OPPORTUNITY – Dual PhD degree in volcano-geophysics with emphasis in Machine Learning

The Department of Earth and Environmental Sciences at the University of Michigan (UM) – Ann Arbor, USA, in collaboration with the Institut des Sciences de la Terre (ISTerre) at the Université Grenoble-Alpes (UGA) and the Université Savoie Mont Blanc (USMB), France, are seeking for a highly motivated B.S. or preferably an M.Sc. student to pursue a dual PhD degree in volcano-geophysics with special emphasis in Machine Learning (ML). We encourage in particular students with a background in geophysics, computer science, physics, or a closely related field to apply.

The PhD student will work in collaboration between the two institutions and will be co-mentored by Zack Spica (UM), Corentin Caudron, and Philippe Lesage (ISTerre). To obtain a dual-PhD degree, the candidate will have to comply with both the rules of the Rackham Graduate School and l’École Doctorale Terre Univers Environnement of UGA and USMB, which will be gained by spending at least 18 months in France. International fieldwork with a special emphasis on Indonesia will be possible during the 4 years (minimum) of the PhD. The candidate will have the opportunity to take advanced classes in ML at UM, a leading research institution in the US. A short description of the project is outlined below.

Candidates should submit their completed application form to the graduate program at UM prior to January 7th, 2021 in order to start in September 2021. This includes three letters of recommendation and TOEFL scores if English is not the native language. For further information about the admission process, the dual degree, and the scientific perspective of the project, please contact us (zspica[at]umich[dot]edu; corentin.caudron[at]univ-smb[dot]fr).

Project description:

Machine Learning (ML) approaches are increasingly being used in volcano-seismology to automatically classify earthquakes. Yet most of these techniques are supervised and rely on existing catalogs, sometimes requiring already labeled data. Recent advances in Earth Science have however taken advantage of unsupervised strategies which allow defining classes using unlabeled dataset.

Taking advantage of numerous existing seismic datasets, the goal of this project is to apply deep learning approaches to better assess precursors signals to forecast future eruptions. We also want to compare new classification/clusters approaches with existing catalogs from local partners and explore the relationships between them. Ultimately, these tools will be implemented for real-time monitoring purposes and will support decision-making at the observatory level. 

Step 1: Deep-learning of volcano-seismic events using existing approaches to improve our understanding of the source processes, with a focus on volcanic tremor (e.g., Scatnet).

Step 2: Bringing our tools to the operational real-time level.

Step 3: ML-based approaches on multi-parametric observations to improve forecasting.

Related readings: 

  • Bergen, K. J., et al. “Machine learning for data-driven discovery in solid Earth geoscience.” Science 363.6433 (2019).
  • Dempsey, D. E., et al. “Automatic precursor recognition and real-time forecasting of sudden explosive volcanic eruptions at Whakaari, New Zealand.” Nature communications 11.1 (2020): 1-8.
  • Seydoux, L., et al. “Clustering earthquake signals and background noises in continuous seismic data with unsupervised deep learning.” Nature communications 11.1 (2020): 1-12.
  • Malfante, M., et al. “Automatic classification of volcano seismic signatures.” Journal of Geophysical Research: Solid Earth 123.12 (2018): 10-645.
  • Carniel, R., and Guzmán S. R., “Machine Learning in Volcanology: A Review.” Volcanoes-Updates in Volcanology (2020).