TS8.1 | 3-D Geological Models as Scientific Tools for Joint Inversion, Uncertainty Quantification, and Machine Learning
EDI
3-D Geological Models as Scientific Tools for Joint Inversion, Uncertainty Quantification, and Machine Learning
Convener: Florian Wellmann | Co-conveners: Peter Lelièvre, Sarah DevrieseECSECS, Colin Farquharson, Clare Bond
Orals
| Tue, 25 Apr, 08:30–10:15 (CEST)
 
Room K1
Posters on site
| Attendance Mon, 24 Apr, 16:15–18:00 (CEST)
 
Hall X2
Posters virtual
| Attendance Mon, 24 Apr, 16:15–18:00 (CEST)
 
vHall TS/EMRP
Orals |
Tue, 08:30
Mon, 16:15
Mon, 16:15
Geological models are key to our understanding of the subsurface by providing both visual and quantitative context. Accurately modelling the significant heterogeneities, discontinuities and uncertainties of geological systems from often sparse data remains challenging. Substantial developments in geomodelling over the past years has helped bridge the gap between input data and resulting geomodel, allowing for the (semi-)automated construction of geomodels, quicker model validation and rebuilding when new data arrives, and efficient testing of multiple hypotheses. Increasing computing power now also allows for effective stochastic simulation of uncertainties in geomodelling, integration of probabilistic inference frameworks, and the unification of geomodelling with geophysical modelling and inversion. Machine learning approaches can be used in every step of the geomodelling pipeline to enhance the process: from automated input data extraction and classification to probabilistic model selection.

We seek contributions from all geoscientists using 3-D geological modelling methods, or developing novel methods to construct such models. Of special interest are: 1) approaches to quantify and communicate uncertainties in the model building process; 2) attempts to make geophysical Earth models and geological models directly compatible and interoperable; and 3) work that combines and enhances geomodelling with machine learning methods. Applications can be in any field of solid Earth sciences to address scientific questions throughout the lithosphere/anthroposphere.

Orals: Tue, 25 Apr | Room K1

Chairpersons: Clare Bond, Peter Lelièvre, Florian Wellmann
08:30–08:35
08:35–08:45
|
EGU23-8856
|
ECS
|
On-site presentation
Gloria Arienti, Andrea Bistacchi, Guillaume Caumon, François Bonneau, Giorgio Vittorio Dal Piaz, Giovanni Dal Piaz, Bruno Monopoli, and Davide Bertolo

Three-dimensional structural modelling of complex metamorphic settings is an extremely challenging task. In these settings, rocks sequences record multiple ductile and brittle events, leading for instance to refolded fold structures, isoclinal folds and dense network of faults. In this contribution, we build a 3D structural model of a portion of the highly deformed core of the Alpine orogeny in the North-Western Italian Alps, by using field data (1:10,000 geological map and a dense database of structural stations) as unique input source. Our model area has an extension of ca. 1,300 km2 and a vertical elevation difference between the highest mountains (e.g., Cervino-Matterhorn) and the valley floors of ca. 4,000 meters, reflecting a truly three-dimensional dataset.

Our workflow expects a first phase of orientation statistics study of the structural field database, followed by structural interpretation in vertical cross-sections and 3D interpolation using implicit surfaces and structural constraints. The implicit approach allows us to propagate field data and geological interpretation through mathematical constraints and to obtain structural interfaces reflecting observations.

After introducing the new 3D structural model of the portion of the North-Western Alps, we discuss the difficulties related to geomodelling using input surface data only, by qualitatively addressing the uncertainty aspects of our workflow. We also focus on the range of geological and structural constraints that fieldwork allows us, reasoning on the distinction between observed and interpreted geological information.

How to cite: Arienti, G., Bistacchi, A., Caumon, G., Bonneau, F., Dal Piaz, G. V., Dal Piaz, G., Monopoli, B., and Bertolo, D.: Three-dimensional modelling of a complex metamorphic nappe stack from field survey only: the case study of the Aosta Valley (Italian NW-Alps), EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8856, https://doi.org/10.5194/egusphere-egu23-8856, 2023.

