When Cell-Tower Infrastructure Watches the Crust: SoftBank GNSS Data Sharpen Japan’s Strain Map

Japan looks still only because human eyes are poor instruments for watching continents move. The islands are being compressed, stretched and sheared continuously as the Pacific and Philippine Sea plates interact with the crust around them. Much of that motion is measured in millimeters or centimeters per year. Over time, however, those tiny displacements help scientists understand where strain is accumulating within the Japanese archipelago.

A study published September 22 in Earth, Planets and Space gives that picture substantially more detail. Miku Ohtate and Yusaku Ohta of Tohoku University, together with Mako Ohzono and Hiroaki Takahashi of Hokkaido University, integrated Japan’s roughly 1,300-station GEONET with more than 3,300 proprietary GNSS reference stations operated by SoftBank Corp. Using observations from November 2019 through December 2023, they retained 3,606 stations after quality control and estimated a nationwide crustal strain-rate field at finer spatial scales than GEONET alone could provide.^[1][2]

What does adding thousands of points actually change?

A GNSS station does more than report latitude and longitude. If the same monument is observed continuously for years, researchers can track whether the ground has moved north, east, up or down by millimeters. Compare the velocity of neighboring stations and the differences reveal where the crust is shortening, extending or shearing. That derivative of motion is the strain rate.

Japan already possesses one of the world’s most important continuous geodetic networks. The Geospatial Information Authority of Japan operates GEONET, built around roughly 1,300 electronic reference stations and a central facility in Tsukuba. Operating since 1996, it supports surveying, positioning services and continuous monitoring of crustal deformation.^[3]

Yet even 1,300 stations leave gaps. If deformation changes over a distance shorter than the spacing between sites, the signal can be smoothed away. The new study fills many of those gaps with stations that were never originally built for earthquake science.

Cellular infrastructure became a geophysical observatory

SoftBank’s reference stations began with a commercial positioning problem, not a seismological one. In 2019 the company announced a nationwide network of more than 3,300 fixed GNSS reference stations, using locations associated with its cellular base-station infrastructure to support centimeter-class RTK positioning. The service became known as ichimill, intended for applications such as agriculture, construction, surveying and autonomous mobility.^[4]

But a fixed antenna continuously receiving satellite signals is also, in principle, a sensitive record of ground motion. Tohoku University began receiving data from the SoftBank network in 2021 and tested whether it was accurate enough for crustal-deformation research. A 2022 study showed that the private network could capture deformation at roughly comparable accuracy to GEONET. Later that year, Tohoku University, SoftBank and ALES established a consortium involving 18 departments across 12 Japanese research institutions to explore Earth and space science applications of the network.^[5][6]

That history makes the 2026 paper interesting beyond seismology. A nationwide commercial infrastructure, built to make machines and vehicles know where they are, is now increasing the density of Japan’s scientific observations of the crust.

From a roughly 39-kilometer locality scale to about 21 kilometers

The improvement can be quantified. The researchers fitted local linear horizontal-velocity models with spatial regularization and optimized how strongly nearby observations should influence each estimate. Their distance-decay parameter, D, averaged about 39 kilometers when GEONET was used alone. With GEONET and SoftBank data integrated, that mean locality scale fell to about 21 kilometers. SoftBank-only analysis yielded about 25 kilometers.^[2]

A smaller value means the analysis can rely on observations closer to each calculation point and therefore resolve shorter-wavelength deformation. It would be too simplistic to call this a literal doubling of map resolution, but the practical direction is clear: the denser network makes more localized patterns recoverable.

The team also separated broad regional deformation from finer-scale features by applying a two-dimensional low-pass filter with a 50-kilometer radius, then examining what remained after that long-wavelength component was removed. This helped expose local strain patterns that can otherwise be obscured by the large-scale deformation driven by plate motion.

Niigata–Kobe, northern Hokkaido, San-in and the Ou Backbone Range

The higher-density field revealed clustered patches of maximum shear strain within the Niigata–Kobe Tectonic Zone, a well-known belt of concentrated deformation across central Japan. Short-wavelength shear-strain patterns also showed spatial correspondence with seismicity in northern Hokkaido and the San-in region. Along the Ou Backbone Range in northeastern Honshu, characteristic strain patterns corresponded with the distribution of Quaternary and active volcanoes.^[1][2]

That does not mean the brightest strain patch identifies the location of the next earthquake. Earthquake occurrence depends on fault geometry, friction, fluids, stress history, rupture interaction and many other factors. Surface strain is one observable part of a much larger physical system.

