a median change within the volcano. Closer examination of this data in this thesis shows that clipping due to event size variations as well as more subtle features controlled by event size strongly influence such analysis techniques. Once it became obvious that source effects were very important influences on seismogram similarity, more sophisticated methods were implemented to better understand them. A moving window cross correlation analysis demonstrated more comprehensively that eruption size is a significant factor in seismogram similarity. A surprising discovery to come out of this research was the existence of time bands of strong correlation, probably arising from scatterers within the volcano at some distance (km) from the lava lake, which are even visible in the late coda (30+ s). Invariance is seen in the corresponding correlation lags associated with these scatterers, indicating stability in medium velocity. III Clustering the whole waveform seismograms also showed that size is the primary determiner of seismogram similarity. However, the fact that several clusters can have the same size (and even be active simultaneously) showed that other variable source processes must be involved as well. Videos of eruptions, as well as infrasound data, showed that other possible source process variations that may be important in this context include location of the bubble in the lava lake, lava lake convection phenomena (i.e., the lava lake state during eruption), and the geometry or pressure of the eruptive gas slug. Overall, similarities in how the initial and coda parts of the seismogram cluster also indicate that seismogram variability is due to source processes. Clustering the infrasound data resulted in completely different clusters then the seismic data. The clusters for the infrasound data were very similar, except for the rate of change of the first motion. This indicates that infrasound clusters are especially sensitive to rupture processes, making them promising for measuring lava lake rupture process. ii ACKNOWLEDGMENTS