Abstract

This presentation will have three parts. I will start with some (preliminary) results from two ocean bottom seismometer (OBS) experiments (UPFLOW and RHUM -RUM). I will additionally introduce an ongoing project using OBS data and machine learning.

The UPFLOW project, funded by the European Research Council (2021-2026), is an off-shore experiment in the Azores-Madeira-Canary region. We recovered 49 (out of 50) OBS in August and September 2022. Here, I will show various waveforms, spectrograms and data quality analysis from the recovered OBS data.

My second topic will cover the preliminary results of a variational autoencoder (VAE) architecture to denoise seismic waveforms to extract more information than previously possible. Data from OBS are inherently more challenging than their land counterpart because of their noisy environment (e.g. microseismic noises). Additionally, anthropogenic (e.g., ships) and animal noise (e.g., Whales) contribute to complex noise signals that can make it challenging to use traditional filtering methods (e.g., broadband or Gabor filters).

Finally, I will present results from the RHUM -RUM (2012-2016) experiment where for 13 months 57 OBS were deployed around the hotspot island of La Réunion. I present a 3-D P-wave tomography model supplemented by a global data set of P-diffracted measurements and a selection of ISC picks.

We observe the Large Low-Velocity Province (LLVP) rising 800 km above the CMB , forming a cusp beneath South Africa. A low-velocity branch undulates obliquely from this cusp region towards the uppermost mantle beneath La Réunion. The La Réunion’s connection to the lower mantle seems to be more complex than previously envisioned. The deep-mantle connections of the Afar and Kerguelen hotspots emerge from the same LLVP cusp beneath South Africa and extend towards the surface through tilted low-velocity branches.