Old GPS Data Flagged Where a Giant Kamchatka Quake Would Hit

Gareth Funning, a geophysicist at the University of California, Riverside, puts the problem simply: “Earthquakes capture headlines when they happen, but for years beforehand the fault is quietly accumulating strain. This strain can be measured.” The trouble has always been turning that measurement into a location. Along a coastline like Kamchatka’s, which patch of the plate boundary is locked and loaded, and which is quietly creeping along without incident? Funning and Axel Periollat, both at UC Riverside, tested an answer using an unusual instrument: satellite positioning data recorded years before anyone knew a magnitude 8.8 earthquake would strike offshore Kamchatka in July 2025.

They built a probability map of the Kamchatka megathrust, the boundary where the Pacific plate dives beneath the Okhotsk microplate, from measurements of how the ground had crept, or hadn’t, over the preceding years. Older approaches to this problem typically start by assuming where a rupture zone should sit, based on past earthquakes, then shrink that guess until it fits the ground motion. The new algorithm works the other way. It systematically tests combinations of locked and creeping patches across a mesh of 1,638 fault elements and keeps whichever combinations actually explain the observed motion, without assuming in advance where the sticking points are.

The exercise is a look backward, not a forecast issued ahead of time. Periollat and Funning built their model in 2026, after the 2025 earthquake had already happened, using satellite data that stopped well before the rupture. Because none of that older data was tuned to fit the 2025 event, whatever locked patch the model flagged is a fair test of the method, and a striking one. Its principal locked cluster, a run of 618 fault elements along the coast, is where both the 2025 earthquake and its 1952 predecessor began. The 2025 hypocenter sat about 35 kilometers down inside that cluster; the 1952 hypocenter, about 21 kilometers down, inside the same one.

The strain arithmetic lines up almost as well as the location does. Assuming a plate convergence rate of about 80 millimeters a year, the 73 years since 1952 would have loaded the locked patch with roughly 5.8 meters of slip debt. Converting that debt into an equivalent earthquake size gives a magnitude between about 8.85 and 8.99, depending on how stiff the surrounding rock is assumed to be, close to the magnitude 8.84 that Kamchatka actually got in 2025. Scaling only the roughly 68 percent of the patch that ends up locked in a typical model run gives a slightly smaller figure, magnitude 8.80, about 90 percent of the real event’s moment. None of this arithmetic tells the researchers when a fault will go. It only says where the energy has been building.

A Sticking Point That Keeps Sticking

Zooming into the big locked cluster turned up something more specific still. Splitting it into five smaller pieces by how consistently each behaved across the sampled models, the team found one small subcluster, locked in nearly every model run at a probability near 0.98, sitting close to where both the 1952 and 2025 ruptures began. Locking that subcluster alone, or together with its two closest neighbors, reproduces the observed ground deformation about as well as locking the entire 618-element patch, while committing to a far smaller area. That is evidence, Periollat and Funning argue, for a persistent asperity: a stubborn patch of rock that keeps re-locking and re-breaking across multiple earthquake cycles, rather than a fault that ruptures wherever happens to be most stressed at the time. Periollat says the overlap affirmed the underlying idea: “We had an idea where the strain was accumulating based on a relatively limited data set. Seeing it work so well confirmed that this approach has real potential.”

Same Patch, Different Tsunami

Locating the locked zone, though, does not explain everything about what happens once it finally lets go. The 1952 earthquake produced a far larger tsunami than 2025 did, even though both events broke roughly the same deep patch of megathrust. The difference, Periollat and Funning suggest, traces to how close the rupture came to the surface. Separate research led by Chi-Hsien Tang at Tohoku University, using satellite radar and tsunami-buoy records rather than GNSS locking models, reached a compatible conclusion: the main slip in 2025 stayed confined to depths of roughly 13 to 46 kilometers, well short of the trench, and slip models pushed closer to the seafloor predicted tsunami waves arriving noticeably earlier than the ones actually recorded. Limited shallow rupture, in other words, is a large part of why the 2025 tsunami fell short of what a magnitude 8.8 earthquake might otherwise be expected to produce, even at a site with a track record of far worse.

The locking model has its own soft spots. Two of its six locked clusters sit far enough north that they depend heavily on a single GNSS station and vanish from the model once that station is excluded, a pattern the team treats as a likely artifact rather than real coupling. A deeper cluster gets a similar discount, flagged as an edge effect of the fault mesh rather than a genuine sticking point. And because every GNSS station used sits on dry land, the shallowest, trench-adjacent part of the megathrust, exactly the part that governs tsunami size, remains the part the model can see least well. Periollat and Funning are already testing the same approach on subduction zones in Japan, Mexico, New Zealand and the Pacific Northwest, and on California’s Hayward Fault, where creeping and locked sections sit side by side much as they do offshore Kamchatka.

Seismology currently offers two forecasts on very different clocks, and this study is squarely the slower one. In the days after the mainshock, the USGS calculated a 24 percent chance of another magnitude 7 or larger aftershock within a week, a forecast that updates daily and is largely spent within weeks. The locking map works on the opposite timescale: a location that may not matter for decades, revised only as new geodetic surveys come in.

For a fault line, and for the coasts nearby, knowing where the strain has piled up is useful. It is not the same as knowing when the bill comes due. Funning puts it more bluntly: “Your peace of mind shouldn’t come from believing we can forecast the exact earthquake. Especially where we live in Southern California, it’s not a matter of if, but when. There is no substitute for preparation.”

FAQ