Reducing wind noise in seismic data using Non-negative Matrix Factorization: an application to Villarrica volcano, Chile

Abstract

Single channel source separation of seismic signals is an appealing but difficult problem. In this paper, we introduce a semi-blind single-channel seismic source separation method to enhance the components of volcanic origin. In this method, the source decomposition scheme is addressed as a Sparse Non-negative Matrix Factorization (NMF) of the time-frequency representation of the single vertical seismic channel. As a case study we present an application using seismic data recorded at Villarrica volcano, Chile, one of the most active in the southern Andes. The analysed dataset is strongly contaminated by wind noise and the procedure is used to separate a component of volcanic origin from another of meteorological origin.