Deep beneath Japan's Kii Peninsula, a borehole strainmeter had been recording Earth's faint tidal breathing for more than a decade. The instrument, buried hundreds of meters in solid rock, was designed to measure minute deformations of the crust—shifts of a few parts per billion. Its calibration curves, which track the solid-earth tides against theoretical predictions, should have been stable. But 14 of 20 curves drifted. Slowly, persistently, they wandered away from the expected shape. It took a careful re-examination of calibration logs, a controlled laboratory experiment, and a surprising culprit—a silica gel plug—to understand why.

When Silica Gel Sabotages the Tide

Borehole strainmeters are among the most sensitive instruments in geophysics. They sit in cased boreholes, coupled to the rock by grout or mechanical arms, and detect changes in volume or linear strain as the Earth deforms under tidal forces, tectonic stress, and slow slip events. To achieve the required sensitivity—on the order of 10−9 strain—the instruments must be isolated from moisture, temperature fluctuations, and other environmental noise. Silica gel, a common desiccant, is often placed inside the instrument housing to keep the electronics dry.

At the observatory in central Japan, operated by the National Research Institute for Earth Science and Disaster Resilience (NIED), the strainmeter had been running since 2010. Its tidal calibration curves—plots of observed tidal amplitudes versus theoretical values for the major tidal constituents (M2, O1, S2, etc.)—showed a pattern. Most curves were stable for the first few months after installation, then began to drift. The drift was small, roughly 1–2% per year in amplitude, but cumulative. By 2023, some curves had shifted by more than 10% from their initial calibration.

Lead author Y. Tanaka, a geophysicist at the University of Tokyo, had been assuming the drift was due to thermal effects or slow electronic degradation. But when he re-examined the calibration logs in 2023, he noticed a correlation with seasonal humidity cycles. The drift accelerated during wet summers and slowed in dry winters. That pattern pointed not to electronics but to something inside the instrument that changed with moisture. The silica gel plug, which sat between the strainmeter's sensing element and the outside rock, was the prime suspect.

Tanaka's team removed one of the suspect plugs and tested it in a humidity-controlled chamber. Over 18 months, the silica gel absorbed water vapor, swelling by roughly 5% in volume. That expansion applied a small but persistent stress to the strainmeter's sensing element, effectively changing the mechanical coupling between the rock and the sensor. The effect was invisible in short-term lab tests, which typically last weeks. Only long-term field data revealed the slow degradation.

The Hidden Weakness in Borehole Strainmeter Design

The silica gel plug was a standard component in the strainmeter design, used for decades in similar instruments around the world. The gel is initially baked dry during assembly, but over months it slowly rehydrates by absorbing water vapor that diffuses through seals. The volume change is modest but enough to alter the strainmeter's calibration. The problem is not unique to this observatory; any borehole strainmeter that uses a hygroscopic material in the mechanical path could be affected.

Design engineers had not anticipated the issue because lab tests rarely run long enough. A typical qualification test might cycle the instrument through temperature and humidity extremes for a few weeks, then check calibration. The silica gel, having been freshly dried, shows no drift in that window. But in the field, the rehydration proceeds over months to years, a timescale that is expensive to reproduce in the lab.

The drift manifested as a change in the transfer function between ground strain and sensor output. For tidal calibration, the transfer function is assumed constant, so any drift appears as an error in the tidal amplitudes. The 14 affected curves showed systematic deviations that grew over time, while the remaining 6 curves—from instruments with a different batch of silica gel or better seals—remained stable. The pattern was consistent with a spatially variable moisture ingress across the observatory.

Tanaka's team also found that the drift was not linear. It followed a logarithmic curve, fast at first, then slowing as the silica gel approached equilibrium with the borehole's humidity. That shape made it easy to mistake for thermal drift or aging of electronic components. Only by correlating the drift rate with local precipitation data did the humidity link become clear.

A Decade of Data Nearly Tossed

The observatory had accumulated over a decade of continuous strain records. Those data were being used to study slow earthquakes, post-seismic deformation, and the deep structure of the Japanese subduction zone. If the calibration drift was not corrected, the data might have been discarded or, worse, used to draw erroneous conclusions about tectonic processes. Tanaka estimated that roughly 70% of the tidal amplitude data from the observatory were compromised to some degree.

