Bogotá rationed water for about a year, from April 2024 to April 2025. The shortage showed up in the Chingaza reservoirs, so that is where the coverage pointed. But the rain that fills those reservoirs was made somewhere else first, by forests and grasslands hundreds or thousands of kilometres upwind, and none of that land appears in the city's water accounting.
That gap has a name in the science, and it is starting to get one in policy. If you run a utility, sit on a basin commission, or price drought for a living, it is worth knowing before it shows up across a negotiating table.
what a precipitationshed is
A precipitationshed is the upwind area of land and atmosphere whose evaporation ends up as a given place's rain. Keys, van der Ent, Gordon and colleagues coined the term in 2012 (Biogeosciences), deliberately mirroring the watershed: rivers connect upstream to downstream, and moisture connects upwind to downwind. They mapped precipitationsheds for seven rainfed farming regions and showed that each region's vulnerability depended on land use in source areas the farmers had never seen, often in another country. The point was not that rain is mysterious. It was that rain has an address, and the address is usually outside the basin.
A watershed tells you where water goes after it falls. A precipitationshed tells you where it came from before it fell. Nearly every water treaty on earth governs the first and is silent on the second.
The scale is not marginal. Globally, about 40 percent of the rain over land began as evaporation from land rather than ocean, and in parts of South America, Africa and Eurasia the share is 80 to 90 percent (Posada-Marín and colleagues, Nature Water, 2024). Agricultural areas in 155 countries rely on forests in other countries for up to 40 percent of their rainfall, and forest moisture supports roughly 18 percent of global crop production (Pranindita, Wang-Erlandsson and colleagues, Nature Water, 2025). In those places the basin on the map is a downstream fragment of the system that actually delivers the water.
the basin on the map
Water law is built on the river basin because the basin is what you can see and gauge. The Colorado River Compact apportions a river between an upper and a lower basin, measured at Lee Ferry. The 1959 Nile Waters Agreement divides flow measured at Aswan between Egypt and Sudan. These are real achievements, and this post is not about renegotiating either of them. For the Colorado, start with what a compact call actually is and what happens to the river after 2026.
The trouble is that the water being divided is not fully made inside the boundary being governed. The Nile is the clearest case. About 85 percent of the surface water reaching Aswan originates in the Ethiopian Highlands, which cover less than a tenth of the basin's area. Some of the moisture that rains on those highlands crosses the West African rainforest first, over Cameroon, Gabon, Congo and the Central African Republic, none of them Nile riparians. Gebrehiwot, Ellison and colleagues (WIREs Water, 2018) proposed, plainly, that the eleven Nile states would be well advised to build a cooperation framework with the rainforest states, because deforestation there is a plausible lever on flows here. Nobody has signed that yet. The paper's own name for the missing layer is the precipitationshed.
| river-basin treaty object | precipitationshed object | |
|---|---|---|
| boundary | topographic divide; a line you can survey | probabilistic moisture source region; a gradient, not a line |
| parties | riparian states and users inside the divide | upwind land managers, often non-riparian, sometimes another country |
| what is measured | flow at a gauge, storage in a reservoir, diversions | evaporation, moisture transport, share of rainfall attributable to a source |
| what is governed | how much each party may take | whether the source keeps producing, and how steadily |
| failure mode | over-allocation; calls and curtailment | the total shrinks or turns erratic while every allocation is honoured |
| existing law | mature: compacts, commissions, courts | thin: recommendations, one declaration without targets, no binding agreement |
The right-hand column is harder to write a treaty around. That is not a reason to pretend it does not exist. It is the reason the treaty has not been written yet.
the amazon is already forcing the question
In May 2026 Amazon Conservation published Keeping the Flying Rivers Flowing, a white paper that reads like an early term sheet for exactly this kind of agreement. Its subject is the moisture that travels from the Atlantic across Brazil and falls on the southwestern Amazon in Peru and Bolivia. Forests and grasslands there depend on recycled moisture for more than 70 percent of their annual rain, and the people living under it have no say over the land-use decisions in Brazil that determine whether it keeps arriving.