08:45–08:55
|
EGU23-9058
|
On-site presentation
Madeline Lee, Suzanne McEnroe, Zeudia Pastore, and Nathan Church

In this work, we conduct 3-D geologic modelling of the Bjerkreim-Sokndal (BKS) layered intrusion's Bjerkreim lobe in southern Norway. The BKS is a folded, double-plunging syncline with an areal extent of 230km2. There are five rhythmic megacyclic lithological units (MCU, I - V) that are divided into zones (a - f) based on the presence or absence of index minerals. The BKS is often used as an analogue for Martian studies due to the presence of strong magnetic remanence, 20 000 nT below background. It has also undergone significant exploration for critical minerals.

Although the BKS is well-studied, there are limitations that have hindered geophysical mapping. The layered units reside along a topographic low, limiting the lowest altitude for airborne surveys. Land use is classified for agriculture and the presence of lakes restrict ground data collection to roads and pathways. Seismic and gravity surveys have been collected over the study area; however, the gravity data is sparse, and the seismic data is restricted to a single profile. These limited studies suggest a BKS depth to base of 4 - 5 km. Drillcore has been collected, however these extend on average to a depth of 30 m. In 2021, a small-scale drone magnetic survey was collected. This survey was to complete low-altitude, high resolution magnetic sampling to complement previous ground magnetic surveys and as a segue for multiscale analysis with regional airborne surveys. Since no single geophysical dataset is sufficient for a complete geologic interpretation, joint modelling is required to better understand subsurface distribution.

A master ground sample database was compiled of over 3000 samples and 11 petrophysical properties. This database consisted of in-situ and in-lab measurements. Principal Component Analysis was conducted to reduce dimensionality and identify which properties should be incorporated into the model. K-means clustering was conducted to identify natural groupings in the data, where average values for these properties were calculated from the dominant cluster. 2-D profiles orthogonal to strike were constructed at 2 km increments along the Bjerkreim lobe with additional intermediate profiles to minimize truncation of 3-D volumes. A combination of 2-D forward and joint inversion modelling was implemented using compiled aeromagnetics and gravity as the observed values. Each MCU zone was modelled as a block with average properties from the cluster analysis and constrained at surface by mapped contacts. Depth estimation routines, including Euler deconvolution, were also executed. The modelled blocks were then wireframed into volumes to create a 3-D representation of the Bjerkreim lobe.

How to cite: Lee, M., McEnroe, S., Pastore, Z., and Church, N.: 3-D Geologic Modelling of the Bjerkreim Lobe, Norway, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9058, https://doi.org/10.5194/egusphere-egu23-9058, 2023.

08:55–09:05
|
EGU23-1860
|
Virtual presentation
Denis Anikiev, Hans-Jürgen Götze, Christian Plonka, Sabine Schmidt, Judith Bott, and Magdalena Scheck-Wenderoth

Modern workflows for construction of 3-D data-constrained Earth’s subsurface models in complex geological environments require sophisticated research software tools capable of handling interdisciplinary data and analysis in both visual and quantitative context. Integration of potential field data – gravity and magnetics – into the model building process is a key component that helps to bridge the gaps in the sparse input data by fitting the modelled response to the measurements. On the basis of IGMAS+ – a free cross-platform potential field modelling software – we show how 3-D model building can be complemented by interactive optimisation (inversion) of the triangulated subsurface model geometry. The optimisation is done by means of Covariance-Matrix-Adaptation Evolution Strategy (CMAES) which proved to be efficient for strongly non-linear problems with high-dimensional parameter space. In order to avoid topology distortions of the triangulated model domain, we use a concept of warping the space containing a model, rather than operating on the model vertices. The space warping implies an elegant solution using a system of virtual elastic springs connecting the lattice nodes. The optimisation workflow is demonstrated on synthetic and real case studies. We also show how an interpreter can interact with the process: visually control and influence the quality of the optimisation on a timeline. The proposed workflow is an efficient tool for automated quick model construction, validation and rebuilding, as well as for testing of multiple modelling hypotheses.

How to cite: Anikiev, D., Götze, H.-J., Plonka, C., Schmidt, S., Bott, J., and Scheck-Wenderoth, M.: Interactive optimisation of 3-D subsurface models using potential fields, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-1860, https://doi.org/10.5194/egusphere-egu23-1860, 2023.

