Ambridge, Pennsylvania  ·  40.584°N 80.221°W  ·  ground 908.6 ft

What's under the house?

Somewhere beneath this hill is a band of limestone two to three feet thick, packed with sea creatures that died about 300 million years ago, when this was a shallow tropical ocean. It's called the Ames. Nobody has ever drilled here to check.

So it had to be worked out.

Cross-section through the hill above Ambridge A slice through the hillside from the Ohio River on the left to the ridge top on the right. Rock layers run roughly level through the hill. The house sits at 908.6 feet. The Ames limestone lies at about 988 feet, which is above the ground at the house — it has been eroded away there — but is still present in the higher ground to the right. 1200 ft 1100 1000 900 800 700 Ohio River the house 908.6 ft 79 ft of air the Ames is gone here Pittsburgh coal 1,218 ft  · highest ridges only Ames limestone 988 ft  · the bed being looked for ↑ ox-blood red marker shale Pine Creek limestone ~853 ft  · marine fossils, still in the hill Mahoning coal  · 748 ft Upper Freeport coal  · 698 ft the bed the whole model is anchored to Brush Creek  · ~808 ft
A slice through the hill, drawn to scale in feet above sea level. The rock layers run roughly level; the hillside cuts across them. Where a layer sits above the ground line it has already been eroded away — which is exactly what happened to the Ames at the house.
The answer

The Ames is 79 feet above the roof.

It was there once. The hill has since worn down past it. To stand on the Ames you'd walk uphill, roughly a quarter-mile to the northeast.

But

Three more fossil layers are still underneath.

Woods Run, Pine Creek and Brush Creek — all marine limestones, all full of shells and crinoids. The nearest is roughly 55 feet below the ground, and the slope below the house cuts straight through them.

What was gathered

The library

Before anything could be worked out, the raw material had to be found, downloaded, catalogued and pinned to real coordinates — so the computer knows not just what a map says, but exactly where on Earth it says it.

21,444maps located and catalogued
18,907scanned underground mine maps
41,930source documents indexed
339 GBof scans downloaded and verified
38,000automated checks, just to find which files were real
1,209historic government topographic maps
25,357passages of geological text made searchable
29geological society field guidebooks
What was measured

The evidence

A model is only as good as the observations under it. These are the numbers that actually constrain where the rock is.

197,451coal-seam depths from Pennsylvania's own mining records, out to 25 miles
62,480points traced by hand off the contours of a 1975 state structure map
17,378observations pooled into the model that was finally adopted, each carrying its own error bar
3,041fossil records swept from two international databases within 50 miles
219,889provenance links — every number tied back to the document it came from
568 ftof hillside covered by laser survey from an aircraft, accurate to about a foot, mapped at one-metre spacing
The hard part

The model was built four times. Three were wrong.

Three geological folds run through this area, and one of them passes directly beneath the study site. A flat sheet cannot bend, and a tilted sheet cannot bend twice. That single fact killed the first three attempts.

A Flat layers Taken from a fossil-hunting website's general diagram. Deleted outright — it kept catching questions it had no business answering. deleted
B Flat layers Built from government reports of 1906 and 1911. Its geometry was wrong, but its measurements were sound and are still in use today. superseded
C One tilted slab A single plane fitted across the whole area. Landed about 90 feet too low, because the ground it was fitted to is folded, not tilted. superseded
D A bending surface A grid that can warp, built by interpolating 17,378 measurements with a separate uncertainty calculated for every cell. It follows the folds. in use
Proof

How we know it's right

Anyone can build a model that agrees with itself. These are tests against things the model had never seen.

19 ft

Hide the answers, then predict them

11,461 real measurements were withheld from the model and then predicted from scratch. Average miss: about 19 feet, on a rock column over 500 feet tall.

8 of 8

Find the folds

The model was asked where the three geological folds should cross the area. It found every one, within a few hundred metres of where the state maps them — and it was never told they existed.

1.2 ft

Against a map from 1911

A US Geological Survey sheet drawn in 1911 shows where the Ames comes out of the ground. Compared at 14,918 separate points, the 115-year-old map and the new model differ by about a foot. Within 500 metres of the house: eleven inches.

2 ft

Against a completely different layer

A coal seam 520 feet higher up is mapped on a nearby hilltop. Measured against the laser survey it sits at 1,222 feet. The model, working from the seam far below it, predicted 1,220.

The part most projects hide

Nine times the work was wrong — and it got caught

Every one of these was found, written down permanently with its reason, and fixed. That record is the reason the fourth model can be trusted at all.

  • →Four historical map sheets sat two kilometres out of position for weeks, while the notes recorded them as "checked visually, good."
  • →Three attempts at measuring the tilt of the rock produced perfectly believable statistics. One was wrong by a right angle — found only by drawing it on a map instead of trusting the summary table.
  • →A rule for telling two kinds of map line apart turned out to be exactly backwards.
  • →A claim that two rock layers "should not be modelled" was retracted the same day, once it emerged the 1911 report simply predated their names.
  • →A celebrated confirmation of the model turned out to be measuring nothing at all — the rock it referenced had eroded away 240 feet above where the line was drawn.
  • →A check that could only ever fail sat flagged as an error for weeks. The check was right; the model was wrong.

Six known problems are still on the books today — sized, dated, and pointed at the work that would close them. None of them are hidden.

Where it all came from

A 120-year span, made to agree on one hillside

Hand-lettered field notes from before the First World War, and laser altimetry from an aircraft — reconciled to the same few feet.

1824 Harmony Society archives 1906 USGS Beaver Quadrangle 1911 USGS Sewickley Folio 1927 Ambridge–Woodlawn bridge cut 1938 WPA mine maps 1963 Beaver County coal survey 1975 PA structure contours 1983 PA correlation chart 2019 aerial laser survey live state mining database live global fossil databases
What it's all for

Tomorrow, someone walks outside and checks.

The model says: go down the slope about 55 feet and there should be a thin limestone full of snails and shellfish. Go a quarter-mile uphill and the Ames should be there. There's even a field test printed in the 1911 report — look for a thick band of ox-blood red clay whose top sits ten to twenty feet below the limestone. No red bed underneath, and it's an impostor.

A prediction you can check by walking to it is the only kind worth making.

Ambridge, Pennsylvania · elevations in feet above sea level (NAVD88) · all figures drawn from the project's own database