The paper traces three seasonal pathways using ERA5 reanalysis, and its first finding is about what already works. The wet-season pathway crosses the Guianas, Venezuela, the Colombian Amazon and Loreto in Peru, where forests sit inside protected areas or officially recognised Indigenous territories and enjoy relatively strong protection; the paper notes that every category of protection, strict reserves, sustainable-use areas and Indigenous lands alike, has measurably reduced deforestation. Much of the dry-season route is covered the same way. The exposure is concentrated where designation is absent. The dry-season pathway already crosses heavily deforested southern Pará, is cut by federal highway BR-319, whose paving could trigger up to five million hectares of additional clearing, and is interrupted by narrow stretches of undesignated public forest where deforestation risk runs high. All three pathways converge over Acre, where planned road links toward Pucallpa in Peru would sit under a year-round moisture corridor. The corridor holds today largely because of the people and designations already on it; the gaps are the problem precisely because the designated land works.
The 2023–2024 drought, the most severe on record in the Amazon, shows what the exposure looks like when it lands. Soy output in Santa Cruz, Bolivia fell 75 percent. Potato harvests in Puno, Peru dropped from 998,000 to 596,000 tonnes in a single year. The Madeira River, which draws about 60 percent of its discharge from Bolivian and Peruvian tributaries, fell to a 122-year low, halting barges at Porto Velho and cutting hydropower. The drought had several causes and the paper says so. What it demonstrates is the shape of the dependence: Brazil's upwind land helps decide Peru and Bolivia's rain, and that rain flows back into Brazil as river.
The policy asks follow. Amazon Conservation proposes a sixth criterion for protected-area designation and conservation finance, contribution to flying-river function, operationalised as Critical Moisture Territories, to guide the fate of roughly 50 million hectares of undesignated public forest where 26 to 30 percent of current Amazon deforestation happens. It notes that Brazil controls about 60 percent of the basin and near-total control over the moisture reaching Peru and Bolivia, that the 2023 Belém Declaration was a step forward in committing member states to avoid the Amazon's tipping point but set no quantitative targets by country and no differentiated responsibilities, and that progress "requires binding bilateral or multilateral agreements — modeled on transboundary water management frameworks". That is a precipitationshed treaty in everything but name, addressed to the Amazon Cooperation Treaty Organization.
The objection a diplomat raises first is the one the paper raises itself. Why would the upwind state agree? A precipitationshed agreement asks Brazil to accept a continuing positive duty to keep forest standing for the benefit of neighbours who are not even riparians of the rivers involved, and the paper is blunt that ACTO's consensus-based decision-making "creates systematic obstacles to effective action by granting veto power to any single member state". The Nile case earlier in this post runs the same direction: upstream states have spent a century declining downstream obligations, and the rainforest states have never been asked at all. A precipitationshed treaty asks for more than a basin treaty, not less. That is why it has not been written, and why the number attached to the dependence has to be one the upwind party can check.
What is that rain worth? Baker and colleagues (Communications Earth & Environment, 2026) estimate that each square metre of Amazon forest adds about 300 litres a year to regional rainfall, which at Brazil's agricultural water price works out to roughly US$59 per hectare and about US$20 billion a year across the Brazilian Legal Amazon. The authors flag the price assumption as deliberately simple. Take it as a floor on what a treaty would be protecting, not as what the forest is worth.
a public scientist put it in one sentence
Jay Gutierrez, a systems biologist who builds graph intelligence for living systems, drew the implication out in public in July 2026: "The next water treaty may not be written only around a river basin, but around a precipitationshed." His follow-on point is the one governments and underwriters should hear. Treaties, insurance, lending and commodity risk do not move on metaphor. They need accountable numbers, with the uncertainty attached.
