Partager cette page

FAULTS_R_GEMS_pilot

Properties of FAULTS, Key Information to Produce Realistic Generic Earthquake Models and Hazard Simulations (Pilot, starting phase)

- I. Manighetti -

Earthquakes, the fault is in the faults…

Automated fault identification and mapping with deep learning (L. Mattéo PhD thesis, 2019) (left: raw optical drone image of a network of small natural faults in Arizona, faults as grey lines; right: automatic deep learning mapping faults as red lines) Automated fault identification and mapping with deep learning (L. Mattéo PhD thesis, 2019) (left: raw optical drone image of a network of small natural faults in Arizona, faults as grey lines; right: automatic deep learning mapping faults as red lines)

Academy 3 highlight

Large earthquakes are one of the most damaging and least predictable natural phenomena on Earth. Academy 3 encourages projects that might contribute to a better understanding of this natural hazard. To reach this objective, FAULTS_R_GEMS^PILOT fosters new collaborations in UCA among geologists, geophysicists, data science and image processing experts, and mathematicians.

The project

Tectonic faults are large “fracture planes” cutting through the Earth’s crust from the ground surface down to a few (1-10) kilometers depth (general case) or even more. At depth in the crust, faults cannot be observed directly and hence are difficult to analyze. By contrast, as their plane generally intersects the Earth surface, they form clear traces in the ground, generally of 1-1000 kilometers length, which can be observed on the field or from airborne and satellite sensors.

Faults are the Earth features that produce earthquakes; an earthquake is the sudden displacement along a fault of the two rock blocks it separates. Yet, surprisingly, although faults are the source of earthquakes, current earthquake models hardly consider them: the fault source is simplified as a 2D, planar, homogeneous surface with no thickness, embedded in an homogeneous, unfractured medium. This dramatically differs from reality since natural faults are non-planar, 3D zones embedded in intensely fractured rock volumes.

The Faults_R_Gems^Pilot project proposes that intimate relations exist between faults and earthquakes, which implies that earthquake modeling cannot be successful until it properly integrates fault properties. We have two objectives: firstly, identify and quantify the fault properties which most strongly impact the earthquake behavior; secondly, develop a new generation of earthquake models that integrate these fault properties.

During the 3 years of the Faults_R_Gems^Pilot project, we have worked on the first objective. Meanwhile, the second task has started in the framework of the larger project Faults_R_Gems since funded by the French National Agency ANR.

To address the first objective, we have 1) acquired large volumes of field, airborne and satellite imaging and topographic data documenting multiple natural fault zones of various scales and earthquake behaviors; 2) adapted deep learning tools to automate the identification and measure of the faults in the image and topographic data; 3) started the analysis of some of the fault properties in the measurements.

Among preliminary results, the new deep learning tools seem efficient at automatic fault identification in image data (Fig.1); scaling relations exist between important constitutive fault elements (such as fault length and thickness, Fig.2), proving the genetic and generic link between those elements. These scaling relations, key to understand the fault physics, will feed physical and mathematical fault models developed by our partners from LJAD laboratory (project DEFIDEF). Ultimately, these fault models will be integrated into the new generation of earthquake models we are developing.

Scaling relation between fault length and across-fault width of fractured zone off the fault (data are measures on natural faults, T. Giampietro PhD thesis, 2019) Scaling relation between fault length and across-fault width of fractured zone off the fault (data are measures on natural faults, T. Giampietro PhD thesis, 2019)

The +

Deep learning tools are adapted for the first time to automate identification, mapping and measure of faults in remote sensing and airborne data. This is also the first time that actual generic fault properties will be integrated into earthquake models. This new approach, coupling faults and earthquakes intimately, should decrease discrepancies between model outputs and natural earthquake behaviors.

What’s next?

If the results we obtain in the next few years (Faults_R_Gems ANR project) confirm that fault properties are primarily generic, their integration should push forward generic earthquake models that could be run for the vast majority of faults and earthquakes worldwide. These new models might thus open a novel avenue in earthquake modeling and hazard assessment.

