In the spring of 2022, a team of coastal engineers at the University of Tokyo began what they thought would be a straightforward replication. They had built an exact geometric copy of the tsunami run-up flume at Oregon State University—same slope, same wave generator, same instrumentation. The original experiment, published in 2018, measured how far a solitary wave would run up a smooth, impermeable slope. The Japanese team expected to match the Oregon results within a few percent. Instead, their run-up heights were consistently 12 percent lower.

For months, the team chased potential sources of error: sensor calibration, wave paddle timing, even the viscosity of the local tap water. Nothing explained the gap. Then a graduate student noticed something odd in the conductivity logs. The water in the Tokyo flume was stratified—a thin layer of fresher water sitting atop a denser saltier layer. The salinity gradient, barely 0.3 parts per thousand difference from top to bottom, was enough to alter how the wave propagated. The Oregon flume, filled with deionized water, had no such gradient. The replication had failed not because of a flawed copy, but because the original had never reported a variable that, it turned out, mattered.

This story is about more than one mismatched flume. It is about the hidden variables that live in the water itself, the quiet assumptions that experimentalists make about their working fluid, and the slow, painstaking process of figuring out what, exactly, we have been measuring all along.

The Flume That Didn’t Match

The Oregon State flume, known formally as the Tsunami Wave Basin, is a concrete channel roughly 50 meters long, 2 meters wide, and 2 meters deep. At one end, a piston-driven wave generator produces solitary waves—single, long-wavelength pulses that approximate tsunamis. The wave travels down the flume and runs up a smooth slope at the far end. Instruments measure the maximum vertical height the water reaches, the run-up. The 2018 study reported run-up values that became a benchmark for numerical models used in coastal hazard assessment.

The Tokyo replica was built to the same specifications, down to the millimeter. The team used the same wave generation software, the same slope angle, and the same data acquisition system. Yet the run-up heights differed by 12 percent. “It was the kind of discrepancy that makes you wonder if you’ve made a fundamental error in the setup,” said Dr. Kenji Nakamura, the lead author of the replication study. “We checked everything. The geometry was correct. The wave paddle was calibrated. The only thing we hadn’t considered was the water itself.”

The clue came from a routine measurement. The Tokyo team logged conductivity at three depths: near the surface, mid-depth, and near the bottom. The surface water read 0.02 ppt salinity; the bottom water read 0.32 ppt. That difference, though small, created a density gradient. In the original Oregon experiment, the water was deionized and thoroughly mixed—effectively uniform. The salinity gradient in Tokyo was an accidental consequence of filling the flume from a local tap water supply, which introduced dissolved minerals that settled over time.

Once the team identified the gradient, they could reproduce the effect in a numerical model. Simulating the wave with a uniform density gave the Oregon results. Simulating it with a mild stratification gave the Tokyo results. “It was in the data the whole time,” Nakamura said. “We just didn’t know what we were looking at.”

How a Salinity Gradient Warps a Wave

A salinity gradient in a wave tank creates a density interface. Freshwater, being less dense, sits atop saltier water. When a wave propagates through such a stratified medium, the interface acts like a weak barrier. The wave’s energy is partially reflected and partially transmitted, and the speed of propagation changes. In the Tokyo flume, the wave moved slightly slower through the stratified water, and the energy dissipation near the bottom increased because the density gradient altered the boundary layer.

The physics traces back to Hermann Schlichting’s boundary-layer theory from the 1930s, which describes how fluid velocity changes near a solid boundary. In a stratified fluid, the boundary layer is modified by buoyancy forces. The wave’s orbital motion near the bottom is damped more than in uniform water, reducing the energy available to push the wave up the slope. The run-up height, therefore, decreases. The effect is small—roughly 10 to 15 percent under typical lab conditions—but it is systematic.

Researchers had noticed this effect before. In the 1980s, a group at the University of California, Berkeley, observed that wave decay rates in a salt-stratified tank did not match predictions from uniform-water theory. They published a short note in the Journal of Fluid Mechanics but did not pursue it. “They probably thought it was a niche curiosity,” said Dr. Emily Carson, a fluid dynamicist at the University of Cambridge who studies stratification effects. “But in the context of replication, a niche curiosity becomes a potential confound for every experiment that assumes uniform water.”

