HT-ATES systems work as a pair of wells, known as a doublet. When excess renewable heat is available, groundwater is extracted from one well, warmed at the surface through a heat exchanger, and injected into the second well, the so-called hot well. During the heating season the direction reverses, and the stored warm water is pumped back up. Ideally, the hot water spreads as an even bubble through the aquifer, and most of it returns through the well screen when it is needed. In practice, however, heat is lost at the thermal front and at the confining layers above and below the aquifer, and buoyancy adds a further complication that has long been recognized: hot, lighter water tends to rise and tilt the thermal front upward during storage.

The new research, led by Stijn Beernink of KWR Water Research Institute and Delft University of Technology together with colleagues, goes a step further. The team realized that density differences do not only drive buoyant flow in the aquifer — they also alter the pressure distribution inside the well itself. When a column of 90-degree water stands next to a column of cool groundwater at roughly 12 degrees, the lighter column gains hydrostatic pressure more slowly with depth. That means the pressure difference between the well and the surrounding aquifer shrinks as you descend, and can even vanish at a certain depth, below which essentially no injection occurs at all.

To quantify the effect, the researchers built axisymmetric numerical simulations using the SEAWATv4 code, which couples groundwater flow with variable density and viscosity, and represented the well as a column of extremely high hydraulic conductivity so that water could distribute itself freely along the screen. They tested injection temperatures of 15, 30, 60 and 90 degrees Celsius in a 40-meter-thick aquifer, running a realistic seasonal cycle of 90 days of injection, storage, extraction and idle time with an annual storage volume of 100,000 cubic meters per well. They also developed analytical equations that describe the same physics, allowing the maximum depth of injection and the peak flow at the top of the screen to be predicted from simple operational parameters.

The results are striking. As injection temperature rises, the flow distribution becomes increasingly lopsided, with the highest inflow concentrated at the top of the well screen — in some scenarios reaching several times the average flow rate, and with essentially no water entering the deepest part of the screen at 90 degrees under low-pressure conditions. The degree of nonuniformity depends on the ratio of pumping rate to aquifer permeability, which controls the injection pressure head. When that ratio is small, the pressure available to push water into the aquifer is modest, the buoyancy term dominates, and hot water infiltrates only through the upper portion of the screen, sometimes just the top 40 to 50 percent.

Viscosity makes matters worse in a self-reinforcing way. Water at 90 degrees is nearly 75 percent less viscous than water at 12 degrees, so the first hot water to enter the top of the aquifer creates a zone of low hydraulic resistance precisely where flow is already strongest. More water rushes through that zone, which heats it further and thins the water even more. In the simulations including both variable density and viscosity, flows at the top of the screen reached up to 3.7 times the average rate. The lowered viscosity also reduces the pressure head needed to inject a given volume, which paradoxically deepens the nonuniformity because less pressure is available to overcome the buoyancy gradient along the screen.

Crucially, the researchers found that the flow distribution is fundamentally asymmetrical between injection and extraction. During injection, buoyancy can choke off flow entirely below the maximum depth of injection. During extraction, however, the pressure difference between the light water column in the well and the denser ambient groundwater grows with depth and can never reach zero, so water is always drawn in across the full length of the screen. The practical consequence is sobering: a well that only injected through its upper half must, during recovery, pull cold ambient groundwater from the bottom half, diluting the extracted heat and directly reducing the thermal recovery efficiency of the system.

The nonuniformity also carries a mechanical warning. Maximum flow velocity at the borehole wall is a critical design parameter for preventing clogging of the screen and gravel pack, and the study shows that assuming uniform flow can underestimate this velocity by a factor of up to 3.7. A well designed on the assumption of even flow distribution might therefore be far more vulnerable to clogging in practice than its designers believed. The team’s screening of published HT-ATES storage conditions, spanning eight earlier studies, indicates that the levels of nonuniformity they quantified overlap substantially with real-world operating conditions, making this a genuinely practical concern rather than a numerical curiosity.

There are also direct implications for how HT-ATES systems are modeled and designed. Many previous simulation studies forced a uniform flow distribution across the well screen as a simplifying assumption. The new work shows this assumption introduces errors of up to about 7 percent in predicted recovery efficiency, in either direction depending on the aquifer’s vertical permeability, and larger errors can be expected in strongly heterogeneous aquifers. Allowing flow to distribute freely, the authors argue, is the safer approach. Where conditions are unfavorable, design remedies exist: operating wells closer to their maximum flow rate, using larger-diameter wells, or deploying multiple partially penetrating screens to spread injection and extraction more evenly across the aquifer.

