Fractures are ubiquitous in geological formations and can often have an impact on subsurface applications such as geothermal energy, groundwater management or CO2 storage. Quantifying the relationship between the uncertainties inherent to fracture networks and the corresponding flow behaviour for these applications remains an open challenge. Simulation studies that are based on outcrop analogues of fracture networks have yielded many new insights about heat and mass transfer in fractured geological formations but are restricted to a limited number of fracture network realizations, simplified assumptions about fracture network properties or deterministic models, making it difficult to analyse a wide range of uncertainties. This study introduces a flexible workflow that generates ensembles of geologically plausible fracture networks that can be based on statistical data from outcrop analogues. The fracture networks are generated using a computationally efficient approach that combines mechanical and statistical methods. The ensembles are then seamlessly linked to multi-purpose flow and transport simulations where the fractures are represented explicitly in a porous and permeable rock matrix. This approach can enable new uncertainty quantification methods, supported by machine-learning-based emulators, to analyse how fracture network properties, such as fracture intensity, fracture aperture or fracture orientation, influence heat and mass transfer in fractured geological formations. The workflow is illustrated using two classic example applications pertinent to fracture network modelling – one based on outcrop data to assess thermal behaviour in geothermal systems, and one synthetic study to analyse the transition from matrix-dominated to fracture-dominated flow – and released as open-source code.

Reducing the uncertainty of reservoir characterization requires to better identify the small-scale structures of the subsurface from the available data. Studying the seismic response of meter-scale, stratigraphic heterogeneities typically relies on the generation of reservoir models based on outcrop examples and their forward seismic modelling. To bridge geological information and seismic modelling, these methods allocate values of acoustic properties, such as mass-density and P-wave velocity, according to discretized properties like layer-type lithology or facies units. This strategy matches the current workflow in seismic data inversion in industry, where modelling workflows are based on lithofacies distributions. However, from stratigraphic modelling, we know that meter-scale heterogeneities occur within certain facies and lithologies. Here, we evaluate the difference on the seismic response between allocating acoustic properties in a grain size–based, semi-continuous manner versus discretized manners based on lithology and facies classifications. To do so, we generate a reference geological simulation that we populate with acoustic properties, mass-density and P-wave velocity, using three different strategies: (1) based on grain size distribution; (2) based on facies distribution; and (3) based on lithology. The method we propose includes the generation of realistic geological simulations based on stratigraphic modelling and the transformation of its output into acoustic properties, honouring the intra-lithology and intra-facies, small-scale structures. We, then, generate seismic data by applying a forward seismic modelling workflow. The synthetic data show that the grain size–based simulation allows the identification of small-scale, stratigraphic heterogeneities, such as beds with strong density and velocity contrasts. These stratigraphic structures are smoothened or may completely disappear in the facies and lithology discretized simulations and, therefore, are not (well) represented in the synthetic seismic data. Recognizing meter-scale, stratigraphic heterogeneities is relevant for the characterization of the fluid flow in the reservoir. However, current discrete and lithology-based strategies in seismic inversion are not able to resolve such heterogeneities because real subsurface properties are not discrete properties but continuous, unless there are stratigraphic discontinuities such as erosional surfaces or faults. This research works towards a better understanding of the relationship between changes in these continuous properties and the observed seismic data by introducing greater complexity into the discretized geological simulations. Here, we use synthetic seismic images with the goal of eventually aiding in fine-tuning seismic inversion methodologies applied to real seismic data. One pathway is to foster the development of inversion approaches that can leverage stratigraphic modelling to get stronger geological priors and replace the standard but inadequate multi-Gaussian prior.

