Underground hydrogen storage modelling: the reservoir questions to ask first

Underground hydrogen storage modelling has to answer a different question from CO₂ storage: not "will the gas stay down for ever?" but "how much can we get back, how fast, how pure, and for how many cycles?" Before any detailed simulation, five reservoir questions decide whether a site is worth pursuing: porous rock or salt cavern, how much cushion gas, how the store behaves under repeated cycles, how much hydrogen is lost, and whether the seal and the wells hold. This guide takes them in turn, with the public research behind each.

Why hydrogen storage is not CO₂ storage in reverse

Hydrogen is stored to buffer renewable electricity: produced when power is plentiful, stored, and withdrawn when it is needed [1]. That means the store is cycled, perhaps seasonally, perhaps much more often, and recovery is the point. A CO₂ store is judged on what stays down; a hydrogen store on what comes back up.

Hydrogen also behaves unlike either CO₂ or natural gas. Its very low density means it takes much more volume than natural gas for the same energy, and its low viscosity makes it highly mobile but a poor displacer of brine, which raises the risk of viscous fingering that can leave pockets of hydrogen that cannot be recovered [2]. Its buoyancy makes it gather in a cap directly under the seal [2]. Much of the CO₂ storage toolkit transfers (seal checks, pressure limits, legacy wells, monitoring [3]), but the answers change.

Hydrogen storage in porous rock vs salt caverns

There are two families of store. Salt caverns are solution-mined voids in thick salt; porous stores are saline aquifers and depleted gas or oil fields [1]. Heinemann and co-authors compare them as follows [2]:

The rest of this guide is about porous stores, where the reservoir questions are hardest. The European HyUSPRe project, coordinated by TNO and run from 2021 to 2024, was set up precisely to assess the feasibility, risks and potential of large-scale hydrogen storage in porous reservoirs in Europe, using laboratory experiments and integrated modelling of geochemical, microbial and flow processes [4].

Cushion gas and cyclic injection

A store never empties. Part of the gas, the cushion gas, stays in place to hold up the pressure so the working gas can be withdrawn at the rates needed. For natural gas storage the rule of thumb is a cushion of 40–70 %; for hydrogen the right ratio is not yet known and is likely to be specific to each site and project. Other gases have been proposed as cushions instead: methane or nitrogen to cut cost, or CO₂ to cut emissions [2]. A cushion of a different gas brings its own question: how much mixes with the hydrogen, and how pure is the gas that comes back?

Cycling also matters mechanically. Repeated injection and withdrawal change the stress in the storage complex, and the rate of cycling sets the rate of deformation; Heinemann and co-authors flag possible compaction, fault reactivation and new fractures in the caprock, and note that relative permeability may change from cycle to cycle, as seen in CO₂ studies [2]. A model that runs only one fill and one withdrawal will miss this.

Hydrogen loss mechanisms

Hydrogen that goes in and does not come back is lost money and, sometimes, a safety or quality problem. The main routes are [2][1]:

• Microbial. Methanogens turn hydrogen into methane, sulphate reducers produce hydrogen sulphide, and homoacetogens produce acetate. Field experience comes mostly from storing hydrogen-rich town gas, not pure hydrogen. It ranges from no detected hydrogen consumption at Beynes, France, to a hydrogen decrease of up to 17 %, with less CO₂ and more methane, over a seven-month cycle at the Lobodice aquifer store in what is now the Czech Republic [2], where methane-producing microbes were identified as the likely cause [5]. In Austria's Underground Sun.Storage project, which added hydrogen to stored natural gas, about 3 % of the injected hydrogen was judged to have been converted to methane [2]. Reservoir temperature, salinity and the available sulphate and CO₂ set how active the microbes can be.
• Geochemical. Hydrogen can react with sulphur species and pyrite to form hydrogen sulphide, which degrades the gas and corrodes equipment; carbonate minerals may limit this [2].
• Diffusion and dissolution. Despite hydrogen's high diffusivity, diffusion losses are estimated at only about 0.1–1 % over a site's lifetime [2]. Hydrogen dissolves poorly in brine, so losses by dissolution are expected to be small [2].
• Trapping and mixing. Residual hydrogen left in the pores as brine moves back in, viscous fingering, and mixing with cushion gas or the gas left in a depleted field all reduce what can be produced at the purity required [2][1].