The findings extend beyond geothermal storage. Similar density-driven nonuniform flow arises in carbon dioxide injection and in aquifer storage and recovery of freshwater in brackish aquifers, where comparable density contrasts of 5 to 25 kilograms per cubic meter occur. The analytical tools developed here — the maximum depth of injection and the maximum normalized flow — could help engineers in those fields as well. As heat networks across Europe and beyond look to the subsurface for seasonal storage at temperatures up to and beyond 90 degrees, this study makes clear that gravity itself must be treated as a design constraint: ignore buoyancy at the well screen, and a significant share of the stored warmth may never come back.

Subject of Research: Buoyancy-induced nonuniform groundwater flow through well screens in high-temperature aquifer thermal energy storage

Article Title: How density and viscosity differences cause nonuniform flow distribution across well screens and impact heat recovery of HT-ATES systems

Article References: Beernink, S., Hartog, N., Vardon, P. J., & Bloemendal, M. (2026). How density and viscosity differences cause nonuniform flow distribution across well screens and impact heat recovery of HT-ATES systems. Hydrogeology Journal. https://doi.org/10.1007/s10040-026-03154-8

Image Credits: AI Generated

DOI: 10.1007/s10040-026-03154-8

Keywords: HT-ATES, aquifer thermal energy storage, well screen, buoyancy flow, groundwater density, viscosity, heat recovery, well hydraulics, thermal energy storage, geothermal energy, SEAWAT, clogging

The study, published in Environmental Earth Sciences, focuses on the Upper Jurassic Malm aquifer, a karstified carbonate formation lying more than 1,400 meters below the surface of the South German Molasse Basin. The Malm is the workhorse of Bavarian geothermal energy: up to 600 meters thick, composed of limestones and dolomites riddled by multiple episodes of karstification, and cut by an extensive network of normal faults created as the Alpine orogeny bent the lithosphere. Its waters reach 80 to 85 degrees Celsius, hot enough to feed district heating grids across the region. But the same fractures and faults that make the reservoir productive also raise a persistent engineering question: how much can operators push fluid through this underground before the rock itself pushes back?

To answer that question, the researchers built a three-dimensional synthetic geological model inspired by the faulted Malm reservoir north of Munich, incorporating a real fault interpreted from the GRAME seismic survey, one of the largest urban geothermal seismic campaigns ever conducted in Europe, which imaged roughly 170 square kilometers of the city’s subsurface in 2015 and 2016. Into this model they placed two competing operational schemes. The first was the familiar hydrothermal doublet: a production well drawing ambient-temperature reservoir fluid continuously, paired with an injection well returning the cooled water year-round. The second was an HT-ATES triplet, a three-well arrangement in which a bidirectional hot well injects water heated to 140 degrees Celsius during six-month summer periods and produces it again in winter, a cold well reinjects the cooled fluid at 50 degrees Celsius, and a buffer well balances the mass by producing ambient-temperature water.

The simulations were run with the open-source PorousFlow module of the MOOSE framework, coupling single-phase saturated fluid flow, heat transport, and linear elastic rock mechanics into a fully coupled thermo-hydro-mechanical model. Water properties were calculated with the IAPWS-IF97 equation of state, and the model was subjected to the strike-slip stress regime of the Molasse Basin, with a maximum horizontal stress gradient of 34.6 kilopascals per meter oriented due north. Each simulation ran for ten years, spanning twenty alternating summer and winter cycles for the triplet and a decade of continuous operation for the doublet, with flow rates of 150 liters per second at every well. The team then evaluated the response of the reservoir and the fault using slip-tendency analysis, a standard geomechanical measure of how close a fault surface is to frictional failure.

The first striking result concerns pressure. Because the Malm aquifer is extraordinarily permeable, around 1.48 by 10 to the minus 12 square meters in the model, pressure perturbations from injection and production remained tiny, on the order of plus or minus 0.1 megapascal after ten years. That means poroelastic effects, the stress changes caused directly by fluid pressure, played almost no role in the mechanical response. Instead, the geomechanics of the reservoir turned out to be governed almost entirely by thermoelasticity: the expansion of rock where hot water is injected and its contraction where cold water is injected. Heating near the hot well increased compressive normal stress and even stabilized nearby fault sections, while cooling near the cold and injection wells reduced normal stress and pushed the fault closer to slipping.

