This thesis develops a geological ensemble-reduction workflow for efficient uncertainty quantification (UQ) in Carbon Capture and Storage (CCS). Using early-time full-physics (FP) injection-rate data and distance-based clustering, representative models are selected to preserve ensemble percentile bounds (P10–P50-P90) within ≤5% relative RMSE. Applied to two 108-member ensembles, injection-rate uncertainty distributions were accurately reconstructed using only 6 and 9 representative models (∼7–10% of the total simulation cost). Early FP rates (≤10 days) proved strongly predictive of long-term behavior and key geological controls. Distance-based generalized sensitivity analysis (dGSA) effectively identified influential parameters governing reservoir response variability at only ~1.3% of the full simulation cost. The approach provides a transparent, low-cost framework for CCS reservoir UQ, enabling robust risk and performance assessment with order-of-magnitude computational savings.

This thesis explores whether vertical stratigraphic cross flow from the A zone to the C1 zone of the Upper Slochteren Member can occur without wellbore crossflow, and under what conditions such recharge is plausible. The study focuses on the B-well in the L-field, Dutch offshore North Sea, where production resumed after a shut-in period, and pressure behavior suggested possible recharge. Four different basecase models were constructed using Rapid Reservoir Modeling (RRM), each representing a different degree of vertical connectivity across the B zone. These models were then transferred to Computer Modeling Group (CMG) software, where dynamic multiphase simulations were performed to assess gas flow between the zones of the Upper Slochteren Member.
The results show that stratigraphic recharge from A to C1 is possible, but highly sensitive to the inter- nal architecture of the B zone. Increased vertical connectivity across the B zone consistently show earlier pressure communication and higher gas fluxes into the C1 zone. Sensitivity analyses were conducted on porosity, vertical permeability, and gas relative permeability to test their influence on flow behavior. These parameters affected not only the rate of gas migration, but also the degree of pressure redistribution across the model, which influences the gas rates even further. Capillary pressure and water saturation were also found to control gas mobility, particularly in low-permeability or heterolithic intervals. The findings do not fully support the hypothesis that stratigraphic recharge explains the observed pressure response in the reservoir, but suggest it may account for part of it. Additionally, the results emphasize the importance of considering vertical heterogeneity and capillary forces when evaluating near-wellbore connectivity.

This study quantifies the impact of mesoscale geological heterogeneity on CO2 storage behavior, using the Pano flood-tidal delta as a case study. A hierarchical workflow was employed, progressing m simple to complex static modeling (L1 to L4) based on outcrop interpretation, followed by flow diagnostics analysis. The results reveal a non-linear relationship between model complexity and system behavior. The mesoscale architecture (L2) primarily controls early CO2 breakthrough risk, whereas the sub-lobe scale heterogeneity (L3) is crucial for reliably predicting long-term storage performance and sweep efficiency.
A key insight from this work is that the level of model complexity systematically biases the simulated flow narrative. Consequently, there is no single ”correct” model; instead, the choice of complexity inherently pre-selects which aspects of storage behavior (e.g., early risk vs. long-term efficiency) will be most accurately represented. This provides a decision-making framework for CO2 storage projects, demonstrating that distinct optimal levels of model complexity exist for specific engineering objectives, such as using L2 for risk identification and L3 for performance prediction—thereby guiding more effective and intentional model deployment.

Sandstones from the Triassic Main Buntsandstein Subgroup represent a promising deep geothermal target in the subsurface of the Netherlands. These sands have a widespread distribution and temperatures that locally reach up to 140-150°C at depths of ~ 4 to 5 km. The Main Buntsandstein Subgroup is a sand-prone interval, but the reservoir quality of these sandstones is known to be heterogeneous as a result of the interplay between depositional and diagenetic processes. Recent drilling campaigns at depths greater than 4 km have confirmed the poor reservoir quality in certain areas. Nonetheless, successful geothermal projects have been realized in the southern Netherlands at shallower depths of 2 to 3 km, where temperatures of 70 to 90°C can meet local heating demands for greenhouses and district heating. To facilitate further geothermal exploration, it is important to understand the geological conditions that control reservoir quality and identify areas with favourable subsurface conditions. The Main Buntsandstein sediments have been extensively studied at a regional scale in the Netherlands. The factors controlling reservoir quality remain however poorly understood in the southern Netherlands, particularly in the southeastern region where data are scarce.