Characterising fractures in geothermal reservoirs is crucial for understanding heat and fluid flow, as fractures control reservoir permeability. Due to data scarcity, estimating fracture network properties remains uncertain. Dynamic data, such as well tests, provides indirect insights into subsurface properties and workflows have been developed to illustrate how uncertainty in fracture data affects flow behaviour. However, they use simplified, randomly generated fracture geometries limiting their applicability to real-world scenarios. This study presents a machine learning workflow for characterizing fractured reservoirs using transient data, focusing on geothermal reservoirs. A comprehensive dataset of 5000 geologically consistent Discrete Fracture Networks (DFNs) was generated using GeoDFN and directly linked to MRST for simulations. The workflow then applies a k-medoids clustering approach, using dynamic time warping (DTW) as a distance metric, to cluster pressure responses with similar transient behaviour. We identified 18 distinct pressure behaviour. Linking clusters to fracture properties reveals that fracture intensity, aperture, and length have the most significant impact on pressure behaviour, while fracture set type was found to be the least important factor. Future work will extend this workflow to temperature transient data and apply advanced machine learning techniques for both forward and inverse modelling of fractured geothermal reservoirs.

This research focuses on investigating the relative performance of a range of machine learning algorithms, namely the artificial neural network, support vector machine, Gaussian process regression, random forest, and XGBoost, for predicting the undrained shear strength from cone penetration test data. This is to assess how machine learning could help us lower the need for laboratory test data. The training dataset compiles 526 data from 12 regions and the testing dataset consists of 20 data from a polder located close to Leiden in the Netherlands. In addition, k-fold and group k-fold cross-validation strategies are both applied to validate the models. The poor performance of the models during group k-fold cross-validation suggests that, while machine learning techniques can perform well when site-specific data are included during training, they struggle to generalize without site-specific data. This highlights the difficulty of capturing soil heterogeneity and suggests that either machine learning methods should be trained on specific sites for which some data are already available, or much larger training datasets are needed.

Many stratigraphic features occur at a scale that is at the edge or below vertical seismic resolution. Thus, they cannot be directly observed in the seismic data, while still having an important effect on the fluid flow within the system. The better understanding of these sub-seismic scale features or heterogeneities can help decrease subsurface uncertainty. Here we present a novel method that integrates forward stratigraphic modelling, petrophysics, and geophysics to decipher the seismic imprint of heterogeneities in wave-dominated, shallow marine environments. The proposed three-stepped method starts with defining geology-related input parameters for BarSim, a stratigraphic forward modelling software that produces models that include stratigraphic architecture, grain size distribution, and facies distribution. Then, the geological data is translated, cell by cell, into petrophysical data (density, Vp, and Vs) using emphirical relationships. Finally, the forward seismic modelling is performed by combining a finite difference approach strategy and angle-dependent full wavefield migration to retrieve the angle gathers This method also allows the generation of large amounts of field-independent data suitable for machine learning applications.

Understanding the impact of fractures on fluid flow is fundamental for developing geoenergy reservoirs. Pressure transient analysis could play a key role for fracture characterization purposes if better links can be established between the pressure derivative responses (p′) and the fracture properties. However, pressure transient analysis is particularly challenging in the presence of fractures because they can manifest themselves in many different p′ curves. In this work, we aim to provide a proof-of-concept machine learning approach that allows us to effectively handle the diversity in fracture-related p′ curves by automatically classifying them and identifying the characteristic fracture patterns. We created a synthetic dataset from numerical simulation that comprised 2560 p′ curves that represent a wide range of fracture network properties. We developed an unsupervised machine learning approach that can distinguish the temporal variations in the p′ curves by combining dynamic time warping with k-medoids clustering. Our results suggest that the approach is effective at recognizing similar shapes in the p′ curves if the second pressure derivatives are used as the classification variable. Our analysis indicated that 12 clusters were appropriate to describe the full collection of p′ curves in this particular dataset. The classification exercise also allowed us to identify the key geological features that influence the p′ curves in this particular dataset, namely (1) the distance from the wellbore to the closest fracture(s), (2) the local/global fracture connectivity, and (3) the local/global fracture intensity. With additional training data to account for a broader range of fracture network properties, the proposed classification method could be expanded to other naturally fractured reservoirs and eventually serve as an interpretation framework for understanding how complex fracture network properties impact pressure transient behaviour.