TS9.2

In geological settings characterised by folded and faulted strata, and where good field data exist, we have been able to automate a large part of the 3D modelling process directly from the raw geological database (maps, bedding orientations and drillhole data). The automation is based upon the deconstruction of the geological maps and databases into positional, gradient and spatial and temporal topology information, and the combination of deconstructed data into augmented inputs for 3D geological modelling systems, notably LoopStructural and GemPy.

When we try to apply this approach to more complex terranes, such as greenstone belts, we come across two types of problem:

• 1) Insufficient structural data, since the more complexly deformed the geology, the more we need to rely on secondary structural information, such as fold axial traces and vergence to ‘solve’ the structures. Unfortunately these types of data are not always stored in national geological databases. One approach to overcoming this is to analyse the simpler (i.e. bedding) data to try and estimate the secondary information automatically.

• 2) The available information is unsuited to the logic of the modelling system. Most modern modelling platforms assume the knowledge of a chronostratigraphic hierarchy, however, especially in more complexly deformed regions, a lithostratigraphy may be all that is available. Again a pre-processing of the map and stratigraphic information may be possible to overcome this problem.

This presentation will highlight the progress that has been made, as well as the road-blocks to universal automated 3D geological model construction.

We acknowledge the support of the MinEx CRC and the Loop: Enabling Stochastic 3D Geological Modelling (LP170100985) consortia. The work has been supported by the Mineral Exploration Cooperative Research Centre whose activities are funded by the Australian Government's Cooperative Research Centre Programme. This is MinEx CRC Document 2020/xxx.

How to cite: Jessell, M.: Current and future limits to automated 3D geological model construction, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-632, https://doi.org/10.5194/egusphere-egu21-632, 2021.

In September 2020, the Corona crisis offered us an opportunity to develop and test a blended real and virtual interdisciplinary field mapping class, as well as revealing the need for, and stimulating development of new web-based tools for structural interpretation.

Universität Mainz’ usual Master’s advanced field mapping, and Universität Tübingen’s usual Bachelor’s mapping classes were replaced with combinations of (i) virtual field mapping of Jurassic-Cretaceous sedimentary units at Molinos, Teruel Province, Spain, and (ii) field mapping of metamorphic rocks in the Mittelrhein Gorge and the Arh Valley, and outcrops of sedimentary rocks near Tübingen, Germany, which the students were mostly able to access on day trips using public transport or by bicycle.

For the Molinos part of the exercise both groups were offered hand specimens containing distinctive fossils, linked to locations (and pseudo-locations) by google .kmz files, a variety of structural measurements also linked via .kmz files, and detailed satellite imagery within which mappable geological units display distinct characteristics. Introductions to the stratigraphy were made in three virtual outcrop sections examined in Google Street View from within Google Earth, and via web-based photogrammetric 3D outcrop models made available on the V3Geo virtual 3D geoscience platform. The students then extrapolated this stratigraphy based on the satellite imagery and .kmz file information.

Our perception, validated by student feedback, is that the real parts of both field excursions were very important since they allowed us to teach and refine mapping and compass methodology and best demonstrate how to analyze 3D geometries of geological structures. Universität Mainz students particularly benefited from being able to visit locations where we had already made 3D outcrop models and offered a digital excursion, in the Ahr Valley (Rhenish Massif). They were able to compare real structural measurements with those derived from the precisely georeferenced 3D models, which enhanced their ability to subsequently obtain such information solely from the models. Although final student maps were of comparable quality to those produced in the field, structural interpretations were hampered by a lack of field measurements. In many cases, the Google Earth DEM is of too low resolution and ways should be found to include higher-resolution DEMs in web-based data sets.

Overall, we think there were advantages compared to traditional field mapping, such as (i) enhanced evidence that methods like ‘structure contouring’ were used in all mapping, (ii) we were stimulated to teach the students to use digital methods to acquire field data, such as StraboSpot and Stereonet11 Apps. We observed these tools, and others we were unaware of, being used in combination with traditional paper and compass during the real mapping exercise. We hope to continue to employ this blended teaching approach even when the Corona crisis passes. This will be facilitated by our development of further 3D outcrop models, .kmz files with key information about outcrops in the Mittelrhein, and especially, web-based (rather than PC-based) tools to extract structural data such as plane and line orientations from 3D outcrop models and enable collaborative work on one data set.

How to cite: Toy, V., Abe, S., Bons, P., Buckley, S. J., Deckert, H., Fenske, S., Kirilova, M., Lewis, C., Mutz, S., Owin, J., Sachau, T., Schuck, B., and Seelos, P.: A blended learning approach to structural field mapping: combining local geology, virtual geology, and web-based tools, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-3581, https://doi.org/10.5194/egusphere-egu21-3581, 2021.