09:05–09:15
|
EGU23-5416
|
On-site presentation
Ali Dashti, Jens C. Grimmer, Christophe Geuzaine, and Thomas Kohl

High-Temperature Aquifer Thermal Energy Storage (HT-ATES) manages the temporal mismatch between heat supply and demand periods. Up to 50 % of consumed energy by residents of metropolitans can be provided through this huge underground battery systems. This study evaluates risks and effects of geological structures for two HT-ATES candidates designed close to populated areas in central Europe. DeepStor, as the first example, is designed to store surplus heat in Meletta beds beneath the campus of the Karlsruhe Institute of Technology (KIT). A synthetic sealing fault is embedded in real topology of the Meletta beds to numerically simulate the temperature and pressure under such structural feature. The synthetic fault is relocated 16 times in the geological model and proved to only increase the pressure value up to 7 % in comparison to fault-free (base case) realization. The real tilted morphology of Meletta beds revealed that hot temperature tends to accumulate in the western side of the model while pressure increase is more notable in the reverse side, i.e. down dip. A simple function fitted to the pressure change and fault to well distance shows acceptable levels of reliability. Another showcase designed for the Greater Geneva Basin confirmed the insensitivity of the temperature and pressure to surfaces morphology of the Malm reservoir with 100 m thickness. Despite modulating the top and bottom contacts of the Malm from flat planes to randomly rugged surfaces, the results remain the same. The upper and lower surfaces are moved ± 8 and ± 10 m, respectively. This insensitivity indicates the local natures of the induced thermal regime in thick reservoirs and dispensability of some expensive exploration campaigns like 3D seismic.

How to cite: Dashti, A., Grimmer, J. C., Geuzaine, C., and Kohl, T.: Modeling two high-temperature aquifer thermal energy storage cases under uncertain geological frameworks, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5416, https://doi.org/10.5194/egusphere-egu23-5416, 2023.

09:15–09:25
|
EGU23-2380
|
Virtual presentation
Roland Martin, Vitaliy Ogarko, Jérémie Giraud, Bastien Plazolles, Sonia Rousse, and Paul Angrand

Gravity inversion methods are able to recover density distributions in the Earth but they need to be strongly constrained using a variety of prior information. Here, we aim at inverting gravity data anomalies constrained by existing geological and density information on orogenic areas such as the Pyrenees where many geological and geophysical studies have been conducted for geophysical exploration purposes and fluid resources recovering of economic interest.

To perform such inversion, we aim at constraining gravity inversions using covariance matrix defined as an interval distribution of possible density values. This covariance-like matrix is obtained by computing the probability of impact of lithological density variations on gravity residuals. Instead of using a Monte Carlo-like approach to sample density values in each rock unit, which may be too computationally expensive (in terms of number of forward calculations, memory and disk storage of all data needed for the probabilistic analysis), we calculate a series of probabilistic metrics associated to different combinations of density variations. For this, we select representative model variations and use partial plane experiment-based probabilistic method approach to estimate the impact of density variations on gravity data misfit. This drastically reduces the number of calculations and requires only a few tens of forward problems evaluations (instead of hundreds or thousands with Monte Carlo-like approach). Based on the impact of each prior lithological density variation, intervals of density variations can thus be estimated for each rock unit. This approach allows to define at low cost all these intervals, which can be interpreted as a reduced covariance matrix. For inversion using these intervals as constraints, we use an initial a priori density model obtained from a prior Vp model obtained by seismic teleseismic time-arrival inversion. To reconcile the so-obtained density model with gravity data, we perform gravity inversion constrained by bounded density intervals estimated from the probabilistic approach we propose. A dynamic Alternate Direction Multipliers Method regularization approach is used to constrain the inversion over such variation intervals. This allows us to obtain inverted models consistent with the geological structures modelled in the area and gravity data.

We apply this inversion technique to the whole Pyrenees chain (southwest Europe) at a 2 km resolution and on a smaller zoomed 1 km resolution area constrained by outer information (density, ADMM variation intervals, …) provided by the 2km coarser inverted model. This way, new geological features can be inferred in the collisional intraplate Iberian-Eurasian region, in the axial zone and basement, and also at depth until the upper mantle. Besides, strong excesses of mass in the northern part and strong negative density contrasts in the south of the Pyrenees are appearing and increasing with depth when compared to previous prior models.

How to cite: Martin, R., Ogarko, V., Giraud, J., Plazolles, B., Rousse, S., and Angrand, P.: 3D gravity data inversion constrained by bounding interval ADMM regularization : Application to density distribution reconstruction of the Pyrenees chain lithosphere., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2380, https://doi.org/10.5194/egusphere-egu23-2380, 2023.

