Detecting earthquakes from the high atmosphere using machine learning

Are you interested in geohazard monitoring and machine learning? The main objective of this project is to develop new methods for the automatic detection of earthquakes from the high atmosphere. In the long term, these methods will be used to develop a new satellite-based tsunami early-warning system.

Large offshore EarthQuakes (EQs) can generate strong tsunamis with devastating consequences for coastal populations. EQs are traditionally monitored using their seismic signature that can be recorded on the ground to provide timely alerts. However, assessing an EQ’s tsunamigenic potential is a difficult task with distant seismic data.

Fortunately, seismic energy can couple to the ocean and atmosphere and propagate up to high altitudes in the ionosphere (~200-400 km altitude) as acoustic and gravity waves. Such waves reach the ionospheric altitudes at about 8 min (acoustic waves) or 15-60 min (gravity waves) carrying information about the source that generated them. The exploitation of this information in acoustic and gravity waves could, in the long term, enable the retrieval of source parameters and assess the tsunamigenic potential of an EQ from the atmosphere.

The first step toward this new and revolutionary way to assess natural hazards is the automatic detection of these signals in ionospheric data. Recently, our team has developed the first Machine-Learning (ML) method to automatically detect EQs from the ionosphere (Brissaud & Astafyeva, 2022) using GNSS satellites within 15 min providing enough time to alert coastal populations.

However, our ML model requires improvements in both its methodology and computational time to prepare its implementation as an automated early warning system at GNSS stations. In this master project, you will be invited to:

  • Implement novel ML techniques such as Convolutional Neural Networks (CNNs) and Graph networks to detect and associate EQ signals in ionospheric data
  • Apply the ML model to recent EQs for further validation
  • Contribute to the publication of research results in scientific journals
  • (if time allows) work on the development of an inversion framework to retrieve EQ location from automatically detected ionospheric signals.

Preferred qualifications:

  • Excellent knowledge of Python, and ML library (e.g., TensorFlow or PyTorch)
  • Signal processing
  • Wave propagation theory
  • Timeseries processing using ML
  • Git

Reference:

Published Sep. 26, 2022 9:45 AM - Last modified Sep. 26, 2022 9:50 AM

Supervisor(s)

Scope (credits)

60