Psst… We received a tip…

You are standing in front of a wall-sized seismic image, and it is beautiful. It’s meant to tell you where to drill for oil, and a bright, continuous band of reflection sits right where oil should be, flat and clean. The funding is signed off. The rig is booked. Somebody orders the good coffee.

Eighteen months and a nine-figure budget later, the project reaches that depth and finds nothing. No reservoir. Just rock that was always going to be rock.

The prediction was based on an echo, a seismic wave that had unfortunately bounced more than once on its way back to the surface and arrived looking falsely perfect. Your processing software, assuming every wave bounced only once, drew it in as solid geology. You drilled a ghost.

This is the oldest trap in subsurface imaging, and a recently authored patent from the company that drills more holes than almost anyone on earth is aimed straight at it. The interesting part is how carefully they are trying to erase the ghosts without scrubbing away the real thing sitting right next to them.

The patent comes from Saudi Aramco, the state oil company headquartered in Dhahran. The system targets a problem called seismic multiples, which is the technical name for those echoes that fake a reflector. 

A reflector is a boundary underground where a seismic wave bounces back. A fake reflector shows a false rock boundary, which may indicate an underground layer that doesn’t actually exist, leading to bad oil drilling decisions.

A seismic survey works like sonar for rock. You thump the surface, sound travels down, bounces off layers, and comes back to receivers that record the timing. A wave that bounces once off a layer is a "primary," and primaries are what you want, because their timing tells you where the layer is.

But some waves bounce more than once. They reflect up, hit a shallower layer, reflect back down, then come up again. By the time they reach the surface they have spent so long underground that the software reads them as a single bounce off something much deeper. That phantom is the multiple, and it can paint a reflector that isn't there or smear one that is.

The standard fix has two moves. First you predict where the multiples should appear, using a method the patent names, a Marchenko-based prediction (a way of modelling all the echoes bouncing around in the shallow layers without having to map each one). Then you subtract that predicted echo from the recorded data.

The subtraction is where it gets delicate. The predicted echo never matches the real one exactly, so the software builds "matching filters," small adjustments that reshape the prediction to fit the recording before subtracting. The trouble shows up wherever a real primary and a fake multiple overlap in the same patch of data. The filter, trying to cancel the echo, eats into the genuine signal too. You remove the ghost and dent the reservoir.

The patent's move is to constrain those filters using the shape of the data. It builds something called a structure tensor, which is a way of measuring which direction the layers are running at every point, then smooths the filters along that grain rather than across it. The filter is told, in effect, to respect the local geology while it works.

That single detail is the whole invention. A filter that follows the structure removes echo energy while leaving primary energy alone in exactly the overlap zones where the old method did the most damage. 

Almost every step of seismic processing assumes the data is made of primaries only. It usually isn't. Multiples interfere most in deep water and in deep targets, which is precisely where the expensive hydrocarbons tend to be.

The cost is that a misread image sends a rig to the wrong depth, and offshore exploration wells routinely run into the tens or hundreds of millions of dollars each. Get the picture wrong and you either drill a dry hole or walk away from a real one because an echo masked it.

Existing subtraction methods force a bad trade. Tuned to remove every multiple, they damage primaries. Tuned to protect primaries, they leave multiples behind. In the overlap zones, the windowed math underneath the standard approach simply can't tell which energy is which.

Seismic multiple removal is a mature, crowded corner of geophysics, and the names are the usual heavyweights of subsurface imaging.

SLB (formerly Schlumberger) and Viridien (the French firm that was CGG until its 2024 rebrand) sit closest to this work. Both run advanced imaging centers that do adaptive subtraction at industrial scale, and both have been pushing higher-density imaging and more advanced processing as their main pitch. The structure-oriented constraint in the Aramco patent is an improvement on the technique they already sell.

TGS, now the largest seismic player after absorbing PGS, brings a vast data library and full-service imaging. Shearwater GeoServices and DUG Technology compete on processing horsepower, and DUG in particular markets heavy compute for exactly this kind of iterative filtering.

The direct category is seismic services, and the numbers depend heavily on where you draw the boundary. Narrow definitions covering acquisition, processing, and interpretation put 2025 at roughly US$9.3 billion, with Fortune Business Insights at US$9.58 billion and Stratistics MRC also near US$9.3 billion. Broader scopes run much higher, with SkyQuest reporting US$31.19 billion in 2024. 

The piece this patent touches is smaller and stickier, of processing and imaging software, where the work happens after the survey and the margins live in the algorithms. Oil and gas applications still anchor the category at around 50.9% of 2025 spending, with deepwater Brazil and Guyana tying survey programs directly to drilling decisions.

Then the frame widens. The same imaging that finds oil is now being repurposed to monitor carbon storage sites, which Mordor flags as the fastest-expanding use case, growing about 16.5% a year as projects like Northern Lights turn seismic monitoring into a regulatory obligation. 

The money in seismic is consolidating, and the deals tell you the survival strategy has shifted from owning the most data to owning the most defensible processing.

The flagship event is the TGS and PGS merger, an all-share deal valued at US$863.74 million and completed on 1 July 2024, creating the largest seismic player in the world with a combined market cap of around US$2.6 billion. TGS took on PGS's streamer fleet and data library, with post-merger synergies running at roughly US$100 million a year. The strategic read is that pure data acquisition has commoditized, so scale and integrated imaging are the only defensible positions left.

On the alliance side, SLB and Viridien have turned partnership into a standing business. In November 2025 the SLB–Viridien alliance signed with Egypt for a 95,000-square-kilometre ocean-bottom-node survey, alongside a separate US$44 million extension of Egypt's national upstream data platform. Viridien, the rebranded CGG, also won a 3,400-square-kilometre imaging project in Algeria's Berkine Basin for a Sonatrach-Occidental consortium.

The infrastructure shift is compute. SLB extended its AI collaboration with NVIDIA to accelerate seismic imaging and reservoir modelling, part of a move from small AI trials to large-scale deployment. ExxonMobil has leaned the same way, using its Discovery 6 supercomputer for 4D seismic imaging.

For the people staring at the wall-sized image, the change is small and concrete. Fewer phantom layers to argue about, fewer real ones smoothed into the background, and an image whose clean parts are more likely to be clean for the right reason. The mechanism is just a filter taught to respect the grain of the rock while it scrubs out echoes.

This week's patent is US 12631778, titled "Method and system for determining a location of hydrocarbon reservoir within subterranean region.”

Come find us on Instagram and LinkedIn while you decide.