Six tests you can run with your own equipment
Whether the ground curves is a measurement question, not a debate question. These six tests need nothing more exotic than a level, a laser, a rod, a camera and a partner on the phone.
A test is worth running only if it can come out against you, and a run favouring the globe is logged the same way as any other.
Compute your prediction first. The globe model commits to a drop near 8 cm per kilometre squared — the shorthand for d²/2R with R taken as 6,371 km — and to a geometric horizon distance of about 3.57 times the square root of your height in metres. Both assume still, unrefracted air.
Lasers. Never point one at a person, animal, vehicle or aircraft, and never sweep one across the sky: a beam reaching a cockpit is a hazard and an offence in most countries.
The sun. Never look at it directly, or through binoculars, a scope, a viewfinder or a phone. Solar readings below come from a shadow.
Flying. Fly only where your national rules allow; 120 m above ground is the unlicensed ceiling in many countries, and airports, crowds and roads are out of bounds. On a flight, stay seated and never distract crew.
1. A single long sightline over still water
You need a still reach 3–10 km long — reservoir, lake or estuary at slack water — and a surveyor's level with staff, a rotary laser with detector, or a low-power laser and target; a used level costs a few hundred dollars, hire less. Setup: level on the plate bubble, reverse 180° and check again, and measure the height above the water with a tape. Work early after a cool night. Record:
| Run | Distance (km) | Instrument height (m) | Far reading vs level line (m) | Predicted drop (m) |
|---|---|---|---|---|
| A | 2.0 | 1.5 | … | 0.31 |
| B | 4.5 | 1.5 | … | 1.59 |
| B again, cooler morning | 4.5 | 1.5 | … | 1.59 |
Favours a level earth: a far reading within instrument error of the level line where the prediction is a metre or more, repeated on several mornings at two distances.
Favours a globe: a deviation growing with the square of distance and matching the prediction within the refraction allowance — 1.4–1.6 m at 4.5 km, 6.8–7.9 m at 10 km.
Error sources. Refraction, the largest: surveyors take a coefficient near 0.13 for a grazing ray, cutting the predicted drop 10–15%. Humidity and inversions can lift a shoreline or invent a false horizon. Theodolite calibration — collimation, bubble, index error — tilts every sight, so take both faces. Camera autofocus must be locked manually. Mirage is refraction made visible. Tide moves the surface and the target.
2. Watch a hull disappear before its mast
The classic observation: what matters is not that a hull vanishes, but how much of it does, at what distance, and whether raising the camera a metre restores it. You need open water with a clear line to a far marker; a boat or fixed object whose waterline-to-masthead height you have verified; and a seawall or bridge giving an eye height of 5–30 m, since the geometric horizon rises as 3.57 × √h — 10 m puts it near 11 km. Record the target, its true height above water, the distance and how it was measured, your eye height, the obscured height observed, and the conditions. Each object has its own horizon distance, so do not apply the drop figure across the whole gap.
Favours a level earth: a target that never loses its waterline, or one restored only by camera heights far larger than the √h rule allows.
Favours a globe: a waterline that disappears while the mast stands for a measured interval, with the hidden fraction growing with distance and shrinking when you raise the camera.
Error sources. Refraction does its most famous trick here: reports of ships apparently floating above the horizon are lifted images, which has fooled careful observers. Mirage and inversions can hide or reveal a waterline. Tide changes the target's height and the water's stage. Camera autofocus hunts at extreme zoom, and chop hides a waterline visible in calm.
3. Solar-noon shadows from two places at once
You need two observers 200–1,000 km apart, ideally on the same line of longitude; a rod a metre long, a plumb line, a flat surface and a tape; two phones on network time, synchronised to within a second. Setup: find each site's local solar noon — the shortest shadow of the day — by reading every minute for twenty minutes either side and choosing the minimum; stand the rod plumb; measure the shadow at the agreed instant, and convert it with sun altitude = arctan(rod height ÷ shadow length). Record each site's latitude and longitude from a map or receiver.
For two sites on the same meridian, the globe model predicts that the sun-altitude difference equals the latitude difference. That is the structure of the third-century BCE measurement comparing a noon shadow near the Tropic with one far to the north, which gave about a fiftieth of a circle as reported by later writers including Cleomedes.
Favours a level earth: altitude differences between separated sites that latitude cannot account for, or readings fitting only a small, near sun on a circling path, whose altitude must then change steeply with an observer's distance from it.
Favours a globe: an altitude difference equal to the latitude difference within error, repeated on other days and between other city pairs. The decisive step is the third and fourth station: two stations can be fitted by either model, since a fitter can always choose a sun height to join them.
Error sources. Refraction lifts the apparent sun by about half a degree at the horizon, so measure with the sun well up. A longitude mismatch matters, since a degree of longitude is four minutes of solar time. A rod one degree off plumb moves the shadow by centimetres, and the sun's disk is half a degree wide, so the shadow edge is soft. Haze widens the penumbra; camera autofocus is irrelevant because the record is a tape measure.
