Across the United States ' coasts, the ground is moving. In some places, it is slowly rising through uplift; in many others, it is quietly sinking through subsidence. These processes are often imperceptible from one year to the next, but over decades their effects can substantially alter the performance of coastal infrastructure. Hurricane Katrina provides a striking example: post-event assessments showed that long-term subsidence and settlement had lowered portions of New Orleans’ levees before the storm arrived, effectively reducing the level of protection those structures could provide.
The implications of subsidence or uplift, of course, extend well beyond levees to roads, pipelines, buildings, drainage systems, and other critical infrastructure. We use levees as an example throughout this post partly because their performance is particularly sensitive to changes in elevation and partly because, as geotechnical engineers, we have a difficult time resisting an opportunity to talk about levees.
But that example also points to a much broader question: how widespread is this problem along the U.S. coast, and who or what is most exposed to it?
That question became the starting point for our study. What began as an exploration of a large subsidence and uplift dataset grew, through collaboration with researchers at Virginia Tech, into a national-scale investigation of both the movement of the land itself and what lies above it. We ask two connected questions. The first is about hazard: how are subsidence or uplift rates distributed statistically, how are they organized in space, and where are the most extreme rates concentrated? The second is about exposure: which communities, roads, railways, power systems, pipelines, and levees sit on top of that moving ground, and are some communities disproportionately exposed?
National Geodetic Survey benchmark H1235 RESET in Merced County, California. Public domain image, U.S. Geological Survey (USGS)
Measuring the moving ground
The measurements come from radar satellites. Satellites such as ALOS and Sentinel-1 pass over the same ground again and again. By comparing radar images from one pass to the next, researchers can track how the ground moves up or down, with uncertainties on the order of millimeters per year. The dataset we used combines ALOS imagery from 2007 to 2011 and Sentinel-1 imagery from 2015 to 2020 with ground-based GNSS (GPS-type) stations. The result is a dataset comprising more than 160 million subsidence estimates and over 30 million uplift estimates across 20 U.S. coastal states, each describing the rate at which the ground surface is rising or sinking at approximately 50-meter spatial resolution and extending up to 100 kilometers inland.
This dataset was built by collaborators at Virginia Tech. That national coverage let us analyze the coastal U.S. To our knowledge, that makes this the first national-scale assessment of coastal subsidence hazard and exposure in the U.S.
Hazard: One nation, several statistical shapes
Nationally, subsidence rates are best described by the Generalized Extreme Value (GEV) distribution. Most of the coast sinks slowly, but a small share sinks much faster. Those rare, extreme rates sit in what we call the "tail" of the distribution. GEV is heavy-tailed, meaning extreme rates are more common than many standard statistical models would assume. That matters because engineering decisions often depend on understanding extremes, not just averages.
Not all of the coast is sinking, though. About 16% of the measurements, roughly 30 million points, show the ground rising, mostly along the West Coast. Uplift is driven by different processes, so we fit it separately; it is best described by the Generalized Pareto distribution.
The national picture also hides important regional differences. State by state, the best-fitting distributions tend to group by coast: Pearson Type III on the West Coast, Generalized Logistic on the East Coast, and predominantly GEV along the Gulf Coast. This spatial variation suggests that a single national distribution may not adequately represent the tails of subsidence or uplift across all coastal regions, particularly where extreme values are most consequential.
Subsidence faster than 0.1 cm per year is widespread, affecting approximately 72% of measured points along the coast. But the picture narrows sharply as the threshold rises. About 7% of points sink faster than 0.5 cm per year and about 1% faster than 1 cm per year, with the fastest rates concentrated along the Gulf Coast.
Exposure: communities and infrastructure on moving ground
We grouped coastal census tracts into three hazard levels by average rate: low (sinking slower than 0.2 cm per year, or rising), moderate (0.2 to 0.5 cm per year), and high (0.5 cm per year or faster). In all, a little over 84 million people live in coastal tracts affected by some level of subsidence. Most are in low-subsidence zones, but around 14 million live in moderate-subsidence areas and about 3 million in high-subsidence tracts. In those high-subsidence tracts, roughly 38% of residents live in communities designated as disadvantaged by the federal Climate and Economic Justice Screening Tool (CEJST), compared with about a third in low-subsidence zones.
To measure overrepresentation, we compare a group's share of the population at a given hazard level with its share of the region as a whole. The baseline is all coastal tracts for national figures and a state's coastal tracts for state figures. An overrepresentation of 100% means double the baseline share. Nationally, Non-Hispanic Black residents are overrepresented in high-subsidence tracts by about 26%, and Hispanic residents by about 12%. State-level disparities are starker. In Louisiana's high-subsidence areas, Native American residents are overrepresented by more than 460%, a share more than five and a half times their baseline. In Delaware, Hispanic residents are overrepresented by 218%. In Virginia, low-income residents are overrepresented by roughly 100%, and people over 25 years old without a high school diploma are overrepresented by 140%.
Build Back Higher
Subsidence is a cumulative hazard. Even after infrastructure is repaired or rebuilt to its original elevation, the ground continues to sink. A levee restored to its design crest today can lose measurable freeboard over the following years, before the next storm ever arrives. Recovery, in a subsiding region, is a moving target, and the protection level a community believes it has slowly drifts away from the level it actually has.
Build Back Higher concept emphasizes that in subsiding coastal regions, design standards should explicitly account for expected future ground motion. Reconstruction elevations, flood-protection design, and long-term infrastructure planning should treat ground elevation as a time-dependent design variable rather than a fixed boundary condition.
What this study can't tell us
A few caveats are worth stating. The satellite record covers 2007 to 2011 and 2015 to 2020, so the rates we analyze describe historical trends. They may not capture how subsidence changes as groundwater management, land use, or climate extremes shift. Measurement uncertainty is on the order of millimeters per year, but it varies with land cover, radar signal quality, and the availability of nearby ground stations. Averaging rates to census tracts was necessary to link hazard to population data, but it smooths over very local deformation that a parcel-level study would catch. Our low, moderate, and high categories are rate thresholds, not engineering failure limits. And the exposure analysis treats population and infrastructure as fixed, when both will change.
What comes next
Because the distributions are heavy-tailed and vary by coast, extreme value methods are a natural next step for sharpening the tail estimates that design codes and flood-risk models depend on.
Several questions remain open. Do the coast-specific statistical shapes we found hold along other coastlines around the world? How will accelerated sea-level rise and continued groundwater stress reshape subsidence trajectories over the coming decades? How should infrastructure design codes evolve to treat ground elevation as a time-dependent variable rather than a fixed one? And what does equitable adaptation look like in regions where the ground beneath disadvantaged communities is measurably sinking faster than the ground beneath others?