Full waveform seismic tomography of the Vrancea region using the adjoint method

The Vrancea region, at the south-eastern bend of the Carpathian Mountains, Romania, represents one of the most particular seismically active zones of Europe. Beside some shallow crustal seismicity spread across the whole Romanian territory, Vrancea is the place of intense seismicity with the presence of a cluster of intermediate-depth foci placed in a narrow NE-SW trending volume below 60km depth. The occurrence of strong earthquakes in the past has raised questions about the nature of this deep intra-continental seismicity and increased the interest in the geodynamics of this earthquake-prone area. The central issue for seismic risk assessment is whether this singular seismogenic volume is geodynamically coupled to the crust. Large-scale mantle seismic tomographic studies have revealed the presence of a narrow, almost vertical, high-velocity body in the upper mantle. So far, two main different geodynamical models have been proposed for the region: (1) A subduction-related process, and (2) more recently a delamination process. High-resolution seismic tomography could help to reveal more details in the subcrustal structural models and to constrain the properties of the Vrancea Seismogenic Zone. Previous efforts have relied on classical ray-theoretical travel-time tomography to model data from local permanent or temporary instruments. Recent developments in computational seismology as well as the availability of parallel computing now allow modelling of the entire seismogram in a consistent way. This enables us to potentially retrieve more information out of seismic waveforms and to keep the modelling more uniform. In this work we want to assess the information gain that can be obtained using an adjoint-based inversion scheme combined with a full 3D waveform modelling, with respect to ray theory based tomography for the Vrancea region. The study is done with a dataset of local earthquakes from the broadband data of the CALIXTO 1999 experiment. This dataset is constrained by the quantity of usable seismic data as well as a poor signal-to-noise ratio, which imposes to carry out the analysis in a frequency band of relatively high frequencies. The adjoint tomographic inversion is implemented with the SPECFEM3D solver (a community code applying the spectral element method) to generate synthetic data and traveltime misfit kernels in a fine computational mesh that satisfies the data frequency band requirements. The computation is run on a BlueGene/Q massively parallel computer available at the CINECA supercomputing centre. We show what information can be retrieved about the tomographic model applying the full-waveform inversion and adjoint methods even applied to a modest-quality, mostly high-frequency content, regional dataset.