We measured 2,289 American landfills from orbit. 167 are sinking fast.

The fastest locally-sinking landfills in America

Ranked by window median — the typical pixel, not the worst one. “Deepest tenth” is the 10th-percentile pixel: the fast core of the bowl. TX (23), CA (20), NC (18) lead the state count. Coordinates are the EPA's own pins, included so anyone can check.

All 167 sites over the bar, and every one of the 2,323 we attempted: the full per-site CSV.

Fresh Kills: the control that failed, and what it proved

Before running the country we pre-registered two control sites that had to pass: Puente Hills in Los Angeles County — for decades the biggest landfill in America — and Fresh Kills on Staten Island. Puente Hills passed (median -6.7 mm/yr, deepest tenth -150). Fresh Kills failed: its window median came back -4.0, short of the −5 bar. The diagnosis matters more than the miss: the EPA pins Fresh Kills at an administrative point, so our window mixed the waste mounds with the wetlands beside them. Centre the same-size window on the mound itself and the median is -45.1 mm a year with 92% of pixels sinking — and NASA's own portal, fitted independently, says -42.7 at the same spot. Our measurement and NASA's agree to a couple of millimetres; the failure was the pin, not the radar. We report the bar as failed anyway, because that is what pre-registration means — and it tells you the list above is conservative: a gate-pinned site must sink hard across a diluted window to make it.

The rules we set before we looked

• Positive controls (P1): Puente Hills and Fresh Kills, bars set in advance. Result: one pass, one verbatim fail — diagnosed above, printed, not re-rolled.
• Negative control (N1): passed. 100 random on-swath points at least 5 km from any landfill: median of medians +0.4 mm/yr (bar: within ±1.5), 9% clearing −5 (bar: ≤10%). That 9% is itself a finding — random American ground contains real local motion — and it is the base rate printed beside the headline.
• Story bar (S1): passed as originally registered. The kill criterion was ≥15 landfills over the bar, excluding the two control sites; the sweep found 166.
• One amendment, made before the national data — and retired by its own control. After the Fresh Kills diagnosis we added a second way to qualify (a deep, spatially large tail without the median). The random-ground controls then showed that arm fires on 20% of random windows, and it flagged landfills at the same rate — base rate, not signal. So nothing on this page qualifies through it: the amended count (683) is reported here verbatim and used nowhere. Sites that qualified only under that arm are marked “second” in the CSV — leads for manual follow-up, not findings.

Method — so the result can be checked

1. Sites: the EPA Landfill Methane Outreach Program database (2024-09 edition) — 2,641 unique landfills, 2,323 with coordinates.
2. Radar: NASA/JPL's OPERA DISP-S1 V1 displacement archive — Sentinel-1 InSAR, 2016–2025, ~30 m pixels — read directly from the cloud archive, one window per site, scaled to the landfill's footprint (~1–2.4 km across).
3. Fit: a velocity plus an annual cycle fitted to every pixel's displacement history across the product's reference-date segments; site statistics (median, deepest tenth, share ≤ −5) are computed across pixels, never averaged first. Per-pixel noise is about ±2.5 mm/yr; a window median at −5 with 30+ pixels is far outside it.
4. Coverage, stated: stacks are missing for 154 of the archive's frames, so 34 sites could not be measured; Alaska and parts of the interior are thinner than the coasts. Every attempted site and its outcome is in the CSV.

Data: OPERA DISP-S1 (NASA/JPL, via the Alaska Satellite Facility); EPA LMOP. We did not produce the radar measurements; we produced the per-landfill join and fits. Analysis dated 2026-08-12.

Is your ground moving?

The same physics answers for a single US address, free.

Does a sinking landfill mean it is leaking or dangerous?

No. Settlement is what landfills do: buried waste compresses under its own weight and decomposes, and engineered caps are designed to move. A steady, broad settlement signal is normal physics, not a verdict on the site’s engineering — though rapid or very uneven settlement is something cap and gas-system engineers monitor, because it can stress liners, gas wells and drainage. This page measures motion; it does not inspect anything.

Why is a famous huge landfill missing from the list?

Three honest reasons. First, coverage: the radar archive is missing stacks for 154 frames, and 34 sites had no usable data at all. Second, the EPA database pins many landfills at the entrance gate or office, so our measurement window can sit partly off the waste itself and dilute the median — Fresh Kills in New York, one of our two pre-registered control sites, failed our bar for exactly this reason while NASA's own portal shows its mound sinking at ~43 mm a year. Third, the bar is deliberately strict: half the site window must be sinking for the median to clear it. The full CSV lists every site we measured, hit or not.

Is this a list of America’s fastest-sinking places overall?

No — it is landfills only, and it is local motion only. The satellite product we use removes everything broader than about 30–50 km by construction, so regional groundwater bowls like Houston or the Central Valley are invisible here on purpose. Every number is motion relative to the ground around it. For fast-sinking places in general, with causes tested one by one, see our separate analysis of unexplained sinking clusters.

Are these vertical velocities?

They are line-of-sight velocities: motion toward or away from the satellite, which looks down at a 30–40 degree slant. If the true motion is straight down, the vertical rate is roughly 1.15 to 1.3 times larger than the number we print — so quoting line-of-sight is the conservative choice. Negative means away from the satellite: down, if the motion is vertical.