Seismic velocity changes in response to snow loading at Mount Ruapehu volcano, New Zealand, using passive seismic interferometry

<p>Passive seismic interferometry has become a popular technique for monitoring volcanoes over the past two decades. Despite this, volcanoes still represent challenging locations to apply the methodology due to the presence of volcano seismicity. Volcanic tremor, in particular, can significantly alter the character of cross-correlation functions. This leads to the possibility of mis-interpreting changes in phase or waveform shape as due to real subsurface processes.</p><p>Mount Ruapehu is one such volcano where volcanic tremor is regularly recorded above 1 Hz. Thus, a previous study applying passive interferometry at the volcano during its most recent eruptive period (2006&#8211;07) focused on lower frequencies to reduce the risk of contamination. In this work, we target the higher frequencies that include volcanic tremor (1&#8211;4 Hz) during approximately the same period (2005&#8211;2009), thus providing an opportunity to monitor changes at shallower depths within the volcanic system. To assess the suitability of the tremor as a repeatable seismic source, we first apply an unsupervised machine learning technique in the form of agglomerative hierarchical clustering of cross-correlation functions. Doing so allows us to form groups of data that share similar characteristics and, unlike commonly used similarity measures, does not require a defined reference period. Through this, we find that cross-correlation functions at higher frequencies are both relatively consistent in time and dominated by seasonal processes (with alternating summer and winter clusters clearly identified).</p><p>Applying the wavelet method to compute travel-time changes in the time-frequency domain reveals snow loading to be the most likely process influencing seismic velocities on the volcano. Amplitudes of +/- 0.5% are recorded at the seismic station closest to the summit, with peak velocites occurring at the same time as maximum snow thickness. In contrast, the seasonal trends recorded at seismic stations with minimal snow cover are of lower amplitude (+/- 0.1%), opposite in sign, and are best fit using a model based on fluid pressure changes in response to precipitation. No obvious short-term changes are detected prior to phreatic eruptions in 2006 and 2007. It is of interest, however, that both eruptions occur approximately one year apart following the initial decrease of velocities in response to snow unloading/melt, suggesting a causal relationship may exist.</p>