The Ocean's Hidden Variables: How Wave Height Changes Everything We Thought We Knew About Ocean Reflectance

The Science

When sunlight strikes the ocean surface, it doesn't simply disappear. Instead, it bounces off—reflected by countless tiny facets on the water's skin, each one tilted at its own angle by the complex choreography of waves. These reflections tell us something profound about the sea itself. Measure enough of them, at enough angles, and you can reconstruct the statistical fingerprint of ocean roughness: how tilted are the waves, in what directions, and how does the wind shape them?

This is the logic behind a striking new study by Guérin, Capelle, and Hartmann, published in 2026, which turns 150 million observations from a satellite instrument into the most detailed map yet of how ocean waves really behave. Their paper, New features of the sea-surface slope distribution revealed by IASI observations, doesn't just confirm what scientists long suspected about wave slopes—it overturns some assumptions, refines others, and opens a window onto ocean physics that has remained stubbornly opaque.

The instrument at the heart of this work is the Infrared Atmospheric Sounder Interferometer (IASI), flying aboard Europe's Metop satellites. IASI was designed to measure atmospheric temperature and humidity with high precision. But Guérin and his colleagues realized it could do something else: by analyzing the solar radiation reflected off the ocean's surface at around 3.8 micrometers—a wavelength in the mid-infrared—they could extract the angles of the wave facets doing the reflecting. Each spectrum encodes information about which slopes were present, in what proportions, and in which directions.

This is not a new idea. The foundational work here was done by Cox and Munk in the 1950s, who used aerial photographs of sun glitter patterns to estimate wave-slope distributions. Their parameterizations—the mathematical descriptions of how likely each slope angle is—have been cited thousands of times and underpin countless studies in remote sensing, climate modeling, and oceanography. But Cox and Munk worked with a limited dataset and, crucially, assumed that wind speed alone determined the wave-slope probability distribution. What Guérin and colleagues show is that this assumption, while useful, is incomplete.

To carry out their analysis, the researchers sorted their 150 million IASI observations into bins organized by two variables simultaneously: wind speed, at 0.5 meters per second intervals, and significant wave height—the vertical distance between the troughs and crests of the waves—at 0.5 meter intervals. The significant wave height, denoted as Hs, is a standard measure of ocean roughness that captures the contribution of larger waves, not just the small ripples that sit atop them. By binning their data this way, the researchers could ask questions that no one had systematically answered before: Does the distribution of slopes change depending on how tall the waves are, even when the wind is blowing at the same speed?

Wind data came from the ERA5 reanalysis produced by the European Centre for Medium-Range Weather Forecasts, which provides hourly global estimates of wind speed and direction on a 0.25-degree grid. The researchers had previously verified that ERA5 winds agree with independent satellite measurements to within about 1 meter per second—a small enough error that it doesn't systematically bias their results. Wave height estimates also came from ERA5, and these were validated against over 16 million buoy measurements, showing an excellent correlation of 0.961 and a typical error of roughly 0.3 meters. For the range of conditions most relevant to their study—moderate winds and typical open-ocean wave heights—these are trustworthy numbers.

The methodology for extracting slope probabilities from the IASI spectra was developed in an earlier study and involves modeling how sunlight of different wavelengths is redirected by wave facets at various angles. By comparing the measured radiance at each pixel with theoretical predictions, the researchers can infer what the underlying distribution of slopes must have been. The resulting dataset is, in the researchers' own words, "the first of its kind"—optical-range measurements of wave-slope statistics that probe the full range of slopes without the filtering effects that plague radar-based methods.

The figure above illustrates the raw data for two contrasting sea states: light winds with small waves on the left (U = 2 meters per second, Hs = 1 meter) and stronger winds with larger waves on the right (U = 9 meters per second, Hs = 2 meters). Hot colors indicate regions where many observations fall—where certain slope angles are particularly common. The difference is immediately visible: the left panel shows a tighter, more concentrated distribution, while the right panel is broader and more diffuse. But the real story, as Guérin and colleagues reveal, is subtler than these gross differences. It lies in the asymmetries: which direction the peaks are shifted, how pronounced the directional preferences are, and how the tails of the distributions behave.

What They Found

The most consequential discovery in this paper concerns a variable that most previous studies ignored entirely: significant wave height. When the researchers sorted their IASI observations by both wind speed and wave height simultaneously, they found that wave height exerts a measurable influence on the slope distribution—independent of wind speed. At a given wind speed, smaller wave heights correspond to more directive wave slopes: the asymmetry between upwind-facing and crosswind-facing slopes becomes more pronounced, the directional preferences become sharper.

