LIVE — free dataset, CC-BY 4.0

landfill_reality_delta

We compared what a US federal program declares about how fast a landfill is filling up against what an independent LiDAR sensor actually measures on the surface, for 330 real landfills. The headline result isn't "declared data is wrong" — it's that we found and measured, for the first time at scale, a real blind spot in surface-LiDAR landfill monitoring itself, including our own method.

What we compared

The EPA's Landfill Methane Outreach Program (LMOP) publishes an annual waste-acceptance rate for thousands of US landfills — how many tons per year each site is declared to be taking in. Separately, NKO's measure_physical_change() — the same method already validated on 4 independent real sites in 2 countries — measures the real physical volume change of a site between two independent public LiDAR surveys (USGS 3DEP).

We multiplied each site's declared rate by the years between its two LiDAR surveys to get an expected volume of new waste, converted through a cited EPA density range (waste gets denser as it settles, so we always show both a fresh and a settled scenario, never one cherry-picked number), and compared that to what LiDAR actually measured on the surface.

The blind spot — the central finding, not a footnote

In 317 of 330 sites (96%), the surface measured far less volume gain than the declared rate implies. Before jumping to "the declared numbers must be wrong" — the least likely explanation, and one we don't make without direct evidence — there's a much simpler physical reason: modern engineered landfill cells are built by first excavating below the original ground level (required by US federal regulation, 40 CFR Part 258, for the composite liner system) before waste ever mounds above it. A surface LiDAR survey only sees height change relative to the original ground. It cannot see a cell filling up below its own rim.

This has always been a known theoretical limitation of surface-LiDAR change detection. This dataset is the first time it's been quantified at scale (330 sites) instead of stated as a caveat on a single site.

Three real, documented cases

CONFIRMS — St. Joseph City SLF, Missouri. Declared 1.66 million tons added 2013–2021, but only +96,411 m³ measured at the surface (−94% to −96%). A February 2025 local news report confirms this exact facility's standard practice: crews routinely excavate new cells roughly 30 feet (9 m) below original grade — the exact mechanism this dataset's central finding proposes, independently confirmed for this specific site.

SUPPORTS — Ocean County Landfill, New Jersey. Declared 4.67 million tons added 2010–2018, −104% to −107% delta (the surface actually contracted). Independent reporting confirms a real, ongoing "phased excavation and base liner construction" program at this facility — below-grade cell construction is this site's documented operating practice, though we can't pin the exact depth to the exact years measured.

RULED OUT — Bridgeton Landfill, Missouri has the single strongest delta in the whole dataset — but we explicitly exclude it as evidence for the blind-spot hypothesis. Missouri's Department of Natural Resources maintains an extensive public record showing the real cause is a subsurface smoldering event (detected since December 2010) that caused genuine surface subsidence — a different, well-documented physical phenomenon, not invisible below-grade filling. We report the number honestly and flag why it doesn't belong in the pattern we're describing — using it anyway would have been misleading, even though it would have made the headline more dramatic.

Full sourcing for all three, and an honest account of which other candidate sites we checked and could not confirm, is in documented_cases.md inside the dataset package.

Checked against a control group before trusting any of this

Before computing a single delta, we fixed the method, thresholds and a control group in writing (47 landfills that closed before their first LiDAR survey, with zero declared acceptance rate — sites where we'd expect to see essentially no growth). Only 8 of those 47 (17%) showed unexplained growth, well under the 50% threshold we'd set in advance to stop and re-check the method. The pattern in the main 330 sites isn't a measurement artifact.

Get the data

330 sites, one row each: declared tons, measured m³, delta under two density scenarios, and an honest label per site (CONSISTENT_WITH_DECLARED / UNRESOLVED_SIGNAL_BELOW_DECLARED / UNRESOLVED_SIGNAL_ABOVE_DECLARED — never "discrepancy", because the tons-to-m³ conversion is a cited range, not a measured constant). Preregistration, the reproducible script, the real EPA source table, and the control-group results are all in the package.

Zenodo (DOI) → Kaggle → 536 free per-site physical_change pages →