The research team faced a choice: discard a decade of observations, or find a way to correct the drift post hoc. They opted for the latter, developing a correction algorithm based on the measured humidity and the known swelling behavior of the silica gel. The algorithm works by modeling the silica gel's water absorption as a diffusion process, using Fick's second law with a time-dependent boundary condition. The model parameters—diffusivity and equilibrium water content—were calibrated from the laboratory experiment on the removed plug. The correction is applied to each tidal constituent separately, adjusting the amplitude by a factor that depends on the cumulative humidity exposure since installation. But the correction introduced additional uncertainty, roughly 0.5% in tidal amplitude, which is comparable to the signal of interest for some deep-Earth studies. A hardware fix was clearly preferable.

The decision to replace the silica gel with a hydrophobic polymer seal came after testing several alternatives. The chosen material, a perfluoroelastomer, is chemically inert and absorbs less than 0.1% water by weight. It also has a lower thermal expansion coefficient, reducing another potential drift source. The new seals were installed in three borehole strainmeters in 2024, and after 18 months of monitoring, the drift had dropped to below 0.1% per year—within the noise level of the tidal calibration.

The fix was cheap, under US$ 100 per instrument, but required opening the borehole casing and re-coupling the strainmeter to the rock. That operation carries its own risks: disturbing the sensor, introducing new stress, or damaging the borehole. The team developed a careful protocol, including a post-installation stabilization period of several weeks, before declaring the fix successful. The protocol involves first measuring the borehole's ambient humidity and temperature profile, then slowly depressurizing the casing to avoid sudden stress changes. The strainmeter is decoupled from the rock using a specialized tool that grips the casing, allowing the sensor to be extracted without twisting. After replacing the seal, the instrument is reinserted and allowed to equilibrate for at least four weeks, during which tidal calibrations are performed weekly to monitor stabilization.

How the Fix Changed the Calibration Protocol

The new protocol, adopted by the Japanese Geodetic Survey in early 2025, mandates the use of hydrophobic seals in all new borehole strainmeter installations. Existing instruments are being retrofitted as opportunities arise, typically during routine maintenance every 5 years. The protocol also includes continuous monitoring of internal humidity, with a threshold of 10% relative humidity triggering an alarm and a recalibration check.

Calibration procedures themselves have been revised. Instead of relying solely on a pre-deployment lab calibration, the protocol now requires an in situ tidal calibration every 6 months for the first 2 years, then annually thereafter. The tidal calibration uses the M2 tidal constituent, which has a well-known theoretical amplitude and phase at any location. Any deviation greater than 0.5% from the theoretical value triggers a review of the instrument's mechanical coupling.

The change was not without controversy. Some geophysicists argued that the drift was small enough to be ignored, especially for studies of transient deformation that do not rely on absolute calibration. Others pointed out that the cost of retrofitting dozens of borehole stations worldwide could exceed US$ 500,000, not including the loss of data during the retrofit. Tanaka's team countered that the alternative—living with unquantified systematic error—is more costly in the long run.

The debate echoes a broader tension in geophysical instrumentation: the trade-off between sensitivity and stability. Sensors that are extremely sensitive to strain are also sensitive to any change in their environment. The silica gel plug was a compromise, offering moisture protection at the cost of long-term drift. The new polymer seal improves stability but may introduce other issues, such as reduced flexibility at low temperatures or higher cost. No fix is perfect, but the silica gel episode has made the community more aware of the hidden assumptions in instrument design.

What Cleaner Tides Reveal About the Deep Earth

With the drift corrected, the strainmeter data now show tidal components that were previously masked. In particular, the fortnightly tidal constituent Mf, which has a period of about 13.66 days, is now clearly resolved. This component is sensitive to the viscosity of the lower mantle, because the tidal bulge at that timescale penetrates deeper into the Earth than shorter-period tides. The corrected data suggest a lower mantle viscosity roughly 10–20% lower than previous estimates, which used data contaminated by drift. However, this result is preliminary and based on data from a single observatory. The uncertainty remains large, and the estimate could change as more stations are retrofitted and analyzed. The new viscosity estimate, if confirmed, would have implications for mantle convection and the dynamics of subducting slabs. A less viscous lower mantle means slabs sink faster and plumes rise more easily, affecting models of plate tectonics and hotspot volcanism. Tanaka's team is now analyzing data from two other retrofitted stations to confirm the signal.

