New inversion algorithm reshapes Earth’s deep electrical maps

Magnetotellurics (MT) has long been one of the few techniques capable of revealing Earth’s electrical structure from the surface to depths of hundreds of kilometers. It measures natural fluctuations in Earth’s electric and magnetic fields, providing information on Earth’s temperature, fluids, and rock composition.

Despite its scientific power, MT imaging faces a major obstacle: computation. Converting field data into a three-dimensional conductivity map requires inversion, a process that repeatedly adjusts millions of resistivity parameters until simulated and observed data match within error limits. For large surveys, this can take weeks or even months.

The 2025 work by Arun Singh and Rahul Dehiya presents a major step forward. Their Radiation Boundary (RB) inversion scheme reduces the number of calculations needed to solve the problem without losing precision. It combines a coarse global simulation with a focused, high-resolution model applied only where real data are available.

This hybrid setup eliminates parameters outside the active survey area, sharply reducing both memory use and runtime. The algorithm allows researchers to model an entire continent on a powerful workstation rather than a computing cluster.

Beyond saving time, the method improves accuracy in the deepest layers of the model. Conventional solvers tend to blur or shift anomalies at depths greater than 50 km (31 miles), but the RB approach retains sharper contrasts and more geologically realistic results.

The science behind the Radiation Boundary method

Traditional MT inversion codes use a single large grid that extends beyond the survey boundaries to account for electromagnetic field decay. These additional “padding” cells often contribute little to accuracy but increase the computational burden.

The Radiation Boundary technique divides the process into two stages. First, a coarse mesh covers the entire region and estimates boundary fields. Then, a fine mesh focuses only on the data-constrained zone and applies the boundary values from the coarse simulation.

This design means the algorithm computes what matters most. The fine mesh can follow the shape of the survey rather than being a uniform block, removing thousands of unnecessary cells. The system of equations becomes smaller and easier to solve while preserving numerical stability.

In comparative tests, the RB solver achieved the same or better accuracy at a solver tolerance 100 times larger than the conventional finite-difference approach. This means it can terminate earlier in each iteration, saving time without compromising accuracy.

Because the RB concept modifies only the forward and gradient steps, it can be incorporated into existing inversion frameworks such as ModEM or jif3D with minimal code changes. The result is a general improvement in efficiency available to many researchers.

Testing the method on synthetic models

To assess performance, Singh and Dehiya used the “Rubik model,” a synthetic three-dimensional checkerboard made of alternating conductive and resistive blocks in a 100 ohm-meter half-space. Simulated MT data from 29 stations and 12 frequencies ranging from 10 to 10 000 seconds were inverted using both conventional and RB algorithms.

The tests produced clear results. The RB1.0 version ran about four times faster than the standard finite-difference solver while achieving identical data misfit. RB2.0, which supports non-rectangular meshes, achieved a speedup of about 6.6 times. The RB2.1 variant recovered deep structures with improved accuracy.

At depths below 50 km (31 miles), the RB algorithm captured sharper boundaries between resistive and conductive zones. Conventional inversions smoothed or distorted these features, particularly where the electromagnetic signal was weak.

Noise tests confirmed that the RB method remained stable when realistic errors were added to the data. Even with 3% Gaussian noise, the modelled and observed responses matched closely. The algorithm also revealed that coarse-mesh inversions can trap results in local minima, while direct fine-resolution inversion on the full data set avoids this risk.

These controlled experiments demonstrated that the RB approach not only accelerates computation but also enhances physical reliability. It delivers a more accurate picture of subsurface structures without requiring extreme precision settings.

Revealing southern Africa’s hidden structure

The algorithm’s capabilities were tested further on the Southern African Magnetotelluric Experiment (SAMTEX), one of the largest MT projects ever undertaken. The survey collected data from about 700 stations across South Africa, Botswana, and Namibia, spaced 20–60 km (12–37 miles) apart and covering periods from 0.01 to 10 000 seconds.

Previous inversions using codes such as ModEM and jif3D were limited to regional subsets because of computational constraints. The new RB2.0 algorithm, however, handled the entire data set using meshes with cell sizes of 30, 15, and 10 km (19, 9, and 6 miles).

At 10 km resolution, the normalized root-mean-square (nRMS) misfit fell to 1.8, compared with 4.0 for the coarsest grid. The computation ran to completion on a workstation equipped with a 2.9 GHz Xeon Gold processor and 192 GB of memory, demonstrating the method’s efficiency.

