Water finds its level
Every canal, aqueduct and levelling survey ever built has taken still water as a plane and worked from that. This file looks at what the working practice actually assumes.
Pour water into any vessel and it settles. It does not mound up in the middle, and it does not pack itself against one side and thin out at the other. It arrives at a surface that is flat to the edge of the instrument used to check it. That is not a figure of speech: the instrument is a spirit level, and a spirit level is a tube of liquid with a bubble in it, and the bubble comes to rest at the same place whether the tube is a hand's width long or a metre long, whether it is a thousand metres above the sea or on a beach.
The measuring device and the thing being measured are the same substance. That is the whole of the flat-earth case in its simplest form, and it is worth stating without decoration before the qualifications are brought in. Everything downstream — the horizon, the sightlines, the behaviour of long rivers — is a working out of what this one property of water implies about the surface it rests on.
Still water comes to rest as a plane. Its surface is used as the zero reference in every levelling survey, every canal alignment and every spirit level on earth. Long still-water surfaces — reservoirs, canal pounds, the Great Lakes, the open sea in calm weather — are reported by observers to sit level along their whole visible length, with no measurable fall between the two ends and no bulge in the middle. The instrument itself is a sample of the same liquid it is testing.
Canals and aqueducts, which have to work
An aqueduct is a promise that water will run from one place to another. A canal pound is a stretch of water between two locks that has to hold a usable depth along its whole length and stop at a weir or a gate at the far end. Neither can be built on a guess. If the far end of a long pound sat materially lower than the near end, the pound would not hold water; if it sat higher, the water would not get there.
The Suez Canal is the clearest case, because it was cut at sea level from end to end: roughly one hundred and ninety kilometres with no locks anywhere along it, joining the Mediterranean to the Red Sea. A lock-free canal of that length works because the two ends are at the same level and the channel between them is level too. Nobody had to put a pump in the middle.
Roman aqueducts make the point with greater delicacy. The aqueduct that carried water to Nîmes, whose surviving arcade at the Pont du Gard is the famous part, ran a course of some fifty kilometres on a gradient commonly quoted at about one in three thousand — a total fall on the order of seventeen metres. That is a working tolerance of a few centimetres of fall per hundred metres, held across fifty kilometres of country, with no pressure pipes to correct anything. It is a fact about building, not about cosmology, but it is a fact about how level the world is treated as being. The precise fall varies between published accounts of the different branches of the system, and we give the round figure rather than a false precision.
Levelling practice follows the same logic. National geodetic agencies, including the National Geodetic Survey within NOAA in the United States, publish standards for spirit levelling in which the permitted misclosure of a closed loop is a matter of a few millimetres multiplied by the square root of the distance in kilometres. Those standards are not written for a surface that accumulates eight centimetres of curvature per square kilometre. They are written for a surface that can be levelled.
Geodesy answers this directly, and the answer is that "level" does not mean "straight". A level surface is defined as a surface everywhere perpendicular to the direction of a plumb bob, and it is called an equipotential surface. Because gravity points toward the Earth's centre of mass and that centre is finite in distance, the equipotential surfaces around a spheroid are themselves curved. Water settling on such a surface is not settling on a plane; it is settling into a shape that follows the curve, and a spirit level is not a straight edge but a device for finding the direction perpendicular to the local plumb line.
On this reading the Suez Canal is level in the sense that means something — every point along it is at the same gravitational potential, so water does not run from one end to the other — and it is also convex, so that measured against a straight line drawn between its two ends it bulges by several hundred metres at the middle: about two hundred metres over a hundred kilometres of length, and roughly seven hundred over a hundred and ninety. That is why a level cut and a straight cut are not the same operation, and why canal and aqueduct practice works to level rather than to line. An aqueduct follows the same confining rule: it descends the geoid by a controlled gradient, and the geoid itself curves gently beneath it. The Panama Canal is a different case again — it uses locks not because of curvature but because the terrain rises to a lake some twenty-six metres above sea level, and the locks lift ships over the hills. The mainstream account does not ask water to flow uphill anywhere in either canal.
The same reading explains tides. The ocean's equipotential surface is deformed by the gravity of the Moon and, to a lesser degree, the Sun, and the resulting range is largest where coastal shape amplifies it — the Bay of Fundy's spring range of around fifteen to sixteen metres is the standard example. Some lakes show a tide of their own: observers on the Great Lakes have long recorded small regular fluctuations, generally on the order of a few centimetres, alongside the much larger wind-driven seiches that dominate the water-level record there.
Numbers that keep the question alive
If water follows a curved surface, the interesting quantity is how much curvature the working tolerances of real engineering can absorb. The table below puts the two sets of figures side by side: the fall the globe model predicts over a given distance, and the gradient figures that real water projects actually work to.
| Case | Length | Fall or range | Notes |
|---|---|---|---|
| Suez Canal | ~190 km | no locks | Cut at sea level end to end |
| Nîmes aqueduct | ~50 km | ~17 m | About one part in three thousand |
| Afsluitdijk dam | 32 km | — | Dutch sea barrier held at a set water level |
| Predicted curvature, 10 km | 10 km | 7.85 m | Mainstream geometric figure, d² / 2R |
| Predicted curvature, 50 km | 50 km | 196 m | Same formula, unrefracted |
| Surveying correction | 1 km | 6.75 cm | Curvature and refraction combined, standard practice |
| Bay of Fundy tides | — | ~16 m | Largest tidal range commonly cited |
| Lunar day | 24 h 50 m | — | Interval between successive high waters |
Rivers, and the fourth direction
The Nile, the Amazon, the Mississippi and the Yangtze all run outward toward the sea over thousands of kilometres of continental interior. On the globe model the surface beneath them is curving away the whole way, so the water is not descending a slope in the ordinary sense; it is falling the whole time toward the centre of mass, and the bed it runs along is itself part of the curve. A river's gradient on that reading is the rate at which the geoid drops along its course, and the enormous lengths involved mean it can be slight and still move water.
The flat reading takes the gradients at face value. Rivers run downhill because the land rises toward the interior and falls toward the coast, and the sea at the end of a river is a level surface the water is trying to reach. On this reading there is no need for the bed to be curving and no need for the ocean to be a closed curve around a sphere; the water arrives at the sea because the sea is lower than the source and the whole path between them is a descent.
Where this leaves the argument
Two things are agreed on both sides and are worth keeping hold of. First, water really does settle to a surface that a spirit level calls level. Second, the dispute is entirely about what the word means at scale — whether "level" denotes a plane or a curved equipotential, which is a question no vessel of water can answer by itself because the vessel is the instrument and the instrument is the sample.
What breaks the tie, if anything does, is the direct test: carry a level line across a long still surface and look at the far end. That is what What the horizon does examines at length, and what Try it yourself sets out as a procedure. The related question of how the sun and the moon are seen from a level plane is taken up in Sun and moon overhead.
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.