This thesis aims to improve the geological understanding of the Main Buntsandstein sediments by conducting an integrated geological study leveraging an extensive hydrocarbon dataset and newly acquired data from geothermal exploration in the southeastern part of the Netherlands. The study characterizes the structural, sedimentological, and diagenetic heterogeneities of the Main Buntsandstein Subgroup and evaluates their control on porosity, permeability, and, ultimately, geothermal potential. The first part of the thesis (Chapter 2) assesses the structural evolution of the Roer Valley Graben and the distribution of the Main Buntsandstein sediments through detailed seismic and well-data interpretation. A 2D palinspastic restoration was performed to evaluate the burial history and basin geometry during the Early to Middle Triassic. The analysis reveals that the central and southern parts of the Roer Valley Graben were active depocenters during the Early to Middle Triassic, while the northern part was a more marginal area where predominantly fine-grained sediments were deposited. After deposition, the sediments were significantly impacted by faulting, with burial depths reaching 4-5 km in the central graben, while the flanks experienced shallower burial of 2-3 km, making them promising targets for future geothermal investigations given the higher likelihood of preserved primary reservoir properties.

Next, in Chapter 3, the sedimentology and lithostratigraphy of the Main Buntsandstein are examined using a subsurface dataset of well cores, gamma-ray logs, and thin sections. The study identifies six lithofacies associations deposited through different fluvial processes with minor aeolian reworking. Overall, the different depositional processes are linked to tectonic and climate changes and led to the development of three distinct types of reservoir architectures, each with its own set of heterogeneities at different spatial scales. At the scale of the study area the heterolithic sediments deposited as result of playa-lake expansions can hamper the vertical connectivity of sandstone units given their confluency at km scale. Within the sandstone units, cemented zones or mud drapes are the most common fluid baffles.

Furthermore, most of the sandstone types do not preserve a primary relationship with reservoir properties because of post-depositional diagenetic processes. The diagenetic processes that control porosity and permeability in the Main Buntsandstein Subgroup were analyzed through detailed petrographic studies of available and a series of new thin sections (Chapter 4). The results show that illite, quartz, and dolomite are the dominant cements in these sandstones, with their influence on reservoir quality varying according to the sedimentary facies. In areas such as the Roer Valley Graben flanks, where the maximum burial history during Triassic and Early to Mid-Jurassic was largely shallower than 2 km, lower compaction and cementation rates favour the preservation of primary reservoir properties.

A structural study was conducted to analyse the distribution and characteristics of natural fractures and to investigate the mechanical stratigraphy of the Main Buntsandstein Subgroup (Chapter 5). This study used a dataset from the West Netherlands Basin as the Roer Valley Graben lacks the needed datasets. It revealed that natural fractures are favourably oriented with respect to the present-day in-situ stress, increasing the likelihood that these fractures are open. Fracture density was found to be higher in the heterolithic sedimentary successions, suggesting a link to the depositional environment and Main Buntsandstein Subgroup stratigraphy.

The geological insights gained from these studies were then employed to assess the geothermal potential of the Main Buntsandstein Subgroup in the Roer Valley Graben (Chapter 6). Porosity and permeability were evaluated, and calculations for Heat Initially in Place (HIIP) and Geothermal Power (GP) were made at well scale. A sensitivity analysis identified reservoir thickness and permeability as the parameters that most influence these calculations. The results are contextualized within the broader geological knowledge developed throughout the thesis, and three case studies corresponding to three types of potential geothermal plays are presented and discussed.

A conclusive synthesis is presented in Chapter 7, aimed at summarizing the main findings of this thesis and discussing how these results should be used in the future to reduce uncertainty and mitigate risks in geothermal exploration within the Main Buntsandstein of the southern Netherlands. To keep pace with the growing heat demand and the transition away from hydrocarbons as a primary energy source, the geothermal industry must make fast progresses. The Main Buntsandstein Subgroup has the potential to serve as a promising reservoir, particularly in regions like the northwestern Roer Valley Graben, where geological conditions suggest more favorable reservoir properties. Future exploration and production in these areas could play a crucial role in meeting the Netherlands sustainable energy targets.

Accurate subsurface characterization is critical for emerging energy-transition projects, yet conventional seismic data often fail to resolve metre-scale heterogeneities that strongly influence reservoir behaviour. This thesis develops an integrated workflow that combines stratigraphic forward modelling with seismic forward modelling to improve the detection of sub-seismic stratigraphic features in wave-dominated shoreface systems. Outcrop analysis shows significant variability in petrophysical and acoustic properties, revealing the limitations of lithology-based seismic approaches. By converting grain-size distributions from stratigraphic simulations into acoustic properties, the workflow produces synthetic seismic data that better represent fine-scale stratigraphy. Angle-dependent seismic analysis shows potential for identifying subtle acoustic variations, though current modelling techniques require refinement. The results demonstrate that linking geological, petrophysical, and geophysical data enhances subsurface resolution and point toward future developments involving more complex models, in-well seismic methods, and machine learning.

