The relevance of full 3D-wavefield simulations for the tomographic filtering of geodynamic models

Tomographic-geodynamic model comparisons are a key component in studies of the present-day thermodynamic state of the mantle. A fundamental prerequisite for quantitatively meaningful comparisons is “tomographic filtering” of the geodynamic model. This means that geodynamically predicted mantle structures have to be modified to account for the spatially variable resolving power of tomographic images, i.e. to mimic the effects of uneven data coverage and regularization. Different approaches for tomographic filtering are available, but it is so far unclear which one will be the method of choice in the context of computationally demanding retrodictions of past mantle flow.

Here, we investigate the impact of the possible filtering approaches in a fully synthetic framework. For the first time in a mantle circulation model (MCM), we simulate 3D-wavefields and seismograms for an entire tomographic earthquake catalogue with over 4,200 events using SPECFEM3D_GLOBE. We use both classic filtering with the resolution operator R, as well as the recently introduced “generalized inverse projection” (GIP; Freissler et al. 2020) to generate tomographically filtered versions of the MCM.

In the GIP method, the generalized inverse operator of a given tomographic image is applied to synthetic seismic data predicted from the geodynamic model, as well as to potential data errors, to obtain the filtered MCM plus the propagated error. Important to note, the same generalized inverse operator is applied to an observed data set to build the tomographic model. A physically accurate prediction of synthetic data, here realized with the seismograms from numerical wave propagation, thus enables GIP filtering to consistently reproduce the tomographic imaging process. This is an important methodological advantage over classic filtering with R, where an unphysically reparametrized version of the MCM is filtered directly in model space and seismic data errors can not be considered.

In our study, GIP-filtered models are computed with cross-correlation S-wave traveltime residuals from the synthetic seismograms, as well as with banana-doughnut kernel and ray-theoretical traveltime predictions. The differently filtered models are compared against each other using statistical measures. By taking the GIP-filtered model that is based on the 3D-wavefield simulations as a reference, we can quantify the impact of reparametrization in classic filtering versus the lack of exact wave physics when using less accurate methods for traveltime predictions in the GIP filtering. Additionally, all filtered models can be compared to the underlying original structure of the MCM.

Detailed knowledge of tomographic filtering effects with different strategies is required prior to efforts on the associated uncertainty quantification in data-driven geodynamic retrodictions of mantle evolution.