Caprock and well integrity

The seal has to hold a gas far more buoyant than natural gas or CO₂. At 1,000 m depth, hydrogen's buoyancy forces are about three times those of CO₂, so even a moderate hydrogen column can push hard on the seal [2]. At the time of that review there were no experimental data on hydrogen breakthrough pressures in shales or other caprocks [2], so capillary entry pressure is a real uncertainty, and screening should test how sensitive the answer is to it rather than assume a value.

Wells are the other path out; at natural gas stores, loss of well integrity has been the source of most leakage events [6]. Every well that reaches the store, active or abandoned, is exposed to hydrogen and to cyclic pressure and temperature changes. Hydrogen can enter common low-carbon casing steels and cause premature cracking, may migrate through cement, and can degrade elastomer seals [6]. A well engineer should check the materials of each well for hydrogen service at the specific site, and old wells inside the pressure footprint should be inventoried, as for CO₂ storage [3].

Underground hydrogen storage modelling: what the model must resolve

Putting these together, a useful screening model for a porous store needs to:

1. Represent the layering and facies that control where a light, mobile gas goes, rather than averaging them away; fingering and gravity segregation are set by the contrast between layers.
2. Simulate many cycles, not one, with the operating window bounded by cushion pressure below and seal and fault stress above.
3. Track gas composition (hydrogen, cushion gas, any native gas) so recovery and purity can be reported.
4. Add first-order loss terms for microbial and geochemical consumption, with ranges, until site samples allow better.
5. Carry the uncertainty through as an ensemble, so the result is a range of recoverable working gas, not a single figure.

This mirrors the integrated approach of HyUSPRe, which coupled flow and transport with geochemical and microbial processes and lab experiments [4], and the open problems set out in the 2021 reviews [2][1].

The questions to ask first

• Is there a cavern option, or only porous rock? Is the store to cycle daily, weekly or seasonally?
• For a depleted field: what gas is left, and will it mix with the hydrogen?
• What cushion gas, how much, and what purity does the customer need back?
• What are the temperature, salinity, sulphate and pyrite content: how active could microbes and reactions be?
• How thick and continuous is the seal, and what is known of its entry pressure for hydrogen?
• Which wells reach the store, and in what condition?

What KAI offers today and what it does not

Plainly: KAI has not carried out a hydrogen storage project and has no published hydrogen storage case. In KAI-GeoEnergy, hydrogen and CO₂ storage studies are in build or available on request. The live preview today is the Reservoir Lab for geothermal doublets, which runs in the browser and queues full simulations on the KAI engine; fault-slip, subsidence and storage modules are still in build.

The published methods that would carry over to a hydrogen study are the ones that keep heterogeneity from plug to reservoir:

• Layering between wells. The Heterogeneity Lab builds facies ensembles conditioned on every log, on Dutch BRO borehole data and TNO GeoTOP, and shows what a simplified model costs in flow.
• Permeability and kv/kh from core. In the Zeerijp-3A case, 78 plug pairs give a median kv/kh of 0.95, while bedding calibrated on the same plugs and upscaled by flow gives 1 : 20 to 1 : 360 at 1 m for 20–40 % mud (bedding simulated). For a buoyant gas, that difference changes how fast it reaches the seal.
• Ensembles with P10–P90. The Utah FORGE case runs flow, heat and fracture opening over 30 years in a 20-member ensemble, with the injection limit set by the minimum stress (simulated, and for geothermal, not storage). It also shows why coarse models mislead: the usual coarse model is +11 °C too optimistic at year 30, and flow upscaled from the fine model cuts that to +2 °C.

KAI works remotely with your team and data; it does not drill, hire equipment or run field acquisition, and it does not run laboratory or microbial tests. If you are weighing a porous hydrogen store, read how KAI approaches subsurface energy or see what KAI-GeoEnergy does today, and tell us what data you hold.