30/10/2021
The Human Resources Strategy for Researchers

Thesis in environmental seismology and machine learning [M / F]

This job offer has expired


  • ORGANISATION/COMPANY
    CNRS
  • RESEARCH FIELD
    Astronomy
    Environmental science
    Geosciences
  • RESEARCHER PROFILE
    First Stage Researcher (R1)
  • APPLICATION DEADLINE
    19/11/2021 23:59 - Europe/Brussels
  • LOCATION
    France › STRASBOURG
  • TYPE OF CONTRACT
    Temporary
  • JOB STATUS
    Full-time
  • HOURS PER WEEK
    35
  • OFFER STARTING DATE
    01/02/2022

OFFER DESCRIPTION

The thesis will mainly take place at the Earth and Environment Institute of the University of Strasbourg (ITES) within the Active Dynamics team, with regular visits to the Research Institute in Computer Science, Mathematics, Automatics and Signal (IRIMAS) ) of the University of Haute Alsace. It will be co-supervised by C. Hibert and J.-P. Malet at ITES, G. Forestier and J. Weber at IRIMAS. C. Hibert and J.-P. Malet will supervise the doctoral student on aspects of seismology, knowledge of the objects studied and the general context of the study in environmental seismology. The expertise of G. Forestier and J. Weber lies in the development and implementation of innovative machine learning methods and will supervise the doctoral student on these themes. The doctoral student will benefit from access to the meso-center of the University of Strasbourg (high performance computing center).

With a large proportion of international students and a significant cultural mix, Strasbourg is one of the most attractive French cities for international students.

Title: Automated exploration and analysis of seismological data flows by supervised and unsupervised machine learning methods.

The fall in the cost of seismological sensors has made it possible to considerably densify the existing observation networks and to deploy instruments for applications that go beyond the study of faults and earthquakes. Seismological networks are now deployed to study many geological objects, such as volcanoes, landslides, glaciers, watersheds, underwater hydrothermal vents or karst rivers. All these contexts have their own endogenous seismicity, carrying information on the physical processes that affect them, and therefore able to provide unique elements to understand their dynamics and functioning. However, this seismicity is often characterized by a very large number of events, with very diverse sources (fracturing, friction, explosion / implosion, fluid circulation, landslides, granular flows, impacts, etc.), resulting in an extreme variety of signals. The major difficulty in exploiting and interpreting seismic streams lies in the quantity of data to be processed, of events to be extracted from continuous records, and in the construction of catalogs of micro-seismic sources. It is necessary to automate these tasks to make the best use of the raw seismological data. Recent developments in artificial intelligence methods constitute a unique opportunity for the community interested in this micro-seismicity, whatever the context. We have conducted several studies on the use of machine learning algorithms for the automatic construction of catalogs which have given encouraging results [Provost et al., 2017; Maggi et al., 2017; Hibert et al., 2017, 2019; Wenner et al., 2021], but have also made it possible to identify scientific and methodological obstacles. These difficulties relate to the need to pre-detect events before classification (by an STA / LTA or Kurtosis type detector for example), to take into account the variability of the physics of sources over time, the ability to identify sources rare or new, the (sometimes extreme) over-representation of certain classes (especially noise) and the diversity of instrument networks (number, geometry, sensitivity, distance to sources, and wave propagation media). The overall objective of this thesis is thus to develop new strategies for systematic and continuous exploration of seismological data streams by the mobilization of machine learning algorithms, making it possible to remove these obstacles and arrive at a generalist method. for applications of micro-seismicity studies.

Several areas of work have been identified: The first axis consists in proposing a strategy to dispense with a prior a-priori detection of seismic signals. The approach envisaged is to process all the raw data (background noise and signals), by deploying machine learning algorithms on sliding windows. This will require adapting the attributes of the seismic signals used in the machine learning approaches already tested [e.g. Provost et al., 2017]. This step of parameterization of the raw data is critical because these attributes constitute the basis of the majority of machine learning approaches. The second axis aims to test semi-supervised or fully unsupervised algorithms such as deep neural networks [e.g. Fawaz et al., 2018a; 2019] or clustering algorithms [Forestier et al., 2010] to automatically constitute seismicity catalogs. The imbalance of the training sets that will arise with a sliding window approach could be solved using synthetic seismic signals [e.g. Forestier et al., 2017; Fawaz et al., 2018b]. This should make it possible to envisage the establishment of a robust, versatile and portable processing chain, capable of being easily deployed in new contexts (networks, studied objects). We will explore the use of transfer learning algorithms [Fawaz et al., 2018c] to achieve this goal. The third area of ​​research aims to exploit the information used by automatic classification algorithms to arrive at a better understanding of the physics of seismic sources. The generated classification models will be explored to identify the properties of the signals (waveform, frequency, polarization, etc.) that contribute the most to obtaining a good classification. By looking at these properties, it is possible to identify links with the physics of the source. The object of this line of research will not be to propose a detailed characterization of these links for all the contexts studied, but rather a methodology to find this information, which can then be used in future studies, in particular to develop and calibrate physical models. These lines of research will be developed initially on data sets acquired on landslides monitored within the framework of the Multidisciplinary Observatory of Slope Instabilities (OMIV). The richness of the seismicity endogenous to these landslides, the multiplicity of sites (Viella, Super-Sauze, La Clapière, Séchilienne, Aiguilles), and the existence of catalogs comprising up to hundreds of thousands of events constitute a perfect opportunity to test the approaches we propose to develop.

More Information

Additional comments

This thesis is carried out as part of the ANR HighLand project. The application must be made via the job portal. Please attach a cover letter, a CV and the name of a reference who may be contacted for a letter of recommendation.

Qualifications: Engineering degree or Master in Earth Sciences / Geophysics or Computer Science and Data Sciences, with expertise in signal processing and an interest in seismology, and imperatively knowledge in programming (Python would be a plus).

Web site for additional job details

Required Research Experiences

  • RESEARCH FIELD
    Geosciences
  • YEARS OF RESEARCH EXPERIENCE
    None
  • RESEARCH FIELD
    Astronomy
  • YEARS OF RESEARCH EXPERIENCE
    None
  • RESEARCH FIELD
    Environmental science
  • YEARS OF RESEARCH EXPERIENCE
    None

Offer Requirements

  • REQUIRED EDUCATION LEVEL
    Geosciences: Master Degree or equivalent
    Astronomy: Master Degree or equivalent
    Environmental science: Master Degree or equivalent
  • REQUIRED LANGUAGES
    FRENCH: Basic
Work location(s)
1 position(s) available at
Institut Terre Environnement Strasbourg
France
STRASBOURG

EURAXESS offer ID: 703020
Posting organisation offer ID: 24857

Disclaimer:

The responsibility for the jobs published on this website, including the job description, lies entirely with the publishing institutions. The application is handled uniquely by the employer, who is also fully responsible for the recruitment and selection processes.

 

Please contact support@euraxess.org if you wish to download all jobs in XML.