The differences between the two operating styles were substantial. After ten years, the vertical displacement at the top of the reservoir reached 28 millimeters at the doublet’s injection well, compared with 19 millimeters at the triplet’s cold well and just 8 millimeters at its hot well. The hot plume and the cold plume also stratified by density: buoyant hot water accumulated near the top of the reservoir, where temperatures at the hot well reached 132 degrees Celsius, while denser cold water sank and spread along the base, where only 118 degrees was recovered. Because the doublet injects cold water continuously rather than cyclically, and injects twice the total volume over the decade, its cumulative thermal perturbation was larger, and its stress-field disturbance grew monotonically rather than partially recovering during rest periods.

The fault-stability comparison was even more decisive. The complexly oriented fault surface, with spatially varying strike and dip, showed initial normalized slip-tendency values ranging from 0 to 0.76, highest where the fault faces the direction of maximum horizontal stress at a favorable 30-degree angle. After ten years of operation, the hydrothermal doublet raised the maximum normalized slip tendency by 0.137 near its injection well, concentrated in the deeper fault sections where the cold plume pooled and the well sat closest to the fault. The HT-ATES triplet, by contrast, increased slip tendency by only 0.03, roughly four times less. Neither scenario crossed the critical threshold of 1.0, meaning no fault destabilization was predicted even under deliberately conservative assumptions, including critically stressed crust, zero cohesion, and wells placed a mere 178 meters from the most vulnerable fault segment, far closer than real projects would ever site them.

Heterogeneity proved to be the wild card. The Malm’s karst networks mean permeability can vary by orders of magnitude between layers, and the team tested moderately and extremely heterogeneous end-member scenarios spanning more than three orders of magnitude in permeability contrast. In the extreme case, a single 30-meter layer at the top of the reservoir, representing just 5 percent of its volume, carried 45 percent of the injected fluid. This concentration of flow reduced thermal recovery efficiency by roughly 10 percent compared with the homogeneous case, which achieved about 90 percent recovery after ten years, and it slightly increased fault slip-tendency changes to 0.04. Where the high-permeability layers sat deeper in the reservoir, closer to the slanting fault, the destabilizing cold-front stress changes shifted downward with them, a reminder that the three-dimensional geometry of karst zones matters as much as their existence.

To probe the long term, the team extended the extremely heterogeneous triplet scenario to fifty years. By then the cold front had finally reached the fault surface, and the maximum normalized slip-tendency change had grown from 0.04 to 0.2, with absolute slip tendency peaking at 0.9 in the most critical sections. Still below the failure threshold, but uncomfortably close, and a clear signal that decades of repeated cold-water cycling gradually erode the geomechanical safety margin. The authors are careful to note the caveats: their linear elastic framework cannot capture fatigue, damage accumulation, or plastic deformation across cycles; permeability was held constant; the model is synthetic rather than site-calibrated; and the two schemes serve different engineering purposes, so a rigorous comparison would need to normalize for delivered thermal energy.

Even with those limits, the implications for Munich and cities like it are concrete. Thermoelastic stress changes of this kind announce themselves in the field through falling injection pressures and rising injectivity near cold wells, exactly the locations where all major induced seismic events in the Munich area have historically clustered, making the vicinity of injection wells the natural place to focus monitoring with borehole strainmeters, extensometers, or distributed acoustic sensing. The predicted displacements, up to 28 millimeters at reservoir depth nearly 1.5 kilometers down, are expected to attenuate strongly toward the surface and are readily detectable with standard downhole instruments. As Munich pushes toward a climate-neutral heat supply and looks to store summer surpluses, whether from geothermal, industrial waste heat, or solar thermal, in its own bedrock, this study suggests that seasonal storage triplets can do the job while shaking the faults beneath the city less than the geothermal plants that have operated there for years.

Subject of Research: Comparative thermo-hydro-mechanical modeling of high-temperature aquifer thermal energy storage and hydrothermal doublet operation in the faulted Malm geothermal reservoir near Munich

Article Title: Thermo-hydro-mechanical response of the operation style in a faulted Malm reservoir: Storage triplet vs. hydrothermal doublet

Article References: Dashti, A., Stricker, K., Hörbrand, T., Habibi, R., Beichel, K., & Kohl, T. (2026). Thermo-hydro-mechanical response of the operation style in a faulted Malm reservoir: Storage triplet vs. hydrothermal doublet. Environmental Earth Sciences, 85(16), Article 404. https://doi.org/10.1007/s12665-026-13141-7

Image Credits: AI Generated

DOI: 10.1007/s12665-026-13141-7

Keywords: geothermal energy, HT-ATES, aquifer thermal energy storage, hydrothermal doublet, Malm aquifer, Molasse Basin, Munich, fault stability, slip tendency, thermoelasticity, induced seismicity, district heating