Underground hydrogen storage in porous formations is a promising solution for large-scale energy storage. Understanding hydrogen flow and trapping at the pore-scale is crucial for assessing storage capacity and recovery efficiency. While pore-scale flow visualisation experiments provide realistic insights, they are resource intensive and technically challenging. Pore-network models offer a computationally efficient tool for simulating multiphase flow in porous media and can serve as a valuable complement to pore-scale experiments. However, their accuracy remains a key uncertainty and must be evaluated for future application. This study evaluates the performance of a quasi-static pore-network model by comparing its predictions against three-dimensional pore-scale hydrogen flow visualisation experiments in a homogeneous Bentheimer sandstone and a layered Clashach sandstone. The model was calibrated to match experimental end-state saturations, and its performance was evaluated through comparisons of spatial saturation profiles and pore occupancy. The novelty of this study lies in the direct comparison of hydrogen displacement between pore-scale experimental observations and pore-network model simulations, providing an assessment of model performance under varying degrees of rock heterogeneity relevant to underground hydrogen storage. The pore-network model shows good agreement with experimental observations for the homogeneous rock, particularly during drainage, and is subsequently used to analyse additional scenarios, including cyclic hydrogen injection and withdrawal, and wettability variations. These simulations provide insights into capillary pressure behaviour and residual saturation trends. In contrast, for the heterogeneous and layered Clashach sandstone, the model fails to capture the trapping and fluid redistribution observed experimentally during imbibition, revealing limitations in modelling fine-scale heterogeneity.

Understanding pore-scale hydrogen displacement and trapping is crucial for developing subsurface hydrogen storage facilities. While pore-scale flow visualization experiments provide critical insights, they are complex and re source-intensive. Quasi-static pore-network models (PNMs) offer a faster alternative for simulating multiphase flow. This study uses a widely employed PNM to simulate hydrogen flow in sandstones, comparing results with pore-scale flow visualization experiments at reservoir conditions.

Two sandstone samples were used: homogeneous Bentheimer and heterogeneous Clashach. Pore networks were extracted comprising pores and throats, and hydrogen-water flow was simulated, modelling drainage and imbibition processes. Results were analysed for fluid saturations and pore occupancies.

For the homogeneous rock, the PNM matches experimental results for both drainage and imbibition, enabling simulations of different wettability conditions and multiple injection and production cycles. For the heterogeneous rock, the PNM reasonably predicts the hydrogen flow path during drainage but fails to accurately predict imbibition. This discrepancy highlights the limitations of PNMs in predicting pore-scale flow in complex rocks.

In conclusion, while PNMs offer a computationally efficient means to simulate hydrogen flow, they cannot currently replace experimental observations for complex rocks. Further validation against experimental findings is necessary to refine these models and expand their applicability for underground hydrogen storage.

Pore-network representations of permeable media provide the framework for explicit simulation of capillary-driven immiscible displacement governed by invasion-percolation theory. The most demanding task of a pore-network flow simulation is the identification of trapped defending phase clusters at every displacement step, i.e. the phase connectivity problem. Instead of employing the conventional adjacency list we represent the connectivity of a phase cluster as a tree accompanied by a set of adjacent non-tree edges. In this graph representation, a decrease in phase connectivity due to a pore displacement event corresponds to deletion of either a tree or a non-tree edge. Deletion of a tree edge invokes a computationally intensive search for a possible reconnection of the resulting subtrees by an adjacent non-tree edge. The tree representation facilitates a highly efficient execution of the reconnection search. Invasion-percolation simulations of secondary water floods under different wetting conditions in pore-networks of different origin and size confirm the efficiency of the proposed phase connectivity algorithm. Moreover, a systematic simulation study of runtime growth with increasing model size on regular lattice networks demonstrates a consistent orders-of-magnitude speed-up compared to conventional simulations. Consequently, the proposed algorithm proves to be a powerful tool for invasion-percolation simulations on large multi-scale networks and for extensive stochastic analysis of typical single-scale pore-networks.