Beyond viscosity, the corrected data also reveal transient deformation signals that precede slow earthquakes. Slow earthquakes are episodic slip events that last days to months, releasing energy equivalent to a magnitude 5–6 earthquake but without generating strong shaking. They are thought to play a role in loading the seismogenic zone and triggering regular earthquakes. The improved calibration allows researchers to detect smaller and earlier precursors, potentially improving short-term earthquake forecasting.

The strainmeter data have also been used to refine the local tidal model, which is a necessary input for correcting gravity and GPS measurements. A better tidal model reduces noise in those other geodetic techniques, creating a virtuous cycle of improvement. The silica gel fix, in effect, has ripple effects across multiple disciplines within solid Earth geophysics.

Lessons for Other Long-Term Geophysical Sensors

The silica gel problem is not an isolated case. At the Plate Boundary Observatory (PBO) in the United States, tiltmeters have experienced similar drifts attributed to swelling of epoxy seals used in their construction. A 2018 study by Hodgkinson et al. documented that 12 of 45 PBO borehole tiltmeters showed annual drifts of 1–3 microradians, correlated with seasonal rainfall. The culprit was identified as water absorption by the epoxy potting compound, which expanded and tilted the sensor. Similarly, at the Hawaiian Volcano Observatory (HVO), piezometers used to monitor magma chamber pressure have shown drift linked to degradation of O-rings in wet environments. A 2020 report by Poland et al. noted that three of eight deep piezometers at Kilauea had to be recalibrated after O-ring swelling caused pressure offsets of up to 0.5 MPa. These examples highlight that moisture-sensitive materials are a recurring issue in borehole instrumentation.

Routine recalibration every 5 years, as is common practice, may miss gradual drifts that accumulate over decades. The silica gel drift was detectable only because the observatory had multiple instruments and a long time series. A single instrument, calibrated only at installation, would have produced data that looked reasonable but were systematically wrong. The lesson is that calibration should be ongoing, with independent checks from tidal models or co-located sensors.

Cross-checking with independent tidal models is one such check. If a strainmeter's observed tidal amplitudes deviate from the model by more than a threshold, it signals a problem. But the model itself has uncertainties, and drift can masquerade as a local anomaly. The Japanese team found that the drift was consistent across tidal constituents, ruling out a model error. That consistency was key to identifying the hardware issue.

Continuous monitoring of internal humidity, as now recommended, adds a small cost but provides a direct diagnostic. If humidity rises above a threshold, the operator can schedule a recalibration or retrofit before the data become unusable. This is a shift from a 'fix it when it breaks' mentality to a predictive maintenance approach, common in engineering but still rare in geophysical observatories.

From a Plug to a Careful Improvement

The silica gel fix cost less than US$ 100 per instrument, yet it saved decades of data from potential discard. The methodological lesson is clear: trust the hardware, but audit it deeply. Small material choices—a desiccant, a sealant, a potting compound—can ripple through global geophysical models, affecting our understanding of mantle viscosity, slow earthquakes, and plate tectonics. The community is now retrofitting 30+ borehole stations worldwide, a process that will take years.

Not everyone is convinced that the fix is universally necessary. Some argue that the drift is so small that it only matters for the most demanding applications, such as tidal calibration itself. For studies of coseismic offsets or slow slip events, which involve strain changes of 10−7 or more, a 1% drift is negligible. The cost and risk of retrofitting may outweigh the benefit for many stations. Tanaka acknowledges this, but he points out that the drift is cumulative: over 20 years, a 1% annual drift becomes a 20% error, which is significant for any quantitative analysis.

The episode also highlights the value of long-term, stable funding for geophysical observatories. The Japanese network has been running for over a decade, sustained by government support. Without that continuity, the drift might never have been noticed. In an era of short-term grants and project-based science, the silica gel story is a reminder that some discoveries require patience, persistence, and the willingness to re-examine old data with fresh eyes.

Ultimately, the fix is a careful, incremental improvement in instrumentation. It allows researchers to ask questions they could not ask before: about the viscosity of the deep mantle, the timing of slow earthquakes, and the fidelity of tidal models. But the work is not done. The next step is to extend the retrofit to more stations and to develop a real-time drift detection algorithm that can alert operators before data quality degrades. That algorithm, currently being tested on archived data from the Japanese network, uses machine learning to identify subtle changes in the tidal transfer function. If it works, it could become a standard tool for all borehole strainmeter networks, turning a lesson learned from a small plug into a permanent improvement in how we monitor the Earth's restless interior.