We apply Self-Organising Maps (SOM) to assess the low level seismic activity prior to small scale phreatic events at Ruapehu volcano New Zealand. The SOM approach allows an automatic pattern recognition, virtually independent from a priori knowledge. Volcanic tremor spectra are randomly presented to the network in a competitive iterative training process, followed by a hierarchical clusterization of the SOM nodes. Spectra are then projected, ordered by time, to clusters on the map. A coherent time evolution of the data through the clusters can highlight the existence of different regimes and the transitions between them. Two Ruapehu events were examined: a phreatic event on 4 October 2006 which displaced the crater lake producing a 4m high wave on the lake edge, and the more energetic 25 September 2007 phreatic eruption. The SOM analysis provides a classification of tremor spectral patterns that clusters into three regimes that we label by colours. The pattern for both eruptions is consistent with a pre-eruption spectral pattern including enhanced spectral energy in the range of 4 to 6Hz — labelled ‘green tremor’. This gives way to spectra having broader energy between 2 and 6Hz, the so called ‘red tremor’ just prior to the eruption. The post eruption pattern includes spectral peaks at generally lower frequencies of 2 to 4Hz — the so called ‘blue tremor’. Clusterization into only three groups yields highly non-unique solutions which cannot explain the variety of processes operating at Ruapehu over long time periods. Regardless, the approach highlights noteworthy similarities that may be explained by a pattern of slow pressurisation under a hydrothermal or magmatic seal – ‘green’ – followed by seal failure – ‘red’ – and subsequent de-pressurisation – ‘blue’ – for the two events studied. Although the application shown here is limited, we think it demonstrates the power of this classification approach.
Read full abstract