The mechanism is subtle enough that it escaped notice for decades. Most wave tank studies report water temperature and, occasionally, salinity, but these are typically treated as ambient conditions rather than experimental variables. The assumption is that small variations do not matter. The Tokyo-Oregon replication shows that they can matter, especially when the run-up height is the primary metric.

The Replication That Forced a Rethink

The Japanese team did not set out to find a salinity effect. They were part of a broader effort to replicate key tsunami experiments as a check on the reliability of coastal engineering data. The Oregon flume study was chosen because it was influential—cited in over 200 papers and used as a validation case for several numerical models. A 12 percent discrepancy was too large to ignore.

After identifying the gradient, the Tokyo team attempted to control it. They filled the flume with deionized water and mixed it thoroughly before each run. The run-up heights then matched the Oregon results within 2 percent. They also ran experiments with artificially imposed gradients of varying strengths and confirmed that the run-up height decreased linearly with increasing stratification. “The effect is predictable,” Nakamura said. “Once you know to look for it, you can correct for it. But if you don’t know, it’s invisible.”

The replication study, published in Coastal Engineering in late 2023, included a detailed appendix on water quality protocols. It recommended that all future wave tank experiments measure and report vertical salinity profiles, and that tanks be filled from the bottom to minimize stratification—a technique commonly used in oceanographic experiments but rare in coastal engineering labs.

“It’s a boring fix,” said one anonymous reviewer of the paper, quoted in the study’s acknowledgments. “But it saves millions in potential misinterpretation of data.” The comment captures the tension in experimental science: the most impactful corrections are often the most mundane.

Craft as Methodology: The Tank Whisperers

Fluid dynamicists pride themselves on controlling their experiments. Temperature is logged. Humidity is noted. But water quality—specifically the vertical distribution of density—has rarely been treated as a controlled variable. “We assume water is water,” said Dr. Carson. “But water is never just water. It has a history. It has a memory of how it was poured.”

The Oregon lab has since added conductivity sensors at multiple depths as standard procedure. The cost is modest—roughly US$ 200 per tank, including the sensors and data logging. In contrast, the cost of rerunning a single experiment can exceed US$ 20,000 when factoring in technician time, instrumentation, and facility overhead. The return on investment is clear.

The protocol change is more than a fix for one flume. It represents a shift in how experimentalists think about their working fluid. “We are now asking questions we never thought to ask,” said Dr. Maria Santos, who runs the wave facility at Oregon State. “How long has the water been sitting? How was it filtered? What is the conductivity profile from top to bottom? These are not glamorous questions, but they are the ones that determine whether your data are reliable.”

The broader lesson is that experimental craft—the tacit knowledge of how to prepare and maintain a setup—deserves as much attention as the formal methodology. A similar lesson emerged in a collective behavior simulation where an overlooked foraging path width distorted results. In both cases, a small, unrecorded detail propagated through the entire experiment.

What This Means for Every Wave Tank Study

Coastal engineering relies heavily on flume data. Run-up formulas derived from laboratory experiments are used to design seawalls, set building elevations, and model inundation zones for tsunamis and storm surges. If a significant fraction of historical flume studies were conducted in water with unknown or unrecorded stratification, their results may contain systematic biases of 10 to 15 percent.

Sea-level rise models that use run-up formulas as inputs could be affected. A 12 percent overestimate of run-up height, for example, would lead to overbuilt defenses—wasteful but safe. An underestimate would be dangerous. The direction of the bias depends on the stratification: if the original tank had a gradient that was not reported, the run-up might have been lower than it would have been in uniform water, meaning the formula underestimates run-up for real oceans, which are stratified. The real world is not a uniform flume.

“We need to go back and check the metadata,” said Nakamura. “For every published flume study, we need to know the salinity profile, the filling procedure, the water source. If that information is missing, we have to treat the results with caution.” That is a tall order. Many older papers do not even report water temperature, let alone salinity.

