Where a river begins

Why can two atlases disagree about where a river begins without either being wrong?

Open two reputable atlases at the Nile and you may find two different sources, hundreds of kilometres apart. Do the same for the Amazon and you will find a live disagreement about which Peruvian headstream counts. It is tempting to conclude that one of them has simply not checked, or that better surveying will settle it.

But consider: we have satellite imagery of every square metre of those basins, and the disagreement persists. So let us ask a more uncomfortable question. Is "where the river begins" the kind of thing that has an answer waiting to be measured — or is it the kind of thing that has to be decided first?

Walk up a river and see what happens. Below, a broad single channel. Higher, it splits into two comparable branches, then each of those splits again, and again. Nothing about the water tells you which branch is the river and which is a tributary; both carry water, both carry the same name only by convention. So the first thing you notice is that a river does not come to a point. It comes to a fan.

To pick one thread from the fan, you need a rule. And several perfectly reasonable rules are available. Which do you choose?

You could take the longest thread — follow, at every fork, whichever branch measured from the mouth is longer. That is the rule that makes the Nile's source the Ruvyironza in Burundi rather than Lake Victoria, and it is the rule that keeps the Nile-versus-Amazon length argument alive. You could take the greatest flow — follow the branch carrying the most water. That is arguably more physical, and it can point somewhere completely different, because a long thin stream and a short fat one are different superlatives. You could take the highest elevation, which favours a glacier or a mountain spring. Or you could take the branch that carries the river's name, which is a matter of local custom and history and often decides the answer on maps regardless of hydrology — the Mississippi is conventionally sourced at Lake Itasca even though its Missouri branch is considerably longer.

None of those rules is wrong. They are answers to different questions. So an atlas disagreement is often not a disagreement about the world at all; it is two publishers applying different conventions and both reporting honestly.

Now suppose we agree on a rule — say, longest thread. Are we done? Follow the chosen branch up and it narrows to a stream, then a rill, then a wet groove in the hillside that flows after rain and not in August, then a boggy patch with no channel at all, then damp ground. Where in that sequence did the river start? Ask yourself what measurement could decide it. The channel head migrates up and down the slope with the seasons and with the year's rainfall; two surveys in different months would place it in different spots and both would be correct on the day.

This is the deeper limit, and it is not about instruments. The quantity we are asking for — an exact point — is finer-grained than the thing itself possesses. The river's upper end is genuinely fuzzy and genuinely mobile. Demanding a coordinate is demanding precision the system does not carry.

Which is why the practical resolutions are conventions, honestly labelled: the Danube is customarily begun at the confluence of the Breg and Brigach, with the ornamental spring at Donaueschingen kept as a ceremonial source; the Amazon's source has been reassigned more than once as expeditions applied different criteria to different Andean headstreams. The published coordinate is real. What it records is the rule, not a discovery.

Two things are being conflated in the original puzzle, and separating them makes the disagreement stop being mysterious. The first is a choice of criterion: longest, largest, highest, or customary. That choice is a human decision, made before any fieldwork, and different bodies make it differently. The second is a limit of resolution: even under a fixed criterion, the top of a drainage network has no sharp terminus to find. The first produces disagreements of hundreds of kilometres; the second produces disagreements of hundreds of metres. Both are real, and neither is settled by better equipment.

The misconception worth correcting gently is that this makes the question meaningless or purely arbitrary. It does not. Once a criterion is stated, the answer is objective, checkable, and can be got wrong — a claim about the longest headstream is falsifiable, and expeditions have in fact overturned earlier claims by measuring more carefully. What is arbitrary is which question we chose to ask, not the answer we then get.

And notice the useful general shape here. When a well-resourced field sustains a long-running disagreement about a simple-sounding fact, the reliable first move is not to ask who has the better data. It is to ask whether the quantity is defined finely enough for data to decide.

• The Nile-versus-Amazon length contest, and how much of it is a measurement question versus a definitional one.
• Strahler stream order: numbering branches by structure rather than arguing about which is the main stem.
• Ephemeral and intermittent streams, and how hydrologists draw legal boundaries around waterways that are dry half the year.

Every term the collection defines is gathered in the glossary.