An optimized parallel LSQR algorithm for large-scale seismic tomography

Seismic recordings represent convolution of a source wavelet with physical properties of the Earth’s interior, thus different components of the seismic recordings (e.g. traveltime of seismic phases, amplitudes and seismic waveforms) can be used to image structures and compositions of the Earth (e.g. Iyer and Hirahara, 1993; Nolet, 2008; Romanowicz, 2003; Stein and Wysession, 2002). By using different inversion techniques, the information extracted from the seismic recordings has been used to invert threedimensional (3D) model that represents the Earth’s physical properties. In past decades, many seismic tomographies with various scales have been produced and used to interpret geodynamic systems (e.g. Becker and Boschi, 2002; Calvert et al., 2000; Gutscher et al., 2010; Pari and Peltier, 1995), plate tectonics mechanisms (e.g. Anderson et al., 1992; Gutscher et al., 2010; Iyer and Hirahara, 1993; Zhang and Tanimoto, 1993), magma chambers (e.g. Lees, 1992; Patane, 2006), and structural details of Earth’s crust and fault zones(e.g. Catchings et al., 2002; Chen et al., 2007b; Roecker et al., 2006; Zhang and Thurber, 2005).