The high-resolution results unveiled several well-known geological features with exceptional clarity. The Kaapvaal Resistor, an Archean cratonic block, appeared as a deep resistive core reaching about 150 km (93 miles). The Damara Belt, a conductive suture between the Congo and Kalahari cratons, was sharply defined and continuous. The Molopo Farms and Southern Kaapvaal Conductors emerged as subvertical zones aligned with ancient shear structures.

At shallower depths, smaller conductive belts and resistive lenses became visible, consistent with mineralized zones mapped by geological surveys. These results confirmed that improved numerical efficiency translates directly into better geological interpretation.

What the results mean for geophysical modelling

The study highlighted several practical lessons for researchers. One concerns solver tolerance, a setting that determines when numerical iterations stop. For the RB solver, a tolerance of 10⁻⁷ achieved the same accuracy that the finite-difference solver required at 10⁻⁹, resulting in large time savings.

Another finding relates to the inversion workflow. Many analysts perform a coarse-mesh inversion first and then refine a subset at higher resolution. The tests showed that this approach can trap the algorithm in a local minimum, reproducing coarse-scale patterns rather than discovering new details. Starting directly from a uniform half-space on a fine grid produced better, more realistic results.

The RB method also makes it practical to run multiple inversions with different starting models or regularization parameters, allowing sensitivity tests that were previously too costly. This flexibility helps researchers evaluate which parts of a model are genuinely constrained by data.

When compared with existing software, the RB-enhanced AP3DMT code produced finer contrasts and sharper anomaly boundaries, especially below 150 km (93 miles). The Damara Belt extended farther eastward than in earlier models, aligning closely with independent seismic interpretations.

Overall, the results demonstrate that improved numerical treatment of boundaries can yield substantial gains in accuracy, not only in speed.

Broader significance for Earth science

Rapid, large-scale inversion has wide implications for understanding the deep Earth. Modern MT networks on every continent produce vast volumes of data, but full 3-D analysis has been limited by computing power. The RB method removes that barrier, allowing researchers to model entire lithospheric domains at high resolution.

This capability supports investigations into mantle composition, crustal fluids, and geothermal gradients. It also improves studies of subduction zones, mineral exploration, and tectonic evolution. In geothermal regions, fine-scale conductivity models help locate heat-bearing formations at depths of 5–20 km (3–12 miles).

The Radiation Boundary principle is not limited to passive MT data. It can be adapted to controlled-source electromagnetic and induced-polarization techniques, where forward modelling time dominates. It also provides a foundation for joint inversion, integrating MT with seismic and gravity data for more complete crustal models.

For geological surveys and research institutions, this means faster turnaround times and lower computational costs. Full continental-scale inversions can now be completed on standard high-end computers instead of supercomputers, expanding access to detailed subsurface mapping.

By linking numerical innovation with geological application, the RB method demonstrates that efficiency and scientific depth can advance together.

The path forward

The Radiation Boundary inversion scheme represents a major step in geophysical computing. It reduces runtime by up to sevenfold, sharpens model resolution, and allows realistic continental-scale inversion on accessible hardware.

Future improvements will likely include GPU acceleration and adaptive meshing, further increasing speed. Integration with parallel frameworks could bring near–real-time inversion for regional monitoring networks.

If adopted widely, the RB framework could become a standard approach in electromagnetic geophysics, much like finite-difference solvers in the early 2000s. Its ability to balance efficiency and physical accuracy sets a new benchmark for large-scale modelling.

By transforming massive electromagnetic data sets into detailed conductivity maps, the Radiation Boundary method brings scientists closer to a clear, three-dimensional understanding of Earth’s deep architecture. It shows that the next leap in exploration may come not from bigger computers, but from smarter algorithms.

In doing so, Singh and Dehiya’s work redefines the limits of what geophysicists can see beneath the surface—an innovation that turns computational mathematics into a new lens on the planet itself.

References:

¹ An efficient 3-D inversion scheme for continental scale magnetotelluric data – Arun Singh et al. – Geophysical Journal International – September 18, 2025 – https://doi.org/10.1093/gji/ggaf371 – OPEN ACCESS

Reet Kaur

I’m a science journalist and researcher at The Watchers, contributing to the Epicenter edition, where I cover peer-reviewed scientific research and emerging discoveries across Earth and space sciences. With a background in astronomy and a passion for environmental science, I’ve worked in shark and coral conservation in Fiji, conducting reef and shark-behavior research, contributing to mangrove restoration, and earning PADI Open Water and Coral Reef Certifications. I bring a blend of scientific rigor and storytelling to illuminate the discoveries shaping our planet and beyond.

This author does not have any more posts.