The research team investigated the structures using an unusually detailed three-dimensional geological framework built from more than 4,000 geologically logged boreholes, spaced roughly 50 to 100 metres apart. The model also incorporated geophysical surveys, groundwater observations and structural mapping, allowing the researchers to represent the geometry of the dykes and the surrounding rock in much greater detail than is typical for regional groundwater studies. The work focused on the Brockman Syncline, a large geological fold within the Hamersley Basin that extends for about 60 kilometres across the Pilbara. The area contains extensive iron-ore deposits and a complicated network of faults, shales, iron formations and intrusive dolerite bodies. Because groundwater flow is concentrated in some highly permeable mineralised layers but restricted by others, even a narrow geological feature can have consequences across hundreds of metres. The researchers used that natural complexity as a real-world laboratory for testing how dykes influence groundwater movement.

At each site, engineers pumped groundwater from a bore positioned close to, or across, a dyke while monitoring the response in surrounding wells. The first test ran for 120 hours at a rate of 60 litres per second; the second lasted 145 hours at 50 litres per second. These were not small laboratory experiments. They were field-scale disturbances large enough to generate broad cones of drawdown—the downward depression in hydraulic head created when water is extracted from an aquifer. In a uniform aquifer, the cone would tend to spread in a relatively predictable and symmetrical pattern. Instead, the monitoring data revealed strong asymmetries and delayed responses in wells on the opposite side of the dykes from the pumping wells. At the first site, the pumping bore fell by about 34 metres, while nearby observation wells recorded declines of roughly 11 and 8 metres. Farther away, the response arrived later and was much weaker: one well showed about 2 metres of drawdown, while another recorded less than 1 metre.

The second test produced an even larger local response. Water levels in the pumping bore declined by approximately 48 metres, and the nearest monitoring bore fell by about 20 metres. Yet a well on the opposite side of the dyke registered only around 0.2 metres of drawdown, despite being positioned where a more freely connected aquifer would have transmitted a stronger signal. Another well, farther from the pumping bore, experienced roughly 1 metre of decline. Such delayed and attenuated responses are a hydraulic signature of partial compartmentalisation. The pressure disturbance was not stopped completely, but it was slowed and weakened as it crossed the intrusive rock. The models could reproduce these patterns only when the dykes were explicitly represented. When the first site was modelled without its dyke, one monitoring well showed no drawdown at all in the simulation, even though the field data clearly recorded a response. The mismatch indicated that the structure was not a minor geological detail but a dominant control on pressure propagation.

To translate those observations into estimates of rock properties, the researchers built transient groundwater-flow models using MODFLOW-NWT, a numerical code widely used to simulate aquifer behaviour. The models represented the geological sequence as 17 layers, with hydraulic properties allowed to vary between formations and, in more advanced versions, from place to place within them. The crucial parameter was horizontal hydraulic conductivity, denoted Kh, which measures how readily water can move horizontally through a material. Higher Kh values indicate easier flow; lower values indicate greater resistance. The dykes were represented with a horizontal-flow-barrier formulation that modifies the hydraulic connection between neighbouring model cells. This approach is designed for thin features that would otherwise be difficult to resolve on a computational grid, although it assumes that vertical flow through the dykes is negligible. The team tested five levels of model complexity, ranging from homogeneous formations with no dykes to models containing spatially variable properties in both the surrounding formations and the dykes.

The best-fitting models were calibrated automatically against the observed water-level changes using PEST-HP and a Gauss–Levenberg–Marquardt optimisation algorithm. In essence, the software repeatedly adjusted the model’s hydraulic parameters until its simulated drawdown curves approached the field measurements. The researchers also used pilot points, distributed locations at which hydraulic conductivity could be estimated and then interpolated across the model using geostatistical techniques. This allowed the simulations to capture heterogeneity caused by changes in mineralisation, lithology and structure rather than forcing each formation to behave as a single uniform block. Regularisation methods constrained the fitted values so that they remained spatially smooth and physically plausible. The results were striking: simply adding dykes to otherwise homogeneous models reduced the calibration objective function by roughly an order of magnitude. Adding spatial variability to formations, dykes or both improved the fit further, reducing the objective function by one to two orders of magnitude compared with models that ignored the structures.

