CAREER: Evolution of the Oceanic Plate & Upper Mantle with Deep Probabilistic Seismic Imaging

This CAREER proposal uses seismology and machine learning to explore questions about the structure and formation of the Earth's crust beneath the ocean. The processes that form and shape the chemical and physical properties of the rocks beneath the seafloor are different from those associated with the rapid and voluminous volcanism generated during the creation of large undersea volcanoes and ocean-islands.

While many aspects of these processes are well understood, questions remain about the details and causes of the boundary between the stiff rocks (lithosphere) and the weak interior (asthenosphere). It is still unclear how the uplift and emergence of large oceanic plateaus and ocean islands is related to deep volcanism or shallow thermal cracking. These questions will be investigated by using machine learning to image the subsurface of the largest ocean-basin using existing seismic data.

Pacific-wide imaging using ground vibrations is expensive and challenging. The seafloor is sparsely instrumented and the data noisy. Artificially intelligent algorithms will be trained to extract useful information from noisy seismic data, and the signals will be used to reconstruct 3D wave speed images of ocean interior.

The improved resolution of the ocean subsurface informs tests of models explaining ocean plate genesis and evolution. Students will be recruited and trained using project-based courses that prepare them for research careers in academia and industry using state-of-the-art tools and resources. The educational plan includes two new project-based courses: “Seismic Signals & Noise” and “Earth Imaging and Machine Learning,” which closely integrate with the research questions and provide learning and research experience for a diverse group of undergraduate and graduate students.

Multi-scale deep probabilistic seismic imaging is the process of reconstructing ‘uncertainty-aware’ images of a planet’s interior by stitching together wave signals extracted from sparse and noisy ground-vibration data (earthquake and ambient noise body-waves and surface waves) using sensor arrays with different space-time footprints (small and large aperture as well as short and long duration deployments). This approach uses artificially intelligent machine learning algorithms to solve data and model prediction steps in the image reconstruction problem.

Step 1, cleaner signals: improved wave detection by denoising the long-period (SS, PP) and short-period teleseismic body-wavefield (Ps- and Sp-RFs). Step 2, improved images: joint modeling of the waves through a realistic and adjustable 3-D mesh takes advantage of waves propagating with different sensitivities. Step 3, fast reconstruction with errors: a probabilistic inverse modeling generates a large ensemble of models from which uncertainties can be estimated.

This framework requires that many images are generated in step 2 and the process iterated. Again, intelligent algorithms, such as generative AI, are used to speed up, learn, and compress the model ensemble. The software and algorithms developed will contribute to the exploration of planetary interiors in situations where data is often sparse and noisy.

This project is jointly funded by the Geophysics program and Education and Human Resources Program in the Division of Earth Sciences and the Marine Geology and Geophysics program in the Division of Ocean Sciences.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.