A new approach for constrained 3-D structural geological modelling using Graph Neural Networks (GNN) has been developed that is driven by a learning through training paradigm. Graph neural networks are an emerging deep learning model for graph structured data which can produce vector embeddings of graph elements including nodes, edges, and graphs themselves, useful for various learning objectives. In this work our graphs represent unstructured volumetric meshes. Our developed GNN architecture can generate spatially interpolated implicit scalar fields and discrete geological unit predictions on graph nodes (e.g. mesh vertices) to construct 3-D structural models. Interpolations are constrained by scattered point data sampling geological units, interfaces, as well as linear and planar orientation measurements. Interpolation constraints are incorporated into the neural architecture using loss functions associated with each constraint type that measure the error between the network’s predictions and data observations. This presentation will describe key concepts involved within this approach including vector embeddings, spatial-based convolutions on graphs, and loss functions for structural geological features. In addition, several modelling results will be given that demonstrate the capabilities and potential of GNNs for representing geological structures.

How to cite: Hillier, M., Wellmann, F., Brodaric, B., de Kemp, E., and Schetselaar, E.: Using Graph Neural Networks for 3-D Structural Geological Modelling, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-12978, https://doi.org/10.5194/egusphere-egu21-12978, 2021.

Structural geological models are often calculated on a specific spatial resolution – for example in the form of grid representations, or when surfaces are extracted from implicit fields. However, the structural inventory in these models is limited by the underlying mathematical formulations. It is therefore logical that, above a certain resolution, no additional information is added to the representation.

We evaluate here if Deep Neural Networks can be trained to obtain a high-resolution representation based on a low-resolution structural model, at different levels of resolution. More specifically, we test the use of state-of-the-art Generative Adversarial Networks (GAN’s) for image superresolution in the context of 2-D geological model sections. These techniques aim to learn the hidden structure or information in high resolution image data set and then reproduce highly detailed and super resolved image from its low resolution counterpart. We propose the use of Generative Adversarial Networks GANS for super resolution of geological images and 2D geological models represented as images. In this work a generative adversarial network called SRGAN has been used which uses a perceptual loss function consisting of an adversarial loss, mean squared error loss and content loss for photo realistic image super resolution. First results are promising, but challenges remain due to the different interpretation of color in images for which these GAN’s are typically used, whereas we are mostly interested in structures.

How to cite: Qazi, M. U. R. and Wellmann, F.: Super-resolution in Structural Geological Models., EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-10333, https://doi.org/10.5194/egusphere-egu21-10333, 2021.

The development of Carbon Capture Utilization and Storage (CCUS) technology paired with existing energy systems will facilitate a successful transition to a carbon-neutral economy that offers efficient and sustainable energy. It will also enable the survival of multiple and vital economic sectors of high-energy industries that possess few other options to decarbonize. Nowadays, just about one-ten-thousandth of the global annual emissions are being captured and geologically-stored, and therefore with today’s emission panorama, CCS large-scale deployment is more pressing than ever. In this study, a 3D model that represents the key reservoir uncertainties for a CCUS pilot was constructed to investigate the feasibility of CO2 storage in the Unayzah Formation in Saudi Arabia. The study site covers the area of the city of Riyadh and the Hawtah and Nuayyim Trends, which contain one of the most prolific petroleum-producing systems in the country. The Unayzah reservoir is highly stratified and it is subdivided into three compartments: the Unayzah C (Ghazal Member), the Unayzah B (Jawb Member), and the Unayzah A (Wudayhi and Tinat Members). This formation was deposited under a variety of environments, such as glaciofluvial, fluvial, eolian, and coastal plain. Facies probability trend maps and well log data were used to generate a facies model that accounted for the architecture, facies distribution, and lateral and vertical heterogeneity of this high complexity reservoir. Porosity and predicted permeability logs were used with Sequential Gaussian Simulation and co-kriging methods to construct the porosity and permeability models. The static model was then used for CO2 injection simulation purposes to understand the impact of the flow conduits, barriers, and baffles in CO2 flow in all dimensions. Similarly, the CO2 simulations allowed us to better understand the CO2 entrapment process and to estimate a more realistic and reliable CO2 storage capacity of the Unayzah reservoir in the area. To test the robustness of the model predictions, geological uncertainty quantification and a sensitivity analysis were run. Parameters such as porosity, permeability, pay thickness, anisotropy, and connectivity were analyzed as well as how various combinations between them affected the CO2 storage capacity, injectivity, and containment. This approach could improve the storage efficiency of CO2 exceeding 60%. The analyzed reservoir was found to be a promising storage site. The proposed workflow and findings of the static and dynamic modeling described in this publication could serve as a guideline methodology to test the feasibility of the imminent upcoming pilots and facilitate the large-scale deployment of this very promising technology.

How to cite: Mantilla Salas, S., Corrales, M., Hoteit, H., Alafifi, A., and Tasianas, A.: Quantifying Uncertainty through 3D Geological Modeling for Carbon Capture Utilization and Storage in the Unayzah Formation in Saudi Arabia, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-12839, https://doi.org/10.5194/egusphere-egu21-12839, 2021.