The storage of carbon dioxide is now regarded as a critical industrial application aimed at mitigating its accumulation in the atmosphere. Reservoirs and aquifers have been identified as viable alternative locations. This report seeks to examine the impact of sedimentological heterogeneity on the development and migration of CO2 plumes over time. The study uses the Roda Sandstone Formation as a case study, primarily due to the presence of carbonate-cemented layers within the Gilbert-delta lobes, which contribute to reservoir heterogeneity commonly observed in subsurface formations. The utilization of a sketch-based modeling approach was employed in constructing the geo-models represented by Rapid Reservoir Modeling (RRM) software since the Roda Sandstone is an exposed section in Isabena Valley in Spain. Furthermore, various realizations are constructed using different parameters of the cemented layers to comprehensively comprehend all potential scenarios. The aforementioned observations pertain to variations in the thickness and lateral continuity of the cemented layers. Additionally, dynamic modeling is also performed by injecting CO2 into the reservoir for 50 years. The simulator utilized for dynamic modeling is the Delft Advanced Research Terra Simulator (DARTS). The findings demonstrate that reservoirs with moderate sedimentological heterogeneity are actually better sites to store CO2 due to the capacity and effectiveness of the storage.

Alluvial deposits in the subsurface are essential for geo-energy production and storage in many regions worldwide. Accurate correlation and characterisation of alluvial stratigraphy requires an understanding of how river channels were spatially deposited, and which geomorphological processes acted upon them. This doctoral dissertation the potential of orbital forced cyclic climate control on fluvial systems and its application to reservoir characterisation and subsurface correlation. It discusses the correlation between cyclic arrangements of floodplain strata and corresponding sandstone occurrence based on an analogue outcrop study. Subsequently the methods and concepts are applied on two subsurface case studies where correlation and characterisation of the reservoir is attempted.

The Southern Chotts and Jeffara Basins are situated within the Saharan Domain of Central Tunisia, North Africa. The Southern Chotts Basin hosts reservoirs within the Triassic, Permian and Ordovician units that contain significant hydrocarbon accumulations whilst the Jeffara Basin contains outcrop analogues of the same hydrocarbon­bearing formations. The basins experienced a late Hercynian shortening phase which involved the uplift of a major topographic high (Tebaga de Medenine). This high, in conjunction with a older regional high, the Telemzane Arch influenced the deposition and geometry of the Permian and Triassic units across both basins. This shortening event is characterised at the scale of hundreds of meters by E­W striking folds into which the mid­late Triassic and early Jurassic units are deposited. The folding is also observed at field scale (10’s meters) through small fault­related folds in the Permian deposits of the Tebaga de Medenine. This late Hercynian phase occurs between the late Permian and early Jurassic in the basins. Fracture data collected from the upper Permian and lower Triassic units (Jeffara Basin) provides an analogue to the fracture networks at depth (Southern Chotts Basin) in the Paleozoic reservoirs. A conjugate fracture system observed in the field (from fracture pavements) corroborates with the interpretation of regional shortening in the basins. Seismic attribute analysis on depth slices in the Paleozoic reservoirs also shows the conjugate system at depth. This analysis is integrated with outcrop fracture data and FMI data from wells to create an open fold distributed fracture model of the system in the basins. This model indicates the main driver for fracture generation in the region is folding and is used to predict the fracture networks at depth. This is undertaken using discrete fracture network (DFN) modelling of the subsurface. This model is integrated with a deterministic model from the seismic time slices to create a hybrid predictive fracture model of the early Paleozoic reservoirs. Analytical aperture modelling of the fracture model demonstrates the fractures varied in openness depending on orientation and fracture length. The conjugate set orientated at 240∘and longer joints detected from seismic attributes presented the widest aperture size. These fractures in the subsurface at implications on the transmissibility of the reservoirs, especially Permian units which have low bulk rock permeability and the lower Triassic (TAGI) sequences which are susceptible to compartmentalisation.

Fluvial reservoirs are difficult to model due to the high permeability contrast between the sandstone bodies and the overbank deposits and the complex geometry of the permeable (sandstone bodies) and impermeable zones (overbank deposits). A set of fluvial meandering models has been generated using FLUMY. The models represent a range of Net-to-Gross ratios and sandstone body geometries. In order to quantify the effect of sample size on effective properties, the models are evaluated based on the statistical moments of the probability distributions of porosity and single-phase permeability as a function of sample size. The porosity and permeability show a high spread at small sample volumes, but the spread reduces as the sample size increases. A normalized standard deviation, the coefficient of variation, has been used as a criterion for the variability of the probability distributions. The coefficient of variation of the porosity and the horizontal permeability show a monotonic decline as a function of sample size. The coefficient of variation of the vertical permeability does not show a monotonic decline. This is caused by a drastic decrease of the mean of the vertical permeability with increasing sample volume. The mean of the horizontal permeability also decreases with increasing sample size, but to a lesser extent. The mean of the probability distributions of permeability as a function of sample size converges much earlier than the standard deviation. This convergence indicates that we can determine the effective properties at the Representative Elementary Volume (REV), without reaching REV. The convergence of the mean could potentially be used as a criterion for the relevant spatial scale of upscaling from the fine scale static model to the coarse scale model. Furthermore, if cells are uncorrelated at a scale where the mean of the permeability is not a function of sample volume anymore, random attribution of properties can be used to populate dynamic grid cells.