09:25–09:35
|
EGU23-14866
|
On-site presentation
Ane Elisabet Lothe, Arnt Grøver, and Ole-Andre Roli

Understanding the sealing capabilities of faults provide vital information for underground carbon dioxide storage to hydrocarbon exploration and production. The sealing properties of faults are dependent on several parameters controlled by lithologies, overlap, throw, fault width, burial history, thermal regime and diagenesis in the sedimentary basin. All these input parameters hold large uncertainties, and different processes will influence the fault permeabilities.

In this work we are using a Monte-Carlo approach, varying the input parameters with a certain distribution, and simulate the fault permeabilities for a North Sea case study. The 3D simulated mean geo-pressures are compared with measured overpressures in sandstone units from wells. To carry out the 3D simulation, the in-house Pressim2.0 software has been used to simulate pressure generation and dissipation over geological time scale. The fluid flow dynamics can be represented and described by pressure compartments laterally delineated by mapped faults from seismic. Lateral flow is modelled between the reservoir units and the vertical fluid flow in the overburden is modelled below, in between and above the reservoir units. Depth-converted maps of the overlying sediments are used to reconstruct the burial history that is adjusted for decompaction.

In this work we will present stochastic probability distribution of key input parameters defining the fault permeability and transmissibility for a study area. The simulated fault permeabilities will be compared with published data. We will also use misfit analysis, to evaluate what fault permeabilities that will be give the lowest misfit/deviation compared to measured overpressures from wells.

How to cite: Lothe, A. E., Grøver, A., and Roli, O.-A.: Innovative stochastic probability distribution of fault permeabilities in 3D geo-pressure modelling, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-14866, https://doi.org/10.5194/egusphere-egu23-14866, 2023.

09:35–09:45
|
EGU23-8434
|
ECS
|
Virtual presentation
Xushan Lu, Peter Lelièvre, and Colin Farquharson
Conventional Occam-style, minimum-structure inversion methods typically do not recover models with distinct boundaries between different geological units. Consequently, the constructed geophysical model can be very different from the true geological model and difficult to interpret in the geological context. This can be especially problematic for geological models with very thin structures that have a large physical property contrast with the background model, and determining the location of which is critical to, e.g., an exploration program. We have developed a new inversion method called surface geometry inversion which can construct geophysical models with distinct interfaces. The algorithm parameterizes the interface between geological units with triangular facets of connected nodes (vertices) and then inverts for the coordinates of these nodes. The algorithm only focuses on the boundary interface of localized anomalies and assumes the background model is known. Consequently, it is useful to have an adequately developed geological model and sufficient physical property data on which to base a background model. After the inversion, a model comprised of the background model and the anomalous region is constructed. We then utilize Markov chain Monte Carlo sampling to obtain statistical information, namely, the mean and standard deviation of the nodal coordinates of the constructed model. The standard deviation of each node is then used as an indicator of model uncertainty. The uncertainty information is useful as it can help us obtain a better understanding about the geological model. When applied to mineral exploration, the uncertainty quantification can also be used to mitigate the risks in drilling activities. We present synthetic transient electromagnetic data inversion examples with thin graphitic fault models. We also present a real-data example where transient electromagnetic data are used to target thin graphitic faults for a uranium exploration project.

How to cite: Lu, X., Lelièvre, P., and Farquharson, C.: Uncertainty quantification of geophysical models constructed by surface geometry inversion using Markov chain Monte Carlo sampling, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8434, https://doi.org/10.5194/egusphere-egu23-8434, 2023.

09:45–09:55
|
EGU23-10084
|
On-site presentation
Colin Farquharson and Mitra Kangazian

The regularization function in minimum-structure, or “Occam’s”, style of inversion can stabilize the underdetermined inverse problem and generate models with certain characteristics. The regularization term measures the amount of structure added to the model using the spatial gradient operators. The method commonly employed for calculating the gradient operators on unstructured tetrahedral meshes calculates the physical property differences across the cell faces of two adjacent cells. However, this method is not able to incorporate orientation information of the geological structures such as strike, dip, and tilt angles into the inversion. Providing this information for the inversion leads to constructing geophysical models that have a sensible representation of the true Earth models, especially when geophysical data with limited depth resolution such as gravity and magnetics data are inverted. Designing spatial gradient operators that allow one to incorporate this geological orientation information into the inversion on unstructured tetrahedral meshes is not as straightforward as for structured meshes due to the geometrical complexity of the unstructured tetrahedral meshes.

