Peakmatch – a New Program for Large-Scale Seismic Multiplet Analysis and Event Classification: An Application to Soufrière Hills Volcano, Montserrat.

Abstract:

The large amounts of data generated by volcano observatories, especially during volcanic crises, can present a challenge for rapid analysis of changing event behaviour. Statistical analysis of seismic events, e.g. spectral analysis and identification of waveform families (i.e. multiplet analysis) can be useful tools in characterising the seismicity and can provide another layer of information for volcano observatories. Additionally, multiplet analysis of existing data from key eruptions can reveal patterns of behaviour that are not evident from individual event metrics (e.g. locations, amplitudes, frequency content etc.). However, large scale multiplet analysis has previously been hindered by the computational limitations associated with cross-correlation of large data sets. Event classification remains crucial to volcano monitoring and to retrospective analysis of key eruption data, but such methods are often not straightforward to implement. We present a program, PEAKMATCH, which can be used for spectral and multiplet analysis of large data sets, and can be used for simple automatic classification of volcanic events. Our optimised cross-correlation approach can efficiently handle the cross-correlation of hundreds of thousands of events making large-scale multiplet studies a reality. Much of the crucial information for classifying an event lies in the frequency domain and by using spectral ratios and cluster analysis we are able to automatically classify events without an extensive training period. We apply our method to seismic data from Soufrière Hills Volcano, Montserrat from the end of the last extrusion period in February 2010 until January 2014 to characterise the multiplet behaviour during an extended period of quiescence. We observe notable changes in multiplet behaviour surrounding the period of ash venting in March 2012.