The upper reservoir intervals of the Lower Jurassic Åre Formation in the Heidrun Field (Offshore mid-Norway) are very heterolithic and have the lowest oil recovery factor of the field despite significant amounts of remaining reserves. One of these reservoir zones is the formation Åre 6.2, which is mainly composed of tide-dominated heterolithic channel belt deposits. It contains particular layers that have excellent properties with permeabilities up to 10 Darcy. These layers are predicted to affect the production results as they can act as ‘thief zones’ within the low permeable heterolithic facies causing large quantities of water to flow through, leading to poor sweep efficiency and early water breakthrough. This study focuses on constructing conceptual depositional models of the Åre 6.2 and building detailed geological models to investigate the effect of the thief zones on overall fluid flow predictions.

Conceptual depositional models were constructed by determining the characteristics of the reservoir and its depositional environment. Seven cored wells were used as the primary data to interpret lithofacies and facies associations. The study showed that Åre 6.2 mainly consisted of structured sandstones and heterolithic lithofacies with features that indicate that tidal process play an important role in the deposition. The influence of tidal process on deposition is further exemplified by the identification of two different types of channel facies associations, which are tidal and distributary channels. The thief zones were found in both facies associations, suggesting that the thief zones were formed during high freshwater discharge into the channels supplying coarse sandy material influx during a phase of high-energy deposition. To make detailed models of the tidal and distributary channels, multiscale modeling techniques were utilized to better represent the reservoir heterogeneities at the lithofacies and facies association scales.

At the lithofacies scale, models were built in SBEDTM and the upscaled values of each lithofacies were obtained by applying the Representative Element Volume (REV) concept. The upscaled values were then used as input in the facies association scale models in order to represent the heterogeneities at the lithofacies scale to the next heterogeneity level. This step is essential since heterogeneities at a smaller scale may affect reservoir flow properties. Two different channel models were built in ReservoirStudioTM based on the conceptual depositional model and using outcrop analogue data from the Gule Horn Formation (Neill Klinter Group) in the Albuen area (Greenland). Flow-based upscaling was used to analyze the model uncertainties and determine a proper upscaling grid size. Finally, streamline simulations were performed to identify the effect of the thief zones. The simulation confirms that the thief zones influence fluid flow in the reservoir zone significantly as most flow was concentrated in the thief zones.

Traditional reservoir modelling utilizes a stochastic approach centered around statistics to generate a 3D representation of the reservoir. While the results of a stochastic approach are favorable, they tend to lack the detail that is necessary to fully understand the different aspects of the reservoir, more precisely, the inclusion of heterogeneities. A possible solution to this problem is the use of process-based models. Process-based models utilize the physical processes involved in the transportation, deposition, and erosion of sediments. As a result, the models generated are more complex in nature and better represent what is found within the subsurface. The use of process-based models within reservoir modelling is a a relatively new process that has yet to be fully utilized due to the difficulty in calibrating the models with well data. Before process-based models can be tested on their viability to include the multi-scalar heterogeneities, the model must be calibrated to match the real-world data. To test this, a combination of field work, lab work, and computer simulations is required. In this experiment, the Roda Sandstone Member, a Gilbert-type delta deposit in Northern Spain, was chosen as the unit of focus. The Roda Sandstone consists of multiple prograding sand lobes, with Roda Y being chosen as the focus of this experiment. The Roda Y sandstone exhibits a predominately medium and coarse grain distribution in the central locations of the sand lobe, and a fine and very fine distribution in the distal locations. Five simulations were run within the process-based modelling software "Delft3D", each with varying sediment input parameters, to observe the effects on the results. The results for the simulations show a strong calibration for each of the five simulations for coarse and medium grained sand, with a percent difference between the model results and the field data of 4-8%. The fine and very fine data contain a higher average difference between the two data sets, ranging from 18-23%. The difference for mud averages around 11%, with predominately more mud being deposited within the simulations. The large differences for the fine and very fine grained can be attributed to the difference in the size and shape of the sand lobe produced by the simulations. In locations were the two data points are equivalent in regards to depositional location within the sand lobe, a high correlation is observed. The results indicate that process-based models have the potential to be a very useful took within reservoir modelling. As this is the first step in a series of steps, additional testing is required for the additional aspects involved in utilizing process-based models to better incorporate heterogeneities within reservoir models.