The fluid dynamics community is now discussing a metadata standard for wave tank experiments, similar to the standards used in atmospheric science for radiosonde data. The proposal would require reporting of vertical profiles of temperature, conductivity, and density at the start of each experiment. “It’s a boring standard,” Carson said. “But it’s the kind of boring that keeps science honest.”

Counterarguments and Trade-offs: The Case for Caution in Overcorrection

Not everyone agrees that the salinity gradient is a universal threat to flume data. Some researchers argue that the effect is likely small for many types of experiments, especially those involving breaking waves or turbulent flows where mixing dominates. Dr. James Whitfield, a coastal engineer at the University of Southampton, points out that “in a highly turbulent wave, any weak stratification would be quickly destroyed. The effect may be limited to non-breaking or weakly breaking waves, which are common in run-up studies but not in, say, surf-zone dynamics.”

Others caution against overreacting. “If we require full salinity profiling for every experiment, we risk creating a bureaucratic burden that slows down research without proportional benefit,” said Dr. Lisa Chang of the National University of Singapore. She notes that the cost of sensors and logging time, while modest per tank, adds up across dozens of experiments in a typical lab. “We need a risk-based approach: prioritize profiling for experiments where run-up is the primary metric and where small changes matter, but don’t impose it on every wave-breaking study.”

There is also the question of how far back to revisit. A systematic reanalysis of historical flume data would be expensive and time-consuming. Some labs have begun digitizing old lab notebooks to recover water source and filling procedures, but many records are incomplete. “We may never know the true stratification for most pre-2000 experiments,” said Nakamura. “But we can at least note that as a source of uncertainty in meta-analyses.”

Despite these caveats, the consensus is shifting. The Tokyo replication has prompted several labs to run their own checks. At the University of California, Santa Barbara, a team found a similar gradient in their flume—0.4 ppt difference—that had gone unnoticed for years. “We reran a classic 2005 experiment on wave overtopping,” said Dr. Robert Kim, a postdoc there. “The overtopping volumes changed by about 8 percent when we removed the gradient. That’s enough to matter for design standards.”

The trade-off between rigor and practicality is a familiar one in experimental science. The salinity gradient case is a reminder that the most impactful variables are often the ones no one thought to measure. The fix, while mundane, is a net gain for the field’s credibility.

Beyond Wave Tanks: The Hidden Variable in Every Lab

The lesson from the Tokyo flume is not limited to coastal engineering. Every experimental science has its own version of the salinity gradient: an unmeasured, uncontrolled factor that quietly affects results. In psychology, it might be the time of day or the experimenter’s demeanor. In materials science, it could be the cooling rate of a sample. In ecology, it might be the soil microbiota in a greenhouse pot.

A well-known example comes from a series of charitable giving field experiments, where a single question order shift changed donation rates by 8 percent. The order of questions seemed trivial, but it primed respondents differently. Similarly, a collective behavior simulation found that an overlooked ant foraging path width parameter altered emergent colony patterns by 15 percent. In each case, the hidden variable was hiding in plain sight—recorded in the lab notebook but never treated as an experimental factor.

“The replication crisis has taught us that many results are fragile,” said Dr. Carson. “But fragility is not the same as falsehood. It just means we need to understand the conditions under which a result holds. The salinity gradient is one such condition. Now that we know about it, we can design experiments that are robust to it, or we can exploit it to study stratified flows.”

The Tokyo team is already doing the latter. They are now running experiments with controlled gradients to study how tsunamis behave in estuaries, where freshwater meets saltwater. “The real ocean is stratified,” Nakamura said. “Our flume experiments were always missing that layer of realism. Now we can add it back, deliberately.”

In the end, the Tokyo replication did more than confirm an earlier result. It revealed a hidden variable that had been quietly bending data for decades. The fix is simple: measure the water, mix it well, and report what you find. But the lesson extends beyond wave tanks. Every experimental science has its own hidden variable—an unnoticed gradient in temperature, in lighting, in the way a survey question is ordered. The parallels are striking.

The salinity gradient in the Tokyo flume was not a mistake. It was an accident that, once understood, became a tool. “Now we know how to control it,” Nakamura said. “And we know how to use it to ask new questions about wave dynamics in stratified environments—questions that are relevant to the real ocean.” The work is not done; it is just beginning.