Winds at the Coast Were Hiding the True Wobble of Global Sea Level

Global mean sea level has become one of the most closely watched vital signs of a warming planet, a single number that rolls up ocean heat uptake and the melting of glaciers and ice sheets into one planetary pulse. Yet for all the sophistication of modern satellite altimetry, the measurements that tell us how sea level wobbles from year to year and decade to decade have been surprisingly noisy. A new study published in the journal Ocean Science by Andrew G. P. Shaw, Svetlana Jevrejeva and Francisco M. Calafat reveals a hidden culprit: winds blowing along the world’s coastlines, which create real ocean signals that tide gauges see but satellites miss. By correcting for these wind-driven differences, the team cut the error in reconstructed sea level variability by 23 percent, a substantial leap forward for a measurement that underpins the global sea level budget.

The core of the problem lies in how scientists reconstruct sea level before the satellite era. Direct global measurements from satellite altimetry only began in late 1992, so everything we know about earlier decades comes from tide gauges, instruments that are fixed to the coast and measure the sea right at the shoreline. The most widely used reconstruction technique, known as reduced space optimal interpolation, infers empirical orthogonal functions, or EOFs, from satellite data and then fits a subset of these spatial patterns to tide gauge records to estimate how each pattern waxed and waned through time. Because long-term trends are otherwise poorly captured, a spatially uniform pattern, often called EOF0, is added to the mix, following an approach first proposed by John Church and colleagues in 2004.

That addition comes at a steep price. Previous work has shown that while EOF0 dramatically improves the estimate of the underlying long-term trend, it cripples the reconstruction’s ability to capture interannual to decadal variability. The scale of the failure is striking. When the researchers compared the detrended, de-seasoned global mean sea level from the gridded CMEMS altimetry product with the widely used Church and White 2011 reconstruction over the period 1994 to 2013, the two series showed almost no resemblance, with a correlation of minus 0.19. Worse, the reconstruction overestimated the variability by more than a factor of two, with a standard deviation of 0.45 centimeters against just 0.21 centimeters for the observed altimetry average.

Getting this variability right matters far more than academic tidiness. The year-to-year ups and downs of global sea level carry information about the global hydrological cycle, revealing how much water temporarily shifts between the ocean and the land. They also influence estimates of the long-term trend and its acceleration, the numbers that matter most for climate policy. Earlier studies found that both the warming-driven expansion of seawater and the addition of meltwater from land ice contribute roughly equally to this variability, and both are strongly tied to the El Niño-Southern Oscillation. Sea level variability is also increasingly significant in its own right, contributing to coastal flooding and erosion as the ocean warms.

The key clue came from earlier numerical experiments by Calafat and colleagues, who showed that the low skill of EOF reconstructions is largely due to differences between tide gauge records and the altimetry data at the nearest offshore points. Satellite altimetry degrades within roughly 10 to 20 kilometers of the coast, and coastal waters host real physical signals that a nadir-looking radar struggles to resolve. When the researchers replaced tide gauges with these altimetry points, effectively using them as virtual tide gauges, the reconstruction captured global variability almost perfectly, achieving a correlation of 0.98 with the true altimetry average. The differences between the two instruments, which the team calls Signal Differences, were clearly the dominant source of error, not the sparseness of the tide gauge network.

Crucially, the new study shows that these Signal Differences are not random noise. Their standard deviations range from nearly zero to 4.5 centimeters, with clusters of high values along continental coastlines, and they display strong regional coherence. Sites along the west coast of North America, the Bay of Biscay and English Channel, northeast Australia and the Río de la Plata basin each correlate strongly with their regional averages, with mean correlations between 0.62 and 0.74. A control test in which the differences were replaced with random values and the analysis repeated 100,000 times yielded a mean correlation of only 0.37, far below what the real data show. The differences are geophysical, not instrumental.

When the researchers probed what these differences actually represent, a telling pattern emerged. At 146 of 229 analyzed locations, the Signal Differences correlated significantly with the tide gauge records themselves, but at only 45 locations did they correlate with altimetry. This means the discrepancies are mostly caused by real oceanographic signals that the tide gauges detect but the altimeter does not, rather than by errors in either instrument. The likely physical mechanism is wind-driven coastal trapped waves, which propagate sea level signals over long distances along continental shelves around the world, from Europe and Australia to the United States. These waves have cross-shelf scales comparable to the shelf width, so their effects are confined to the coastal zone and may not be fully captured by altimetry.

Because the differences have a geophysical origin, they can be modeled. The team correlated the Signal Differences with the zonal and meridional components of 10-meter winds from the ERA5 reanalysis and with the Southern Oscillation Index, a normalized pressure difference between Tahiti and Darwin that tracks the El Niño-Southern Oscillation. Strong, coherent correlation patterns emerged along the Pacific coast of the United States, Australia, the United Kingdom and Japan, with absolute correlations of 0.5 or more at many sites. Building on this, the researchers constructed three multiple linear regression models at each tide gauge, progressively adding wind components, the Southern Oscillation Index and the tide gauge record itself as predictors. The full model achieved a globally averaged correlation of 0.67 with the Signal Differences and explained 49 percent of their variance, with wind alone accounting for 26 percent.

The payoff came when the best-performing model was used to correct the tide gauge data before running the reconstruction. The team derived the regression parameters over the satellite era from 1994 to 2020, then extrapolated the corrections back to 1941, extending the improvement deep into the pre-altimetry past. The corrected reconstruction’s standard deviation of variability dropped from 0.30 to 0.25 centimeters for the satellite period, matching the true altimetry value exactly, and its correlation with the observed record doubled from 0.20 to 0.40. The difference between reconstruction and truth shrank by 23 percent. Importantly, the long-term trends were statistically unchanged, at around 2.0 millimeters per year for the full corrected record, confirming that the correction refines the variability without distorting the trend.

The improvement was most visible during the strongest El Niño events. Uncertainties in the corrected reconstruction shrank during the very strong El Niños of 1997-1998 and 2015-2016, and the corrected curve tracked the true altimetry record more closely through both episodes, evidence that uncorrected reconstructions had systematically mishandled the ENSO signature. The authors caution that some Signal Differences may still reflect poor-quality altimetry or tide gauge data rather than genuine ocean physics, and they note that non-linear vertical land motion at some sites remains a stubborn complication; correcting for it using contemporary gravitational, rotational and deformation data did not further improve the result. Future refinements may incorporate river discharge effects, the angle of wind approach to the coastline, and targeted testing of which low-difference sites actually need correction. For now, the study offers something the sea level community has lacked: a physically grounded way to make the pre-satellite record of our planet’s rising oceans finally tell the whole story.

Subject of Research: Improving short-term variability in Global Mean Sea Level reconstructions by correcting wind-driven tide gauge and altimetry differences

Article Title: An improvement to short term variability in Global Mean Sea Level reconstruction

Article References: Shaw, A. G. P., Jevrejeva, S., & Calafat, F. M. (2026). An improvement to short term variability in Global Mean Sea Level reconstruction. Ocean Science, 22(5), 2673-2689. https://doi.org/10.5194/os-22-2673-2026

Image Credits: AI Generated

DOI: 10.5194/os-22-2673-2026

Keywords: global mean sea level, sea level reconstruction, tide gauges, satellite altimetry, empirical orthogonal functions, coastal trapped winds, ENSO, Southern Oscillation Index, coastal trapped waves, sea level budget, climate variability, Ocean Science