Towards a High Resolution 3D Attenuation Model of the Upper Mantle

Understanding the distribution of seismic attenuation in the mantle is important for discriminating between chemical and thermal heterogeneity as well as interpreting models of elastic structure derived from data at different frequencies. While mapping of the 3D elastic structure of the upper mantle has benefited from approximate first-order perturbation techniques, modeling of the 3D distribution of seismic attenuation has lagged behind due to difficulties in accounting for purely elastic effects of scattering at sharp interfaces and (de)focusing, which can obscure the anelastic signal. Our most recent upper mantle Q model, QRLW8 (Gung and Romanowicz, 2004), was derived from long period seismograms in the time domain, comprising both fundamental mode surface waves and overtones, using an iterative waveform inversion technique, in which we solve for elastic and anelastic structure in successive steps. Until now, the theoretical framework for both the forward and inverse parts of our inversions has been the non-linear asymptotic mode coupling theory with focusing (Romanowicz, 1987; Li and Romanowicz, 1996), which limits our ability to retrieve 3D Q structure at wavelengths shorter than degree 8 in a spherical harmonics expansion of the model. In order to make further progress, it is necessary to 1) obtain better constraints on lateral gradients of structure in the elastic part of the model and 2) utilize a more accurate forward modeling theory. For the latter, the coupled Spectral Element Method (CSEM, Capdeville et al., 2003) allows a complete description of the 3D wavefield, and is computationally economical when applied to a restricted region of the earth (i.e. the upper mantle) and relatively low frequencies (here down to 80 s). Our intention is to combine this forward modeling approach with the use of approximate finite-frequency kernels in the inversion to construct a higher resolution 3D Q model of the upper mantle. To ensure good and uniform global data coverage, we have assembled a dataset of 3 component seismograms from 50 recent events observed at more than 100 stations of the IRIS/GSN, GEOSCOPE, GEOFON and various regional broadband networks. We present preliminary results that explore the influence of the forward modeling theory and the elastic model on the resulting Q model.