To quantify this, the researchers computed two asymmetry coefficients. The upwind-crosswind asymmetry (UCA) measures how much more likely a wave facet is to face into the wind than across it. The upwind-downwind asymmetry (UDA) measures the bias between the upwind and downwind faces of waves—the leeward and windward sides, in other words. Their key finding: at 7 meters per second wind speed, the maximum UCA drops from about 2.0 when the significant wave height is below 1 meter to roughly 1.3 when it exceeds 3 meters. The pattern holds across wind speeds. Young, short-fetched seas—those dominated by locally generated waves with relatively small heights—have more strongly directional slope statistics than mature, developed seas where large swells have built up.

This is a result with no obvious precedent in the literature. Most previous optical studies didn't have enough data to bin by wave height; most radar studies couldn't probe the full slope range without filtering effects. The researchers offer two physical mechanisms to explain it. The first involves wave-wave interactions: large, long-period waves can damp the short-scale ripples that contribute to slope statistics on their leeward faces, effectively smoothing those regions and sharpening the contrast with the rougher windward faces. The second involves statistical compounding: when the ocean surface is characterized by multiple wave systems superimposed on each other—a common situation in the open ocean—the resulting slope distribution is a convolution of their individual statistics, which tends to smooth out directional preferences. When the significant wave height is smaller, fewer wave systems are superimposed, and the directivity is preserved.

The chart above makes this relationship concrete. It shows how the upwind-downwind asymmetry coefficient varies with wave tilt angle—essentially, how steep the waves are—for a wind speed of 7 meters per second and several different values of significant wave height. The vertical distance between curves represents the effect of wave height: smaller Hs values (the uppermost curves) sit higher, indicating stronger asymmetry, while larger Hs values fall toward the bottom. The effect is most pronounced at moderate tilt angles, around 10 to 20 degrees, where the contrast between young and developed seas is largest.

A second major finding concerns the mean square slope (MSS)—the statistical measure of overall ocean roughness, capturing how much the surface tilts in all directions. The classical Cox-Munk model predicted that MSS increases linearly with wind speed, a relationship that has been validated many times. But Guérin and colleagues find something more nuanced: at wind speeds below about 6 meters per second, MSS also depends on wave height, decreasing by roughly 5% for every additional meter of significant wave height. At very low winds, this effect is large enough to be important. A sea with 1-meter waves at 3 meters per second wind should have notably smaller MSS than a sea with 3-meter waves at the same wind speed—contrary to what the Cox-Munk model would predict.

This result, which the researchers describe as "original," has implications for how we interpret satellite measurements of ocean roughness. If MSS varies with both wind and wave height independently, then a single MSS measurement cannot uniquely identify the wind speed without additional information. The relationship between wind and roughness is not a simple one-to-one mapping; it depends on the sea state, on the history of the wind, on the fetch and duration of the storm that generated the waves. This complicates retrieval algorithms that use ocean reflectance to estimate wind speed—the researchers cite studies showing that wave height does indeed influence those retrievals—and suggests that future algorithms should incorporate sea-state information.

This figure captures two key dependencies simultaneously. The top panel shows how the tilt angle at which directivity is maximized varies with wind speed and significant wave height. At low winds, the most directive slopes are relatively gentle—around 10 degrees—while at high winds they become steeper, approaching 30 degrees. Importantly, for any given wind speed, larger wave heights shift the most directive tilt toward larger angles by about 3 to 4 degrees. The bottom panel shows the corresponding UCA values at those optimal angles. The pattern is clear: maximum directivity peaks around wind speeds of 1 to 2 meters per second, declines to a minimum near 4 meters per second, and then rises again at higher winds. At all wind speeds, larger wave heights reduce the directivity.

The third major finding concerns the tails of the slope distribution—those rare, steep slopes that correspond to breaking waves and extreme tilts. The central region of the slope probability distribution is approximately Gaussian, meaning it follows the familiar bell curve, with moderate slopes being common and very gentle or very steep slopes being rare in a predictable way. But the tails deviate sharply from this Gaussian expectation. Guérin and colleagues find that the probability of steep slopes decays exponentially, not algebraically—a faster falloff than Gaussian would predict, but still one that produces meaningful probabilities for slopes that are physically extreme.

