Speaker(s): Paul Baville

Date: Thursday 15th of October 2020, 1:00 pm.


Assisted well correlation aims at complementing sedimentological expertise with computational rigor to increase automation, improve reproducibility and assess uncertainties during stratigraphic correlation. We propose a computer-assisted method which automatically generates possible well correlations based on facies interpretation, dipmeter data and knowledge about depositional environments.

This method uses facies interpretations and progading or backstepping trends deduced from the vertical stacking of depositional environments. These data are translated into a paleo-geographic variable inferred from depositional environments, e.g. the position along a proximal-to-distal transect. Assuming that wells have a global distality due to their position with respect to the overall basin geometry within the considered stratigraphic interval, we can interpolate a three-dimensional surface constrained by well-markers and dipmeter data acquired along wells. These surfaces represent chronostratigraphic surfaces. In a first approximation, the depositional dip direction is assumed to parallel sediment transport direction and the depositional strike direction being at a right angle to the former.

Well correlations are computed using correlation costs between all possible marker combinations aggregated by the Dynamic Time Warping algorithm. These correlation costs are based on the shape of the relative paleo-topography. Additionally, proximal facies interpreted in a distal well cannot be associated with distal facies interpreted in a proximal well, and conversely distal facies interpreted in a distal well may be likely associated with a proximal facies interpreted in a proximal well. Along the depositional strike, the method tries to associate identical or close facies with respect to distality.

Speaker(s): Zoé Renat

Date: Thursday 01st of October 2020 - 01:00 pm.


Time reversal allows us to locate earthquakes by back-propagating seismic waveforms recorded at a set of receivers in depth. Doing so, the seismic energy focuses, pointing out the actual seismic source. Interestingly, this method can handle noisy recordings and complex geological settings.

From a theoretical point of view, it requires the receivers to form a closed surface at depth, known as time-reversal mirror. Obviously, this condition cannot be satisfied in practice, but this is not our purpose to discuss the limitation here. Placing ourselves in the ideal framework, we implement a perfect time-reversal mirror to study the effect of the two force terms which generate the backpropagated field. One force-term is the traction at the mirror; the other is a dipole derived from the displacement at the mirror. The implementation of these two terms is performed in SpecFEM2d.

We show that the two force-terms are not necessary to generate the wavefield. With only one, the wavefield still focuses but also moves apart outside the closed surface. This result is encouraging because the traction is difficult to measure in practice.

Speaker(s): Pauline Collon

Date: Thursday 24th of September  2020 - 01:00 pm.


Il y a quelques années nous proposions d'étudier les réseaux kartsiques en les considérant comme des graphes (Collon et al., 2017, dans Geomorphology). Cette approche nous permettait d'étendre les classiques analyses géométriques à des analyses topologiques des réseaux en s'appuyant sur des indicateurs bien connus issus de la théorie des graphes. En effectuant ces calculs sur une 30aine de réseaux réels, nous fournissions ainsi des valeurs de référence et identifions les métriques les plus pertinentes.

Afin de permettre au plus grand nombre d'utiliser ces travaux et d'étendre les gammes de valeurs réalistes, nous avons depuis développé Karstnet : utilisant la bibliothèque NetworkX, ce petit programme python permet en quelques clics d'obtenir une analyse complète d'un réseau karstique tel que nous la proposions dans l'article.
Il est disponible sur Github en open-source avec sa batterie de tests unitaires et ses notebooks.

==> Ce séminaire est donc l'occasion de revenir sur cette recherche en vous (re)présentant les concepts, l'outil, et tout ce qui s'ouvre désormais au développement collaboratif grâce à ce projet.

Speaker(s): Francois Bonneau

Date: Thursday 20th of August, 2:00 pm.


Fracture networks (FN) are systems of complex mechanical discontinuities, which dramatically impact the physical behavior of rocks. Their statistical characterization is an important first step of stochastic modeling. It is, however, a big challenge because field data are sparse and incomplete and present several biases due to sampling (censoring, truncation, orientation).

The present paper concentrates on the statistical analysis of outcrops, which often may be considered as planar sections through three-dimensional FN. For the corresponding planar FN there exist well elaborated statistical methods, which yield first-order characteristics such as fracture density or fracture length distributions. Using ideas from stochastic geometry, in particular the theory of fiber processes and marked point processes, we develop second-order characteristics such as pair correlation function and mark correlation functions, which describe the variability of planar FN and their inner spatial correlations. Surprisingly, one of these characteristics is closely related to characteristics used in statistics of fractals applied to FN.

We demonstrate the application of our ideas by field outcrops already published in the literature.

Speaker(s): Corentin Gouache

Date: Thursday 09th July 2020 - 02:00 pm


Most of the seismic active regions are localised along plate boundaries, thus spatial distribution of earthquakes is also clustered along these boundaries. Moreover with high accumulation rates (>1cm/y) one can expect the area of the next large earthquakes. These findings aren't true for Stable Continental Region (SCR) like French mainland. In these territories, seismic events seem to be uniformly distributed in space. Furthermore, some observations express a dependency behaviour between SCR earthquakes, as in seismic sequences but over hundreds of years. These findings are taken in account into the Generator of Earthquakes for Barely seismic areas.

Speaker(s): Paul Baville & Capucine Legentil

Date: Wednesday 01st July 2020 - 11:00 am


We will briefly present how Bezier interpolator works and its utility to generate chronostratigraphic lines (2D) or surfaces (3D) from well markers (dip and strike data). Once these lines generated, it is possible to extrapolate them using a signed distance function within a 2D meshed model (3D is the next step).

Speaker(s): Guillaume Caumon

Date: Thursday 18th June 2020 – 11:00 am


In this talk, I briefly review some key machine learning principles and report on some recent work using convolutional networks for the structural interpretation of seismic images.

Speaker(s): Capucine Legentil

Date: Thursday 11th June 2020 – 11:00 am


This seminar is the occasion to present my PhD work on a method to robustly introduce boundaries in a triangulated geomodel, that allows the integration of new data, simulation results or geometry perturbation to reflect the subsurface uncertainties. The 2D geological model is locally updated, meaning that only a given region is modified and that the rest of the model remains identical. The area that can be modified is either specified as an input parameter or defined automatically. The input data is a triangulated surface storing the geological structure and physical properties. I will focus on the insertion of a horizon implicitly defined by an iso-value of a scalar field. The output is an updated mesh, which contains the new horizon. Distinctly from current model modification, the modifications are made to the mesh and aim to keep it valid throughout the transformations. The representation of geological structures as well as the computational support (mesh) can impact physical simulations. Once the local remeshing has been made, the impact of these updates on the model's physical behavior is evaluated thanks to the comparison of wave propagation profiles. These profiles are simulated using a Discontinuous Galerkin method.

Speaker(s): Marcus Apel

Date: Thursday 04th June 2020 – 2 :00 pm


In the Spring of 2018, a group of leading oil and gas companies met to discuss how contemporary cloud technology could be used to transform the current complex data and application environment. The Open Subsurface Data Universe™ (OSDU) Forum, a Forum of The Open Group, was created with the objective of enabling new cloud-native data-driven applications with seamless access to the full range of subsurface and wells data as well as supporting existing applications and data frameworks. The core principle of the OSDU solution is to separate data from applications. This will be achieved by developing a common data platform with standard public APIs and to involve global cloud hosting vendors to build working implementations. The focus of this presentation will be on data models and data architecture for geomodelling.