Project information

Scientific domain Earth Sciences Theme Earthquake Hazard	Key words Faults Earthquakes Remote Sensing Deep Learning Scaling Relations Fault and Earthquake Modeling
Students inolved Mattéo Lionel (Phd, Géoazur-Inria-TAS) Mercier Antoine (M2R, Géoazur) Serreau Laure (M2r, Géoazur) Giampietro Tiziano (Phd, Géoazur) Vincendeau Alexandre (M1, Géoazur)	Total budget about 957 k€ including : ~ 500 k€ from ANR, ~ 100 k€ from Isis program for satellite data acquisition, 3 2 k€ from CNRS interdisciplinary mission, ~ 300 k€ from Ministry of Research, OCA, TAS and UCA funding for 2 PhDs and 1 post-doc Budget Academie 3 : 37 k€
Partner laboratories INRIA, UCA LJAD, UCA Thales Alenia Space (TAS) CNES Geosciences Montpellier International collaboration Arizona State University (Chelsea SCOTT, Ramon ARROWSMITH, Tyler SCOTT) UNAVCO, USA (Christopher CROSBY) Université de Pise (Giancarlo MOLLI)	Project members Ampuero Jean-Paul, Géoazur, UCA Auroux Didier, LJAD, UCA Berolo Wanda, Géoazur, UCA Bochra Mejri, LJAD, UCA Boese Maren, ETH Cappa Frédéric, Géoazur, UCA Delvit Jean-Marc, CNES Dominguez Stéphane, Géosciences Montpellier Gaucel Jean-Michel, Thales Alenia Space Gaudemer Yves, Institut De Physique Du Globe De Paris Hutchison Allie, Géoazur-Eth Leclerc Frédérique, Géoazur, UCA Malavieille Jacques, Géosciences Montpellier Manighetti Isabelle, Géoazur, UCA Mathieu Sandrine, Thales Alenia Space Pantz Olivier, Ljad, UCA Perrin Clément, Institut De Physique Du Globe De Paris Tarabalka Yuliya, Inria Sophia (UCA) And Luxcarta Zerubia Josiane, Inria Sophia, UCA

Isabelle Manighetti

GEOAZUR, Université Côte d’Azur

Project valorization

Publications:

Mattéo L., Tarabalka Y., and Manighetti I. (2018). Digital surface model generation from multiple optical high-resolution satellite images. In Image and Signal Processing for Remote Sensing XXIV (Vol. 10789, p. 107890W). International Society for Optics and Photonics.
Hutchison, A. A., Böse, M., & Manighetti, I. Improving early estimates of large earthquake's final fault lengths and magnitudes leveraging source fault structural maturity information. Geophysical Research Letters, e2020GL087539.
Jafrasteh, B., Manighetti, I., & Zerubia, J. (2020). Generative adversarial networks as a novel approach for tectonic fault and fracture extraction in high resolution satellite and airborne optical images. ISPRS-International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 43, 1219-1227.
Lionel Mattéo, Isabelle Manighetti, Yuliya Tarabalka, Jean-Michel Gaucel, Martijn Van Den Ende, Antoine Mercier, Onur Tasar, Nicolas Girard, Frédérique Leclerc, Tiziano Giampetro, Stéphane Dominguez, and Jacques Malavieille, Automatic fault mapping in remote optical images and topographic data with deep learning, in press at Journal of Geophysical Research - Solid Earth, March 2021.

International conferences:

Lionel Mattéo, Isabelle Manighetti, Onur Tasar, Frédérique Leclerc, Nicolas Girard, Yuliya Tarabalka, Tiziano Giampietro, Chelsea Scott, Stéphane Dominguez, Martijn van den Ende, Ramon Arrowsmith, Alexandre Vincendeau, Tyler Scott, and Jacques Malavieille, Automatic mapping of fractures and faults in optical images and Digital Surface Models with deep learning, AGU Fall meeting, Dec 2019, San Francisco
Allie Hutchison, Maren Böse, Frédérik Massin, Isabelle Manighetti, Incorporating fault maturity-related earthquake slip characteristics into final rupture length determination for Earthquake Early Warning, AGU Fall meeting, Dec 2019, San Francisco
Chelsea Scott, Tiziano Giampietro, Lionel Mattéo, Cassandra Brigham, Frédérique Leclerc, Ramon Arrowsmith, Isabelle Manighetti, Automatic measurement of scarp height along normal faults from high resolution topography in the Volcanic Tablelands, California, Geological Society of America meeting, Sept. 2019
Giampietro T., Molli G., Malavieille J., Manighetti I., Leclerc F., Antinori A., Structural architecture and damage zones in relay zones of normal fault systems: the case study of Vecchiano (Pisa) and experimental modelling, CRUST meeting, Perugia, Italy, July 2019.
Manighetti I., Leclerc, F., Serreau, L., Mattéo, L., Scott, C., Scott, T., ... & Crosby, C. J. (2018, December). Tectonic Damage Zones: How Large Across Faults? Insight From Empirical Scaling Relations in Granite Dells Fault Zone, Arizona, USA. In AGU Fall Meeting Abstracts.
Giampietro, T., Gaudemer, Y., Manighetti, I., Leclerc, F., Billant, J., Dominguez, S., ... & Crosby, C. J. (2020, December). Relative Slip in Damage Zones: Insights from Bishop Tuff Normal Faults. In AGU Fall Meeting 2020. AGU.
Jafrasteh, B., Manighetti, I., & Zerubia, J. (2020). GAN and U-net abilities to automate tectonic fault mapping in remote sensing optical images. In AGU Fall Meeting 2020. AGU

Diffusion :