A few methods have been proposed for calculating the gradient operators for unstructured tetrahedral meshes that allow one to incorporate orientation information into the inversion framework and obtain more sensible geophysical models. Most of these methods consider a cell along with its nearest neighbours as a package and commonly use an l2 norm for the measure of the regularization term. These methods work well, however, the constructed models using these regularization methods are not as sharp as expected if an l1-type measure of roughness is used instead of an l2 norm.

In this study, the method that calculates the spatial gradient operators across the cell faces between two adjacent cells is extended such that structural orientation information can be incorporated into the inversion. The synthetic gravity examples demonstrate that this method allows models with desired strike and dip directions to be built successfully. Also, the constructed models using this proposed method have sharper boundaries compared to the constructed models that consider each cell as a package with its neighbours for the scenarios in which an l1-norm measure is employed in the regularization term.

Keywords: Gradient operators, inversion, structural orientation information, unstructured tetrahedral meshes.

How to cite: Farquharson, C. and Kangazian, M.: A comprehensive study of gradient and smoothness regularization operators on unstructured tetrahedral meshes, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10084, https://doi.org/10.5194/egusphere-egu23-10084, 2023.

09:55–10:05
|
EGU23-13340
|
ECS
|
On-site presentation
Sofia Brisson, Josefine Ziegler, Nils Chudalla, Florian Wellmann, and Christoph von Hagke

Thermokinematic modeling often relies on prescribed geometric and kinematic models at depth without considering their uncertainty. This does not allow for the proper quantification of the relative contributions of different drivers to the exhumation signal. Considering uncertainty of structural data in thermokinematic models would help understanding how much shortening associated with the observed cooling signal occurred. 

The aim of this work is to combine probabilistic structural modeling with thermokinematic forward simulations to investigate the related uncertainties. For this purpose, the Bavarian Subalpine Molasse is particularly suited as a test case, as it connects the Alpine orogen with its foreland, and should shed light on the strain distributions during the latest stages of Alpine mountain building. 3D implicit geological modeling of the Bavarian Subalpine Molasse triangle zone was carried out and combined with a systematic random sampling approach to automatically generate an ensemble of geometric models in the range of assigned uncertainties. In addition, a probabilistic 3D kinematic forward model is constructed. A link can then be obtained between kinematic model parameters and present-day geometry in comparison with field observations at the surface and also in comparison to the range of geometric uncertainties in the 3D geological model. Results show that the uncertainty (quantified as information entropy) is distributed as a function of structural complexity, depth, and data density throughout the geometric model, and additionally where fault slip ranges are large in the kinematic model.

In a next step, these models are combined with a thermokinematic forward model to integrate thermochronological measurements from previous campaigns, and eventually own measurements, to obtain an integrated picture of foreland evolution and associated uncertainties over space and time.

How to cite: Brisson, S., Ziegler, J., Chudalla, N., Wellmann, F., and von Hagke, C.: Combining thermochronological data with 3D probabilistic kinematic modeling of the Bavarian Subalpine Molasse for uncertainty estimation, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-13340, https://doi.org/10.5194/egusphere-egu23-13340, 2023.

10:05–10:15
|
EGU23-15650
|
ECS
|
On-site presentation
Evaluating RJ MCMC for parameter dimensionality adjustment in geological models
(withdrawn)
Nils Chudalla, Florian Wellmann, Christin Bobe, Sofia Brisson, and Zhouji Liang

Posters on site: Mon, 24 Apr, 16:15–18:00 | Hall X2

Chairpersons: Clare Bond, Peter Lelièvre, Florian Wellmann
X2.208
|
EGU23-7556
|
ECS
|
Alberto Albarrán-Ordás and Kai Zosseder

The growing use of multiple practical applications in the subsurface, especially in the more densely populated urban areas, demands both the development of new methods and the adaptation of existing approaches for predicting the geological heterogeneity. To this should be added the related uncertainty, since the reliability of the prediction is critical for practical decision purposes. Many of the urban areas are built in detrital depositional environments characterized by the sediment texture of the clastic mixture, which refers to the grain sizes of the particles.