A multiscale network integration approach introduced by Jiang et al. (2013) is used to generate a representative pore-network for a carbonate rock with a pore size distribution across several orders of magnitude. We predict the macroscopic flow parameters of the rock utilising (i) 3-D images captured by X-ray computed microtomography and (ii) pore-network flow simulations. To capture the multiscale pore size distribution of the rock, we imaged four different rock samples at different resolutions and integrated the data to produce a pore-network model that combines information at several length-scales that cannot be recovered from a single tomographic image. A workflow for selection of the number and length-scale of the required input networks for the network integration process, as well as fine tuning the model parameters is presented. Mercury injection capillary-pressure data were used to evaluate independently the multiscale networks. We explore single-scale, two-scale, and three-scale network models and discuss their representativeness by comparing simulated capillary-pressure versus saturation curves with laboratory measurements. We demonstrate that for carbonate rocks with wide pore size distributions, it may be required to integrate networks extracted from two or three discrete tomographic data sets in order to simulate macroscopic flow parameters.

The physics of multi-phase displacement processes in the individual pores of a connected pore-network of a rock ultimately controls how oil, gas and water move in reservoir rocks and how readily they can be produced. These pore scale processes, including piston-like displacement, snap off, film-flow and fluid redistribution have been studied traditionally in pore-network simulations as well as in 2D micro-model experiments. However, recent advances in X-ray computed micro-tomography (CT) techniques now enable us to visualize and monitor these processes in 3D during in-situ core flooding experiments at pore-scale resolution. This provides new information on the spatial and temporal evolution of oil and water phase clusters and films. In this paper, we present results of a suite of two-phase fluid displacement experiments performed on a dolomite core plug. The experiments consist of a series of fluid injections and in-situ CT scans of the core in certain time steps during the drainage and imbibition displacement processes. The fluid phases are brine and a mineral oil. A simple, low-cost and highly X-ray transparent design for core flooding cells is introduced. Our experiments and CT images allow us to visualize the 3D fluid structures of each phase during fluid displacements in carbonate rocks with excellent clarity. Piston-like displacement and snap off mechanisms have been captured clearly in 3D. In addition, the formation, collapse and reorganisation of brine films surrounding oil blobs in individual pores were clearly visualised. However, the formation of oil films, which could provide connectivity for the hydrocarbon phase at low saturations, could not be observed in these experiments. The observed displacement processes and the particular oil-water/rock configurations seen in the displacements suggest the rock is preferentially water wet.

Pore-network modelling, or digital petrophysics, is an emerging technology which allows computing physically consistent two- and three-phase flow functions such as relative permeability and capillary pressure curves at arbitrary wettability. Considering the difficulty in measuring these functions reliably in the laboratory, pore-network modelling is increasingly used to guide and complement costly and time-consuming SCAL programmes. Although pore-network modelling is now well-established for clastic reservoirs, applying it to carbonate rocks is significantly more challenging due to their complex and multi-scale pore structure, comprising micro- and macro-pores as well as cracks and fractures. We have hence developed a novel method to integrate pore-networks, which have been extracted at multiple scales directly from CT images at different resolutions, into a single pore-network model that can be used in subsequent calculations of two- and three-phase flow functions. This method has been validated by comparing computed two-phase relative permeability and capillary pressure curves for a multiscale network to the corresponding laboratory measurements. However, in this approach it is important to accurately consider the impact of the spatial distribution of fine network elements, which are extracted from high-resolution images. We therefore show the impact of the spatial correlation of high-resolution porosity on single- and two-phase fluid flow and propose a model that allows us to simulate the spatial correlation between coarse-scale pores and fine-scale porosity in the absence of a suitable CT image that segments the rock into three phases (pores, sub-resolution matrix porosity, and solid). We demonstrate the impact of the spatial correlation of microporosity on absolute and relative permeabilities by applying this model to various datasets, including CT images of an off-shore carbonate reservoir. This demonstrates that the spatial correlation of microporosity is one of the key factors controlling recovery from carbonate reservoirs and that our new method allows us to quantify it.