Speaker: Giusi Ruggiero
Date: Friday 29th of September 2023, 1:15pm.
Abstract:
In this talk I review some basic notions and mathematical tools in the stochastic modeling of uncertainties and their quatification for large-scale computational models discussed during the 'Uncertainty Quantification Workshop' held at the Isaac Newton Institute of Mathematical Sciences at the University of Cambridge.
- Details
- Category: Seminar
Speaker: Rachit Gautam
Date: Thursday 21st of September 2023, 1:15pm.
Abstract:
Toba caldera, the Earth's largest Quaternary volcanic complex, is located in north Sumatra at the forearc of the Sumatra subduction zone, along the Great Sumatran fault and the Investigator Fracture Zone. Its latest and largest known eruption to date, known as Youngest Toba eruption, occured 74,000 years ago and opened the central caldera hosting Lake Toba. In this study, I analysed earthquake waveform (passive seismic) data recorded by 42 three-component seismometers located across the caldera to measure and map total attenuation, scattering attenuation and absorption attenuation separately in 3D space. The coda normalisation method was used to measure total attenuation, S-wave peak delay measurements was used to quantify scattering attenuation and late time coda quality factor as marker for absorption attenuation. I was able to successfully perform total attenuation and peak-delay (scattering) analysis which produced high resolution and highly detailed 3D tomographic maps of my study area up to the depth of $\sim$15 km. The results of my study show that the crust beneath Toba caldera and the area south of it is highly heterogeneous. Multiple areas exhibiting high attenuation anomaly could be located at different depths beneath the southern part of Toba which I interpret as Magma bodies or the reservoirs containing magma derived materials (magmatic fluids, gases, etc.).
- Details
- Category: Seminar
Speaker: Juan Sebastian Osorno Bolivar
Date: Thursday 20th of July 2023, 1:15pm.
Abstract:
The restoration techniques that provide means for structural geologists to better understand the tectonic history of present-day geometries have been developed in a geomechanically consistent approach for more than 50 years. These usually involve linear elastic material properties, although the small deformation assumption does not hold in most case studies. Nonlinear elastic and plastic rheologies can model large deformations but they do not yield a time reversible scheme because of energy dissipation phenomena which cannot be recovered at geological timescales. Viscous deformations of geological layers described by Stokes equations have been shown to circumvent these obstacles. The goal of the present paper is to develop an inversion scheme for the effective viscosity of the geological materials at play, relying on the exploration of the admissible space based on an objective function related to the horizontality of the top surface. To perform the inversion, methods that have a strong performance solving ill-conditioned problems are considered: Covariance Matrix Adaptation Evolution Strategy (CMA-ES), Particle Swarm Optimization (PSO), Simulated Annealing, Random Search and Grid Search, all available in the GNIR library (Mazuyer et al., 2018). We apply the proposed approach to a synthetic 2D model of a faulted graben generated in the FAIStokes program (Schuh-Senlis et al., 2020).
- Details
- Category: Seminar
Speaker: Guillaume Rongier
Date: Thursday 13th of July 2023, 1:15pm.
Abstract:
Recent developments in deep learning have turned the spotlight on the entire field of machine learning. This has led to more and more studies comparing predictions from geostatistical and machine learning methods, sparking debates about whether machine learning will take over geostatistics in geological applications. This implicitly implies an opposition between those two fields. In this context, a statement made by Williams in 1998 contains an intriguing perspective: "In the Bayesian approach to neural networks, a prior on the weights of a network induces a prior over functions. An alternative method of putting a prior over functions is to use a Gaussian process (GP) prior over functions. This idea has been used for a long time in the spatial statistics community under the name of 'kriging' although it seems to have been largely ignored as a general purpose regression method." If the method known as kriging in geostatistics is the same as the method known as Gaussian processes in machine learning, how is it that geostatistics and machine learning are so often introduced as contending fields? Can we find similar relationships with more recent geostatistical approaches such as multiple-point simulation? What does this mean for the future development of both fields? This talk will deconstruct the idea that geostatistics and machine learning are two completely separate fields: their methods share a lot more similarities than usually acknowledged, and geostatistics can be seen as a sub-field of machine learning specialized on predictions with spatial data. Since 1998, Gaussian processes have become an integral part of machine learning, and both fields stand to gain from further exploring their similarities.
- Details
- Category: Seminar
Speaker: Enrico Scarpa
Date: Thursday 6th of July 2023, 1:15pm.
Abstract:
This seminar will mainly consist of a review and explanation of some field observations depicted from well-exposed outcrops in deep-water systems. It’s based on several papers about turbidite channels and one by Bell, Kane, Pontén, Flint, Hodgson & Barrett (Marine and Petroleum Geology 2018), entitled: "Spatial variability in depositional reservoir quality of deep-water channel fill and lobe deposits ". The full paper is available at Marine and Petroleum Geology.
- Details
- Category: Seminar
Speaker: Iman Rahimzadeh Kivi
Date: Thursday 29th of June 2023, 1:15pm.
Abstract:
Widespread deployment of CO2 capture and storage (CCS) in deep geological formations is identified as an essential component of any efforts to mitigate the climate change crisis. Most assessments conclude that CO2 storage rates in the order of several gigatons per year will be needed to achieve carbon neutrality by 2050. However, concerns exist about the long-term fate of CO2 underground and the possibility of leakage back to the surface. I will present research that investigates what would happen to CO2 injected at climate-relevant rates of gigatons per year over geological time scales (million years), much longer than any assessments performed so far. I will discuss the subsurface CO2 dynamics that can be simplified without loss of generality to vertical CO2 flow and transport concerning long-term CO2 migration through geological layers. Such a model enables us to draw reliable constraints on the CO2 leakage potential over geological time scales at affordable computational costs. Simulation results show that repetitive layering of caprocks, even if pervasively fractured and a few tens of meters thick, will significantly reduce the CO2 leakage risk, ensuring a secure road toward achieving climate targets.
- Details
- Category: Seminar
Speaker: Eric Galin
Date: Thursday 22nd of June 2023, 1:15pm.
Abstract:
Modeling realistic and controllable large-scale landscapes is essential for creating virtual worlds. The challenge stems not only from the complexity of landforms, the variety of details and patterns at different scales, the need for geomorphological and hydrological realism, and the complexity of ecosystems but also from the need to control the shape and location of landforms to follow the designer’s intent. In this talk, I will present some recent research that aims at bridging the gap between simulation providing realism and interactive editing giving control.
- Details
- Category: Seminar
Speaker: Giusi Ruggiero
Date: Thursday 15th of June 2023, 1:15pm.
Abstract:
Uncertainty quantification in seismic imaging is important for a proper interpretation of the structural elements (e.g., faults and horizons) within the investigated subsurface. Especially in seismic full-waveform inversion (FWI), which is a highly non-linear problem and hence prone to non uniqueness, evaluate uncertainties associated with the estimated subsurface parameters is essential for interpreting inverted models. In this work, we first address uncertainty estimation in elastic FWI by calculating the posterior covariance matrix based on the data-misfit Hessian matrix. In particular, in order to make the computation tractable for large scale problems, we rely on a low rank approximation of the Hessian, which avoids the prohibitive computation of the full matrix. The resulting estimate of uncertainties will be used in a comparison between the homogenized FWI model and a set of homogenized geological models for seeking the best among multiple possible structural interpretations of a given seismic image.
- Details
- Category: Seminar