This exponential decay, the researchers note, appears "quite universal": the same functional form fits the tails across all wind speeds and wave heights they examined. This is a striking simplification. Despite the complexity of ocean wave dynamics—the interplay of wind forcing, gravity, surface tension, breaking, and foam—the statistical tail of the slope distribution obeys a single law. The physical origin of this universality remains to be fully clarified, but the researchers discuss several candidate mechanisms: the geometry of specular reflection, which naturally filters extreme slopes; the contribution of whitecaps and breaking waves, which add a separate component to the reflectance; and the compounding of multiple wave systems.

Another curious asymmetry emerges in the steep-slope regime. For gentle and moderate slopes, the probability distribution is negatively skewed along the wind direction: the peak of the distribution sits slightly downwind of center, reflecting the fact that wave faces tilted downwind are more common than those tilted upwind. But for the steepest slopes—those approaching the angle of maximum wave steepness before breaking—the asymmetry reverses. Larger probabilities are observed in the downwind direction, and similar probabilities are found in the upwind and crosswind directions. This is a regime that has been "practically unstudied" before, and the researchers are careful to acknowledge the difficulty of interpreting it: at these slopes, contributions from foam, spray, and breaking waves complicate the optical signal, and the assumptions of their retrieval method may not fully hold.

Finally, the researchers propose revised parametrizations for mean square slope versus wind speed that depart from the Cox-Munk relationships at moderate wind speeds between 5 and 8 meters per second. The Cox-Munk formulas, derived from aerial photographs in the 1950s, assumed a linear relationship that has been widely used ever since. But with 150 million observations and the ability to bin by wave height, Guérin and colleagues can resolve deviations from linearity that earlier studies missed. Their new formulas don't dramatically alter the predictions at high winds—where Cox-Munk remains a reasonable approximation—but at moderate winds they improve the match to observations. For scientists using MSS as an input to retrieval algorithms or climate models, these refinements could reduce systematic errors.

Why This Changes Things

The sea surface is where the ocean meets the atmosphere, and how it reflects light and heat is a problem that runs through nearly every branch of Earth system science. Satellite sensors that measure ocean color, sea surface temperature, or atmospheric composition all depend on accurate models of ocean reflectance—models that require the slope distribution as input. Climate models that simulate the energy budget of the planet need to know how much solar radiation the ocean reflects versus absorbs. Retrieval algorithms that convert satellite measurements into geophysical variables like wind speed assume a particular relationship between those variables and the ocean's optical properties. When the slope distribution deviates from what those models assume, errors propagate.

The Cox-Munk parameterizations have been the bedrock of this field for seventy years. They aren't wrong, exactly—they capture the first-order behavior of ocean roughness with wind speed. But they are incomplete in ways that matter. A 5% error in mean square slope might seem small, but when integrated over the global ocean and over years of satellite observations, it accumulates into a meaningful bias in climate records. If MSS depends on wave height as well as wind, then retrievals that use MSS as a proxy for wind will systematically overestimate winds in regions where seas are developed (high waves) and underestimate them where seas are young (low waves). The researchers' finding that this effect is strongest at low wind speeds is particularly important: these are exactly the conditions where satellite wind retrievals are already most uncertain, and where errors have the largest relative impact.

The discovery that steep slopes follow a universal exponential distribution is also significant, though its practical implications are less immediate. If the tail behavior is truly universal, it simplifies the mathematics of reflectance models and reduces the number of free parameters that need to be fitted. More fundamentally, it hints at some underlying order in ocean wave dynamics that is not yet fully understood. The exponential decay of probabilities at extreme slopes suggests that the processes governing wave breaking and the associated reflectance are dominated by a single physical mechanism, one that operates regardless of whether the sea is a glassy calm at 2 meters per second or a churn at 12 meters per second. Identifying that mechanism would be a genuine advance in physical oceanography.

The asymmetry findings—specifically, that young seas are more directive than developed seas—have practical ramifications for remote sensing of the ocean and atmosphere. Many retrieval algorithms implicitly assume that the slope distribution depends only on wind speed. If it also depends on sea state, on wave age, on the history of the wind field, then those algorithms will introduce errors that vary with location and time in ways that are hard to predict. The open ocean's vast subtropical gyres, where long swells from distant storms are common, may present a different reflectance regime than the marginal seas near continents, where locally generated wind waves dominate. Coastal zones, where fetch is limited and wave ages are typically young, may be systematically different still. These are variations that matter for the interpretation of satellite data and for the validation of climate models.

