Summary of the context and overall objectives of the project (For the final period, include the conclusions of the action)
Seismic tomography images of the Earth's interior are key to the characterisation of earthquakes, natural resource exploration, seismic risk assessment, tsunami warning, and studies of geodynamic processes. While tomography has drawn a fascinating picture of our planet, today's individual researchers can exploit only a fraction of the rapidly expanding seismic data volume. Applications relying on tomographic images lag behind their potential; fundamental questions remain unanswered: Do mantle plumes exist in the deep Earth? What are the properties of active faults, and how do they affect earthquake ground motion?
To address these questions and to ensure continued progress of seismic tomography in the 'Big Data' era, we work on new technological developments that enable a paradigm shift in Earth model construction towards a Collaborative Seismic Earth Model (CSEM). Fully accounting for the physics of wave propagation in the complex 3D Earth, the CSEM is envisioned to evolve successively through a systematic group effort of my team, thus going beyond the tomographic models that individual researchers may construct today.
We develop the technological foundation of the CSEM and integrate these developments in studies of large-earthquake rupture processes and the convective pattern of the Earth's mantle in relation to surface geology. The CSEM project bridges the gap between regional and global tomography, and deliver the first multiscale model of the Earth where crust and mantle are jointly resolved. The CSEM will lead to a dramatic increase in the exploitable seismic data volume, and set new standards for the construction and reproducibility of tomographic Earth models.
Key methodological developments of this project include a novel stochastic optimisation method for seismic waveform inversion and wavefield-adapted spectral-element meshes. The combination of both allows us to reduce computational requirements by up to two orders of magnitude. This paradigm shift in waveform inversion allows us to assimilate vastly more data, thereby improving tomographic resolution and ground motion predictions.
The most concrete outcome of this project is the CSEM itself. While its first generation has been published nearly 3 years ago, the significantly improved second generation is nearly finished.
Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far (For the final period please include an overview of the results and their exploitation and dissemination)
Our work mainly consists of both technological developments and the construction of the Collaborative Seismic Earth Model (CSEM) itself. So far, the results have been disseminated in 21 published research papers and in numerous conference presentations (around 10 per year at all major geophysics conferences, including AGU and EGU). Furthermore, we explain our results for a broader public on our YouTube Channel (YouTube Channel of the Seismology and Wave Physics Group at ETH).
Concrete elements of our work include the following:
[1] We developed the first prototype of the Collaborative Seismic Earth Model (CSEM). This allows researchers for the first time to constrain seismic Earth structure on multiple scales in a collaborative fashion, thus taking advantage of our distributed human and computational power.
[2] We developed a novel evolutionary full-waveform inversion, which can automatically update a 3-D Earth model when new data become available. This is based on a specialised form of stochastic gradient descent where adaptive mini batches are being used. An additional advantage is the ability to use vastly more data than in conventional full-waveform inversion approaches.
[3] We developed a numerical modelling and inversion tool that uses wavefield adapted finite-element meshes. The method exploits the fact that wavefields are comparatively smooth in azimuthal direction, thereby allowing us to stretch elements perpendicular to the propagation direction. Effectively, this allows us to drastically reduce the number of elements, and therefore the required computational resources.
[4] Based on points 1 - 3, we are in the process of finalising the second generation of the CSEM, which includes novel regional updates in South America, the western United States, South-East Asia and Africa. All of these are being consolidated by a global full-waveform inversion. Results have been presented already at various conferences and will be published soon in a research paper