The novel Di models method was conceived to accurately forecast the three-dimensional lithological composition of detrital systems by means of predicting the fictive grain size distribution of the clastic mixture through a geostatistical framework. The input data used are the direct soil observations from drilled materials described in the field according to the standards for soil description. These data are subject to systematic imprecisions in the lithological descriptions linked to the inherent generalizations of the standards used and to the subjectivity of on-site personnel.

In this context, the incorporation into the geostatistical framework of the above-cited uncertainties linked to systematic imprecisions in the input data is addressed. This process focuses on integrating the uncertainties detected in the semi-quantitative and qualitative descriptions of soil observations from drilled materials by capturing the lower and upper limits of the fictive GSD of the clastic sediments inferred from the soil descriptions. In terms of the underlying random variables, this implies the introduction of lithological noise with two equiprobable sets of input data in the simulation. Subsequently, the concepts of entropy and joint entropy are applied for uncertainty quantification of the main outputs of the Di models method, i.e., the partial percentile lithological models and the Most Uniform Lithological Model. A simulation experiment consisting of seven model setups was conducted to test the ability of the uncertainty measures with various underlying random function models and to evaluate the impacts of incorporating uncertainties from imprecise input data. The results show that the incorporation of the uncertainties in the input data into the geostatistical framework of the Di models method overcomes potential bias caused by ignoring the inaccuracies in the input data, thus providing a more realistic assessment of uncertainty. Moreover, the uncertainty measures provide very useful scalar measures for quantifying uncertainties in the grain size distributions, comparing between average uncertainties and for better understanding how the implementation parameters of the geo-modelling process influence the property forecast and the underlying uncertainties. The proposed uncertainty measures can be used to support practical decisions made based on the implementation of the Di models method.

How to cite: Albarrán-Ordás, A. and Zosseder, K.: 3-D stochastic geological modelling of the sediment texture in detrital systems: prediction of fictive grain size distributions and uncertainty quantification, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-7556, https://doi.org/10.5194/egusphere-egu23-7556, 2023.

X2.209
|
EGU23-16394
Florian Wellmann, Miguel de la Varga, Zhouji Liang, and Yang Jian

Geological models can be constructed with a variety of mathematical methods. Generally, we can describe the modeling process in a formal way as a functional relationship between input parameters (geological observations, orientations, interpolation parameters) and an output in space (lithology, stratigraphy, rock property, etc.). We evaluate here the potential of using not only the output value (prediction) itself, but also its partial derivative with respect to the input parameters to gain insight into the interpolation process, to speed-up model calibration, and to enable high-dimensional uncertainty quantification.

The calculation of this partial derivative through the complex modeling procedure requires additional work – however, this step has been greatly simplified due to progress in automatic differentiation approaches in recent years. Specifically, all modern machine learning frameworks enable a calculation of the derivatives, as this is an essential component of training in deep neural networks. We can benefit from these developments for specific geological modeling functions – and if we take specific care in the numerical implementation.

In this presentation, we discuss the basic principles between differentiable geomodelling methods, show implementation methods, and discuss potential difficulties and open challenges. We exemplify the advantage of the additional work through efficient implementations of sensitivity analyses and gradient-based sampling methods for uncertainty quantification in geological models.

How to cite: Wellmann, F., de la Varga, M., Liang, Z., and Jian, Y.: Differentiable Geomodeling: Opportunities and Challenges, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16394, https://doi.org/10.5194/egusphere-egu23-16394, 2023.

X2.210
|
EGU23-9576
|
ECS
Josefine Ziegler, Sofia Brisson, Florian Wellmann, and Christoph von Hagke

Thermochronological data and kinematic models are often combined to retrace exhumation, cooling or fault activity. However, structural uncertainty is often neglected in thermokinematic models, which can lead to bias when interpreting data. Here we aim to test the influence of structural uncertainty on the interpretation of low-temperature thermochronological data in Mount Rigi in the Subalpine Molasse, a key region in the Swiss foreland fold-thrust belt of the European Alps. The region has been incorporated into the Alpine orogenic wedge in the Miocene, which led to the development of a triangle zone at the leading edge of deformation. An extensive low-temperature thermochronological data set exists, including apatite fission track as well as apatite (U-Th)/He data, which contains outliers not easily to be explained with the existing kinematic models.

