Earthquake Detection, Classification and Forecasting with Machine Learning

Natural earthquakes, wind, waves and anthropic activity create a rich background ambient noise that is monitored 27/7 by national and international broad-band seismic stations. Recently, Machine Learning (ML) tools like CNNs, auto-encoders and clustering techniques have opened interesting avenues, allowing to enhance traditional tools, and to harvest a vast quantity of data by automating the tedious processing by an operator.

Scrutinizing subtle signals hidden within noisy time series allows to understand the earthquake cycle, to detect and locate a myriad of microearthquakes and to discriminate different types of seismic sources (natural earthquakes, explosions, collapses, landslides, industrial or anthropic noise). Moreover, it may allow to identify slow slip motion and non-volcanic tremor which sometimes indicate the active nucleation phase of a large earthquake, as observed in mega-thrust events of Chile and Japan.

The aim of this PhD is to develop and generalise the ML approach for the analysis and classification of seismic signals, to test its generality, and explore the geophysical origin of the detected signals in the framework of the seismic cycle. Classification through clustering will aim at detecting different class of earthquakes and separating them according to region, tectonic style, depth,  intra- versus inter-plate location, fast and slow rupture, potential for hazard and for tsunami generation.

In addition, the student will investigate the potential integration of the ML in (1) seismic Early Warning protocols, (2) real-time scenarios of probabilistic forecasting and risk mitigation, (3) automation and improvement in the detection, location and classification of seismic events in regional seismic catalogues (it has been shown that up to 60% more events can be detected by using CNN methods).

The target sites will be zones that have high seismic activity and good quality regional seismic data. Examples of optimal target sites are East coast of Japan, West coast of Chile, California.

In summary, the aim is to fully develop the potential of ML in seismology to:

-identify early signs of earthquakes based on change in seismic signal behaviour

-differentiate natural seismic events from other seismic signals

-classify types of earthquake signals that can be attributed to particular seismic scenarios


The student will select seismic data available from the Global Seismic Network (GSN) broadband stations and regional strong motions networks such as K-Net and Kik-Net (Japan), to create a catalogue of earthquake-related signals and classify them by region, type of focal mechanism and depth. The catalogue will then be used to train a CNN and to conduct ML analysis ex-tending current investigations that successfully tested auto-encoders, k-means clustering, multi-dimensional scaling and boosted learning trees to classify seismic signals.

Project Timeline

Year 1

Review existing literature. This field is progressing very fast and the best type of ML implementations need be identified in the most recent studies

Identification of regions of interest for the training and testing of the ML workflow

Acquisition and organisation of public domain data

Set up a database, embed metadata for flexible access and selection by ML methods

Initial development of ML tools

Year 2

Continue development of ML tools

Setup an efficient pipeline for efficient and automated workflow

Initiate training and testing of the ML on the database

Year 3

Finalise development phase

Test of generalisation for the method onto data from different regions than those where the algorithm was trained

Dissemination of the results

Year 3.5

Pursue the test of generalisation, including signals with lower signal to noise ratio

Dissemination of the results

Write-up of the PhD report

& Skills

The student will acquire training and skills in:

Handling of Big Data

Sourcing of material from public databases

Organising and classifying seismic events

Building a structured catalogue with metadata information

They will also learn:

Essentials of earthquake source physics and tectonic context where earthquakes happen

Use of Python libraries such as Keras, Tensorflow, pyTorch for Machine Learning

High-level statistics for advanced analysis of time series

References & further reading

✦ Galea, A. and Capelo, L. (2018). Applied Deep Learning with Python: Use scikit-learn, TensorFlow, and Keras to create intelligent sys-tems and machine learning solutions. Packt Pub-lishing, 2018.
✦ Goyal, P. and Pandey, S. and Jain K. (2018). Deep Learning for Natural Language Pro-cessing: Creating Neural Networks with Python. Apress.
✦ Guerin-Marthe, S., Nielsen, S., Bird, R., Gia-ni, S. & Di Toro, G (2018). Earthquake Nuclea-tion Size: Evidence of Loading Rate Dependence in Laboratory Faults. Journal of Geophysical Research: Solid Earth 124(1): 698-708. https://doi.org/10.1029/2018JB016803
✦ Hulbert, C. et al. (2018). Similarity of fast and slow earthquakes illuminated by machine learn-ing. Nature Geoscience 12: 69-74.
✦ Karim, F. et al. (2018) LSTM fully convolu-tional networks for time series classification, IEEE access 6, 1662-1669.
✦ Pattanayak, S. (2017). Pro Deep Learning with TensorFlow: A Mathematical Approach to Advanced Artificial Intelligence in Python. Apress.
✦ Rouet-Leduc, B., Hulbert, C. & Johnson, P. A. (2019), “Continuous chatter of the cascadia subduction zone revealed by machine learning”, Nature Geoscience 12(1), 75-79.
✦ C. W. Johnson, Y. Ben‐Zion, H. Meng, and F. Vernon, ‘Identifying Different Classes of Seismic Noise Signals Using Unsupervised Learning’, Geophys. Res. Lett., vol. 47, no. 15, Aug. 2020, doi: 10.1029/2020GL088353.
✦ L. Seydoux, R. Balestriero, P. Poli, M. de Hoop, M. Campillo, and R. Baraniuk, ‘Cluster-ing earthquake signals and background noises in continuous seismic data with unsupervised deep learning’, Nat Commun, vol. 11, no. 1, p. 3972, Aug. 2020, doi: 10.1038/s41467-020-17841-x.
✦ D. Snover, C. W. Johnson, M. J. Bianco, and P. Gerstoft, ‘Deep Clustering to Identify Sources of Urban Seismic Noise in Long Beach, California’, Seismological Research Letters, vol. 92, no. 2A, pp. 1011–1022, Mar. 2021, doi: 10.1785/0220200164.

Apply Now