Noto showed the value of density before the nationwide map did

The scientific value of SoftBank’s network had already emerged in the Noto Peninsula. Earthquake activity there intensified from around the end of 2020, and a magnitude 6.5 earthquake struck in May 2023. GEONET recorded anomalous deformation, but the number of national stations close to the swarm was limited.

Researchers added SoftBank stations and were able to map uplift and deformation in greater spatial detail, supporting work that linked the prolonged swarm to fluid migration and slow slip.^[8]

After the magnitude 7.6 Noto Peninsula earthquake on January 1, 2024, the consortium again used SoftBank reference-station data alongside established observations to estimate horizontal and vertical coseismic displacement and fault slip. Results were supplied to bodies including Japan’s Earthquake Research Committee.^[7]

The 2026 study extends that regional proof of usefulness into a nationwide, multi-year strain-rate field.

GEONET’s three decades remain the foundation

The private network does not replace GEONET. Japan’s national system provides something a commercial deployment cannot instantly reproduce: decades of continuity, stable geodetic control, institutional quality management and publicly accessible data. Since operations began in 1996, GEONET has transformed the study of plate coupling, earthquake deformation, volcanic inflation and postseismic motion.^[3]

The 2011 Great East Japan Earthquake showed how consequential that measurement capability can be. The crust moved so extensively across eastern Japan that the national coordinate framework itself had to be revised. In October 2011, GSI introduced the Japanese Geodetic Datum 2011, JGD2011, because the old coordinates no longer adequately represented the ground after the earthquake.^[9]

The public-private strategy therefore works by layering density onto continuity. GEONET supplies the long-lived geodetic backbone; the private network fills spatial gaps.

More than 3,300 private stations does not mean every station was accepted

Density alone is not enough. Commercial reference stations are not operated under exactly the same constraints as dedicated scientific monuments. Equipment changes, local surroundings, data gaps and noise all matter. The researchers performed quality control and retained 3,606 stations in the final nationwide strain analysis rather than simply using every available site.

They also tested robustness by randomly removing 10% of stations and repeating the analysis 100 times. The principal strain patterns remained recoverable, while the exercise helped quantify uncertainty associated with station distribution. That is important because a finely textured map is only scientifically useful if its main structures do not disappear when a subset of observations changes.^[2]

The map also contains some post-earthquake motion

There is another limitation. The authors describe their target as an interseismic strain-rate field, but the November 2019–December 2023 interval includes postseismic deformation in some parts of Japan. Large earthquakes can continue moving the crust for months or years after the main rupture. Those signals can overlap with the slower background deformation researchers are trying to characterize.^[2]

That is another reason not to read the map as a simple hazard ranking. A high value can reflect tectonic loading, earthquake-cycle effects, volcanic processes or combinations of them. Interpretation requires the earthquake history and geology of each region.

Reusing communications infrastructure as science infrastructure

One of the most consequential ideas in the project is that Japan did not need to build several thousand new earthquake observatories from scratch. SoftBank had already deployed the stations to support centimeter-class commercial positioning. Scientific reuse raises the effective spatial density of the geodetic network at far lower marginal infrastructure cost.

The consortium is looking beyond earthquakes. Its stated research areas include volcanoes, atmospheric water vapor, heavy rainfall, ionospheric disturbances and correction of synthetic-aperture radar data. GNSS signals carry information not only about antenna position but about the atmosphere and ionosphere they pass through.^[5]

There are practical questions in relying on commercial infrastructure for long-term science: data continuity, metadata for hardware changes, quality assurance, access conditions and the business life of the network. Those questions do not negate the value. They define what a durable public-private observing system would need to manage.

The goal is better earthquake physics, not a prediction headline

Earthquake science often advances through better constraints rather than dramatic forecasts. How rapidly is each part of the crust deforming? Where do strain concentrations align with active faults or seismicity? How do volcanic regions alter the deformation field? Can models of inland earthquake occurrence reproduce what is actually measured at the surface?

The study’s gridded strain-rate products are released as digital supplementary data, allowing other researchers to compare them with earthquake catalogs, active-fault maps, seismic-velocity structure, volcanoes, InSAR and other datasets. The lasting value may therefore be less the publication of one high-resolution image than the creation of a reusable quantitative layer for future studies.^[2]

GEONET began turning the slow movement of Japan into measurable data three decades ago. SoftBank built a much denser network in 2019 for an entirely different commercial purpose. In 2026, putting the two together is allowing scientists to see finer structure in the same moving archipelago.