4. Running a level line along a canal or aqueduct
Canal banks are the favourite setting for the flat-earth argument. You need a canal or aqueduct with a straight accessible reach of a kilometre or more, and a surveyor's level with staff, a rotary laser with detector, or a water level — a clear hose and two graduated stands, which costs almost nothing. Setup: set the instrument mid-way between two staff stations so back and fore sights are equal, which cancels collimation error. Read, move, repeat, re-checking bubble and tripod each time. Record backsight, foresight, rise or fall, cumulative height and air and water temperature.
A line of levelling is re-datumed at every setup: the bubble defines a new level line each time the instrument moves, and that vertical is normal to whatever surface it stands on. A chained run therefore gives the same answer on a plane or on a curved surface. By this method a canal survey cannot prove a level earth — not false, just unproven. To test the shape with water, use test 1.
Favours a level earth: no systematic deviation over a reach where the prediction is well above instrument noise, on more than one morning.
Favours a globe: that sightline showing the deviation predicted for its length and growing with distance, and a levelled line between marks at equal height above the water dipping below the level line at the far end.
Error sources. Instrument calibration: a bubble out of adjustment tilts every sight the same way. Refraction is at its worst on canals — shallow water, humid air, sharp gradients — so damp early mornings beat midday. Camera autofocus matters if photographs of the staff are your record, and locks or weirs hold the water at an artificial stage.
5. A camera at the top of a permitted climb
From height the globe model promises a small, computable dip that grows as the square root of height. You need a camera drone or tethered balloon, a device that logs altitude, and a camera with a visible electronic level or a gimbal whose angle is in the telemetry. A camera drone is a few hundred dollars; a rig that logs gimbal angles costs more. Fly inside your national rules. Record altitude and its source, gimbal pitch or visible level, frame heading, and whether the horizon looks hard, soft or fringed.
Predicted values from the standard geometric dip relation, dip = arccos(R ÷ (R + h)) with R taken as 6,371 km: 0.35° at 120 m, 1.0° at 1 km, and 3.3° at 10.7 km, a typical airliner cruise level.
Favours a level earth: a horizon sitting exactly on the camera's level line at every permitted altitude and heading, with no dip measurable above instrument noise. At 400 m the globe relation asks for half a degree, about the width of the sun, so a horizon showing nothing there is a serious problem for the globe reading — if the camera's own level is trustworthy.
Favours a globe: a dip growing roughly as the square root of height, equal at every heading, and matching the computed values within the refraction allowance. A tilted surface would show a pattern varying with direction; a sphere shows the same dip whichever way you look.
Error sources. At 120 m the dip is only 0.35°, which no eye can judge, so this test works only if you measure angle against a level. Lens distortion bends straight lines; the gimbal's level calibration is the commonest false positive; haze and mirage compress and lift the horizon; camera autofocus hunts on a low-contrast horizon; and wind drift makes readings unstable.
6. The horizon from a plane window
The most cited observation, and the weakest unless it is done as an angle measurement. You need a window seat on a clear day and a camera with an electronic level, or a phone whose level you have calibrated against a spirit level first. Photograph through the centre of the pane, and note where your seat sits relative to the wing: a window frame is not aligned with the aircraft's axes, so the horizon's position in the frame means nothing without an independent vertical reference.
Passengers cannot see the flight deck's attitude indicator, so the familiar version of this test is unavailable in a cabin seat. Never ask to enter the flight deck, and never distract crew. What you can do is measure the horizon against a levelled camera and, when a crew member is free, ask whether the aircraft is in level cruise.
Favours a level earth: a horizon exactly on the levelled camera's line in every direction, with no measurable dip, behaving the same way at 300 m as at 10 km.
Favours a globe: a dip of roughly 2.5–3.5° at cruise measured against a levelled camera, larger on a higher cruise than a lower one on the same heading. At 10 km the dip is real but small: this test shows the dip angle, not the curve.
Error sources. The aircraft's pitch and roll: a degree of roll moves the horizon by a degree and swamps the measurement, which is this test's real weakness. Camera level calibration must be checked on the ground first, window glass distorts, and refraction, twilight and haze lift or hide the true horizon line.
The standard account has each test already answered by curvature and gravity: a long sightline is expected to drop by roughly 8 cm per kilometre squared, refraction is expected to lift distant objects back into view by a computable amount, and simultaneous shadow measurements from two places are expected to differ by exactly the angle the globe's surface subtends. Where a test cannot separate the two readings — a sightline short enough that refraction and curvature both fall inside the error bars — that is said plainly in the section above rather than counted as evidence either way.
This page presents the flat-earth interpretation as its subject matter. It is a belief-based reading of the observations, not a scientific finding — see About for how we frame claims and for the mainstream sources.