Global Attenuation Model of the Upper Mantle
Thursday, 17 December 2015: 10:50
307 (Moscone South)
We present a three-dimensional shear attenuation model based on a massive surface wave data-set (372,629 Rayleigh waveforms analysed in the period range 50-300s by Debayle and Ricard, 2012). For each seismogram, this approach yields depth-dependent path average models of shear velocity and quality factor, and a set of fundamental and higher-mode dispersion and attenuation curves. We combine these attenuation measurements in a tomographic inversion after a careful rejection of the noisy data. We first remove data likely to be biased by a poor knowledge of the source. Then we assume that waves corresponding to events having close epicenters and recorded at the same station sample the same elastic and anelastic structure, we cluster the corresponding rays and average the attenuation measurements. Logarithms of the attenuations are regionalized using the non-linear east square formalism of Tarantola and Valette (1982), resulting in attenuation tomographic maps between 50s and 300s. After a first inversion, outlyers are rejected and a second inversion yields a moderate variance reduction of about 20%. We correct the attenuation curves for focusing effect using the linearized ray theory of Woodhouse and Wong (1986). Accounting for focussing effects allows building tomographic maps with variance reductions reaching 40%. In the period range 120-200s, the root mean square of the model perturbations increases from about 5% to 20%. Our 3-D attenuation models present strong agreement with surface tectonics at period lower than 200s. Areas of low attenuation are located under continents and areas of high attenuation are associated with oceans. Surprisingly, although mid oceanic ridges are located in attenuating regions, their signature, even if enhanced by focusing corrections, remains weaker than in the shear velocity models. Synthetic tests suggests that regularisation contributes to damp the attenuation signature of ridges, which could therefore be underestimated.