Numerical simulations of subsurface processes are essential to the success of many geoengineering projects. These simulations often contain significant uncertainties due to imperfect knowledge of material properties and their spatial distribution, boundary conditions, and initial conditions. However, efficient implementations for the quantification of uncertainties for such simulations are big challenges in Computational Geoscience, mainly due to the curse of dimensionality. Process simulations often involve solving high-dimensional Partial Differential Equations (PDE) by using discretization methods such as Finite Difference (FD) or Finite Elements (FE) methods. Although such methods often give good approximations, they are computationally intensive and expensive and therefore infeasible in the applications such as MCMC where thousands of evaluations of the forward simulation are required. Previous work by Degen et.al. (2020) has addressed this problem by using a model order reduction method, the so-called reduced basis (RB) method. However, the method has limitations when considering complex (i.e., hyperbolic and non-linear) PDEs. In this work, we aim to employ the recently developed Fourier Neural Operator (FNO) (Li, 2020) as a tool to implement efficient approximation of PDEs in the application of Geothermal reservoir simulation. FNO involves a Fast Fourier transform to directly learn the mapping from the input function to the output function. FNO has the advantage of being independent of the resolution and complexity of the governing PDE. Our preliminary results show that FNO can provide good approximation results in solving four-dimensional PDEs and thus can be used as a tool for further probability studies of the parameters of interest.

How to cite: Liang, Z., Degen, D., and Wellmann, F.: The Application of Neural Operator in subsurface process simulation, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-12940, https://doi.org/10.5194/egusphere-egu21-12940, 2021.

3D geological modelling of complex metamorphic terrains that underwent a sequence of ductile and brittle deformation events is an extremely challenging task. Difficulties start from the input data that are frequently sparse and heterogeneous in quality and distribution. In projects based on field data only (without significant subsurface data) uncertainties are even more pronounced, but, in our project, we had the rugged topography of the Western Alps on our side, with elevations ranging from c. 1200 m to c. 3200 m and very continuous outcrops. Other problems, that we address in this contribution, arise during the modelling process. We tested different commercial software packages and some open-source research libraries and we found that no one is capable of modelling our complex structures out-of-the-box. This is not surprising since generally these codes, and particularly the commercial ones, are geared towards modelling gently deformed sedimentary sequences. However, it is possible to overcome a large range of obstacles by “fooling” implicit structural modelling algorithms, simply “cheating” on the geological meaning of model entities. This means (1) developing a conceptual model of polyphase ductile and brittle deformation, (2) finding geological/mathematical entities that are at the same time implemented in the code and able to represent the complex structures, and finally (3) carrying out the implicit modelling. For instance, tectonic contacts between large-scale tectono-metamorphic units can be treated as unconformities (and not as faults) to obtain a realistic representation. In some cases, also conformal lithological boundaries can be considered as unconformities with the goal of allowing larger thickness variations. In other situations, a “fake” stratigraphy where the same units are repeated several times can be used to model sequences of isoclinal folds and thin tectonic slices. In this contribution, some of these modelling solutions are compared in terms of (1) their straightforward implementation, and (2) their ability to generate models that properly fit the very detailed geological maps available in our study area (c. 60 km² mapped at 1:5.000-1:10.000 with a dense set of structural stations).

How to cite: Arienti, G., Bistacchi, A., Monopoli, B., Dal Piaz, G. V., Dal Piaz, G., and Bertolo, D.: 3D geomodelling in the complex metamorphic and poly-deformed units of the Italian Western Alps (Conca di By, Aosta Valley, Italy), EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-15515, https://doi.org/10.5194/egusphere-egu21-15515, 2021.

Expectations for geological models for underground characterization rise with complex engineering tasks. In this project we examine a target area as a potential site for the gravitational-wave observatory “Einstein Telescope” in the Meuse-Rhine Euroregion (Netherlands, Belgium, Germany).  The Einstein Telescope will be the world’s most sensitive observatory of its kind. It consists of a triangular shaped facility connected by 10 km long arms in 200-300 m depth. A high accuracy 3-D structural geological model is required to constrain the best position of the Einstein Telescope with geophysical and geotechnical methods.

We use an implicit modeling approach based on surface points and orientation data for modeling. This data is extracted from seismic surveys and well logs available in the region. The application of probabilistic methods in this workflow allows to propagate uncertainty of the input data into a resulting model suite, allowing to define a measure of uncertainty for the final model. Specific local difficulties that were encountered during the modelling process, including data management, the representation of complex fault networks and scaling issues will be discussed.

We will show 3-D geological models for the Meuse-Rhine Eurogregion to significantly improve our geological understanding of the target area. This improved understanding is crucial for finding the optimal position for the Einstein Telescope. Data is managed using the open-source library GemGIS. Models are created using the open-source library GemPy.

How to cite: Chudalla, N., Wellmann, F., Jüstel, A., and von Harten, J.: Implicit geological modeling for the Einstein Telescope (Meuse-Rhine Euroregion), EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-15814, https://doi.org/10.5194/egusphere-egu21-15814, 2021.