Non-negative Matrix Factorization: an application to Erta ’Ale volcano, Ethiopia

Non-negative Matrix Factorization (NMF) is an emerging new technique in the blind separation of signals recorded in a variety of different fields. The application of these techniques to the analysis of volcanic signals is new to date. Volcanic tremor, the continuous seismic signal recorded close to a volcano, often consists of a mixture of signals having different and independent sources, both volcanic and non-volcanic, possibly including anthropogenic ones. In this paper we show that NMF is a suitable technique to separate such a mixture of foreground / interesting / target ”signals” from background / interference / undesired ”noise”. The encouraging results obtained with this methodology in the presented case study, separating high convection foreground signal from low convection background noise at Erta ’Ale lava lake, support its wider applicability in volcanic signals separation.