There is also a more fundamental point about scientific methodology. The Cox-Munk parameterizations were derived from a few hundred aerial photographs of sun glitter—a small sample, given the diversity of ocean conditions worldwide. They have been refined over the decades with new data, but always within the same framework: the assumption that wind speed is the only relevant control variable. Guérin and colleagues' work demonstrates that this assumption is testable and, in certain regimes, wrong. By leveraging 150 million satellite observations, they can resolve variations in the slope distribution that would have been invisible in earlier datasets. This is a case where the sheer scale of modern Earth observation makes new science possible—not just more precise versions of old science, but new qualitative insights that change the questions we think to ask.

For the broader public, these findings speak to something more general: the ocean is not a simple system. It does not respond to the wind like a被动 mechanical surface, with roughness scaling neatly and predictably with forcing. It has memory, history, its own internal dynamics. The waves that reach a point in the ocean have traveled thousands of kilometers, interacted with each other, been shaped by storms and calms long past. A snapshot of wind speed cannot capture that history, and that history matters for how the surface looks, how it reflects light, how it exchanges heat and momentum with the air above. The sea is more complicated than we knew—and understanding that complexity is progress.

What's Next

The most immediate implication of this work is practical: retrieval algorithms and climate models should be updated to incorporate the dependence of slope statistics on significant wave height. The researchers have provided parametrizations for how MSS varies with both wind and wave height, and for how the asymmetry coefficients depend on these variables. Implementing these refinements will require cooperation between the remote sensing, physical oceanography, and climate modeling communities, but the path is clear. The data exist; the formulas exist; what remains is adoption.

But there are deeper questions that this work opens up. The researchers propose two mechanisms to explain why younger seas are more directive, but they cannot decisively choose between them. The first hypothesis—that long waves damp short-scale ripples on their leeward faces—predicts a specific spatial pattern of slope statistics across individual waves, one that could in principle be tested with detailed in situ measurements or high-resolution airborne imagery. The second hypothesis—that statistical compounding from multiple wave systems smooths out directivity—makes different predictions about how the asymmetry should vary with the number of superimposed wave systems. Testing these predictions would require data on the full wave spectrum, not just Hs, which remains rare at global scales but is becoming more available with the maturation of satellite altimeters like SWOT.

The exponential tail behavior also demands explanation. The researchers suggest several candidate mechanisms but stop short of a definitive identification. This is frontier territory: the steep-slope regime involves processes like wave breaking, air entrainment, and spray formation that are inherently nonlinear and difficult to model from first principles. Numerical simulations of the ocean surface, at resolutions fine enough to resolve capillary ripples and gravity waves simultaneously, might eventually clarify which mechanism dominates. Laboratory experiments in wave tanks could also help, by isolating individual processes and varying them systematically. The prize would be a physical understanding that explains not just the existence of the exponential tail but its universality—the fact that the same decay constant appears regardless of wind speed or wave height.

There are also methodological questions worth pursuing. The IASI observations analyzed here come from a single overpass time (9:30 AM local solar time) and a limited range of viewing geometries. A full characterization of the slope distribution would require observations spanning the full diurnal cycle and the full range of sun angles, not just those available from a polar-orbiting morning satellite. Extending this analysis to afternoon overpasses and to instruments with wider swath widths would increase statistical power and expand the range of probed slopes. The researchers acknowledge that their retrieval method introduces some scatter in the estimated probabilities, particularly in the tails; improving the retrieval physics, or using complementary data sources to constrain the results, would sharpen the conclusions.

The finding that MSS decreases with wave height at low winds is surprising and potentially important, but its physical origin remains speculative. The researchers suggest two possibilities: short-scale damping by long waves, or the effect of compounding statistical processes. Both are plausible, but neither is proven. Understanding this effect is not merely academic: if it is real, it means that the standard practice of using MSS as a proxy for wind speed is biased in a direction that depends on sea state. For offshore wind energy applications, which increasingly rely on satellite wind retrievals calibrated against MSS, this could matter. For climate science, where biases in sea surface temperature and ocean heat content are already a concern, it adds one more complication.

Finally, this study points to a broader truth about how scientific knowledge advances. The Cox-Munk parameterizations have served the community well for seventy years. They were carefully derived, widely tested, and broadly correct. But they were derived with the data available at the time, and those data had limitations. As new instruments have come online—IASI, the MODIS sensors, the Sentinel satellites, the upcoming missions that will observe the ocean at higher resolution and with greater frequency—the opportunity arises not just to refine the old estimates but to question the old assumptions. That is what Guérin and colleagues have done. Whether their specific findings stand the test of time, or are modified by future work, the approach is the right one: use the new data, ask the new questions, and be willing to update the textbooks.