The conductivity estimates revealed that not all dykes were equally impermeable. At the first site, the western dyke had Kh values between approximately 1.9 × 10⁻⁶ and 1.8 × 10⁻⁵ metres per day, making it an especially strong barrier compared with the more permeable iron-rich formations around it. The eastern dyke was more conductive, with estimates ranging from 3.3 × 10⁻³ to 2.6 × 10⁻¹ metres per day, but it still exerted a measurable influence on the drawdown pattern. At the second site, the dyke’s Kh ranged from 1.3 × 10⁻³ to 2.8 × 10⁻³ metres per day. Across all calibrated models, however, the general contrast remained: the dykes were one to eight orders of magnitude less conductive horizontally than the surrounding formations, apart from already low-permeability shale units. The variation likely reflects differences in mineral composition, weathering, fracturing and the condition of the contacts between the dykes and their host rocks. A solid, clay-weathered dyke may block flow, while a fractured dyke or its damaged margins could potentially create pathways along the structure.

One of the study’s most important insights concerned uncertainty. A model can produce a close match to observations while still assigning misleading confidence to its parameters. When the surrounding formations were treated as homogeneous but the dykes were allowed to vary, the calculated uncertainty in dyke conductivity sometimes fell sharply. At first glance, that might suggest the pumping tests had tightly constrained the properties of the dykes. But the researchers found that the flexible dyke parameters were often absorbing errors caused by oversimplified surrounding formations. In other words, the model was using the dyke as a mathematical patch for geological complexity it had not been allowed to represent elsewhere. When heterogeneity was included in both the formations and the dykes, uncertainty reduction became much more localised. The dyke properties were most tightly constrained near monitoring wells and in sections where measurable drawdown occurred. Regions far from observations remained substantially less certain, demonstrating that apparent precision can depend heavily on where data are collected.

That result has immediate consequences for mining and water management. Dewatering an open-pit mine can draw groundwater toward pumping systems, alter hydraulic gradients and affect connected aquifers. If a low-permeability dyke is omitted from a groundwater model, simulations may predict that drawdown will spread farther and faster than it really does on one side of the structure. Conversely, if a fractured dyke or its surrounding damage zone is treated as an impermeable wall, the model could underestimate preferential flow and groundwater inflows. The new work therefore argues for a balanced approach: geological structures must be represented explicitly, but model complexity must also be matched to the amount and distribution of available data. The study’s posterior-variance maps offer one way to identify which parts of a dyke are well constrained and which require additional monitoring. That information could guide the placement of future observation wells, helping engineers collect data where it would most reduce uncertainty rather than simply adding more measurements near existing infrastructure.

The researchers caution that the findings describe cross-dyke flow and do not fully resolve the possibility of water moving along dyke margins or through fractured sections. The horizontal-flow-barrier representation simplifies the geometry of a feature that may vary in thickness, composition and weathering state. The uncertainty analysis also relies on a locally linear approximation, which may understate uncertainty in a strongly nonlinear groundwater system. Nevertheless, the field evidence and simulations converge on a clear message: thin geological intrusions can control groundwater behaviour at the scale of entire pumping tests. Dykes are not merely lines on a geological map; they can reorganise the movement of pressure and water through the subsurface. By combining high-resolution geological reconstruction, long-duration pumping experiments and automated numerical inversion, the study provides one of the clearest field-scale demonstrations yet of how such structures function as hydraulic barriers. For the mines, aquifers and communities that depend on reliable groundwater predictions, what lies beneath the surface may be just as important as the water being pumped from it.