To diminish bias in the thermokinematic model we first estimated the geological uncertainty by computing and comparing 1000 stochastically generated 3D implicit geometric models, varied within an assigned uncertainty range assigned to geological input parameters. Model generation is performed with the stochastic geological modeling engine implemented in the Python package GemPy. In a second step a kinematic model was created which was altered in areas of high uncertainty found in the first step. With this setup, minimum and maximum values of cooling associated with shortening within the fold-thrust belt can be determined. The remaining cooling signal (if present) must hence be associated with other drivers, such as erosion caused by drainage reorganization, or uplift and erosion due to deep seated processes. Additionally, hydrothermal fluids could be held responsible for explaining individual data points. With this research, we hope to give new insights into the temporal evolution of heat flow in foreland basins and show how including geological uncertainty can lead to better constrained time-temperature histories.

How to cite: Ziegler, J., Brisson, S., Wellmann, F., and von Hagke, C.: Understanding outliers in thermochronological anomalies in the Swiss Subalpine Molasse and how they are linked to geological uncertainty, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9576, https://doi.org/10.5194/egusphere-egu23-9576, 2023.

X2.211
|
EGU23-12964
|
ECS
Anne-Sophie Mreyen, Miguel de la Varga, and Frédéric Nguyen

Prior geological knowledge is crucial in a wide range of geophysical case studies aiming at data processing and interpretation. In this regard, geological modeling methods can be considered as efficient tools to facilitate petrophysical parametrization and constrain geophysical inversions as a priori information over a defined model space. We use the open-source library GemPy (De la Varga et al., 2019) which offers a community driven alternative based on the potential-field method and specialized in probabilistic modeling.

In this work, we show an exemplary case study of high-resolution marine geophysical data comprising single-trace reflection and underwater refraction seismics validated with geotechnical data. Two approaches are suggested: (1) digitalization of geophysical imagery, i.e., spatial information from interpreted horizons and inherent uncertainties, and (2) pseudo depth migration of picked reflection seismic travel times using a simple velocity model from parallel recorded underwater refraction data. Next to a deterministic “best-fit” solution, the model is interpreted following probabilistic distributions of input data and classified after their identified certitudes (e.g., depth range of an observed seismic horizon), where prior knowledge is optionally included using Bayesian Networks. Finally, global uncertainties are estimated by multiple model realizations allowing for improved data assessment and enhanced decision making.

The further outlook of this study is the creation of a variety of digital twins taking into account realistic conditions in terms of both, the geological environment as well as data acquisition, as a solid prior for future data exploitation by inverse processes of wave propagation in shallow marine environments.

How to cite: Mreyen, A.-S., de la Varga, M., and Nguyen, F.: Marine geomodels from high-resolution seismic reflection data – a model ensemble approach, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12964, https://doi.org/10.5194/egusphere-egu23-12964, 2023.

Posters virtual: Mon, 24 Apr, 16:15–18:00 | vHall TS/EMRP

Chairpersons: Clare Bond, Peter Lelièvre, Florian Wellmann
vTE.2
|
EGU23-10065
Zhouji Liang, Miguel de la Varga, and Florian Wellmann

Uncertainties in geological modeling have drawn increased attention in recent years. This is due to the fast-evolving computational power and more demanding evaluation of the modeling procedure.

However, these uncertainty quantifications (UQ) often face high dimensionality. Advanced mathematical methods have been developed to significantly improve the efficiency of the UQ process. Still, many of the methods require not only the forward evaluation of the quantity of interest but also the partial derivative information to guide the posterior exploration (e.g., HMC, SVGD). Differentiable geological modeling methods have been introduced and have become an appealing tool to efficiently evaluate the partial derivatives w.r.t. the input parameters using Automatic Differentiation (AD) techniques.

To successfully apply AD to geological modeling several challenges need to be addressed. One of these challenges is the aliased effect due to discretization. In this work, we will introduce a method to generate a trainable geological model under the framework of gravity inversion using the implicit geological modeling method. We present a smooth-step function in the scale value domain and adopt an order-reduction method to provide a visual evaluation of the trainability of the generated model. This work provides the fundamental step to the application of advanced derivative-informed UQ and optimization methods.

How to cite: Liang, Z., de la Varga, M., and Wellmann, F.: Trainable geological model under the framework of model-based gravity inversion, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10065, https://doi.org/10.5194/egusphere-egu23-10065, 2023.