That is the question behind Fluvion, his public engine for following moisture from a defined upwind forest to a downwind beneficiary and stating the dependence as a number with a grade beside it. The first corridor, Amazon to the La Plata soy belt, is priced and backtested; we cover it in we've priced rivers. not the ones in the sky. The second corridor is the one that matters here. The same moisture map, re-read at two Andean water towers, asks how much of Quito's and Bogotá's rain the Amazon makes.
His published readings are deliberately unequal. For Quito, the Amazon is the single largest source of rain on the páramo grasslands that feed the city in a typical year, about a third of it in the public data, and the prevailing wind carries Amazon moisture straight up the eastern slope; he frames Quito as Amazon-facing and rates the attribution at medium confidence. For Bogotá the Amazon is a minority source, roughly 13 percent, behind the closer Orinoco basin at roughly 21 percent, and he frames it as mixed-source at low-to-medium confidence rather than forcing the claim. Both attributions pass his first evidence gate with caveats attached. The map works in grid cells of roughly a degree, uses a 2008–2017 climatology with no year-to-year signal, and any water-supply valuation stays locked until observed river flow clears the same baseline test the soy corridor passed. None of it is a forecast, and none of it explains any single drought, including Bogotá's. We collaborate with Jay. The engine, the corridors and the caveats are his, and we are not claiming they are wired into anyone's treaty or anyone's pricing, including ours.
What the Andean work shows is the object a negotiator would need: a named source region, a mechanism, a flow with a share and a stated confidence, a beneficiary, and only then an obligation. Everything before the obligation is science and accounting. The obligation is where the treaty starts.
The reservoir level is a stock metric; it tells you how much water remains. The precipitationshed is the graph; it tells you whether the rain keeps arriving.
what a basin-only view costs you
Here is the loss that basin-only governance carries, in the terms each reader already uses.
A utility plans against gauges, reservoir curves and a rainfall record. Every one of those sits downstream of the precipitationshed. If upwind land cover changes, the record you calibrated on describes a source that no longer exists, and you find out through the reservoir, a year late, as Bogotá did.
An insurer prices agricultural or hydropower drought on basin history. The tail you underwrite is partly set by forest condition two thousand kilometres away. The public evidence Fluvion leans on suggests degraded forest does not make less rain on average; it makes rain less steadily, and his page treats even that as a correlate still being validated, not a proven driver. If it holds, it moves variance and the drought tail, not the mean, which is exactly the part of the distribution you are paid to get right.
A government negotiates with the parties inside the divide. If the source is outside it, the negotiation can succeed on its own terms and still fail the population it serves. Posada-Marín and colleagues assessed 379 transboundary basins with an upwind lens and found that the volume of vegetation water requirements sitting under very high risk rose from about 20,500 to 32,900 cubic kilometres a year once governance and environmental performance in upwind source regions were counted. Read that metric carefully: it is the water that crops and natural vegetation need to grow, the paper's own definition, not human withdrawal, which runs at roughly 4,000 cubic kilometres a year worldwide. The point is the ratio, not the absolute. Counting the upwind term moved the assessed risk by more than half. The risk was already there. The basin view could not see it.
There is partial precedent for governing an atmospheric object. The 1979 Geneva Convention on Long-Range Transboundary Air Pollution regulates what blows across borders using source–receptor modelling that is probabilistic by nature, has been extended by eight protocols, and is widely credited with cutting European sulphur emissions. But the analogy stops at the physics. Air-pollution law imposes a negative duty on the party causing harm: stop emitting. A precipitationshed treaty would ask an upwind state for a positive, continuing duty to keep producing a benefit for people outside its borders and outside its river basins. International law has governed a probabilistic atmospheric flow before. It has not yet governed one in that direction, and anyone drafting the first such agreement should say so.
widen the map, then fund the source
None of this means the river basin stops mattering. Upstream still decides a great deal of downstream water; we have written about how upstream land decides downstream water and about why you can't seed a cloud that isn't there. The precipitationshed is the layer above both: the reason there is water in the headwaters to argue about.
The practical move for a beneficiary does not wait for a treaty. Once a source is named, whether a Critical Moisture Territory in Acre, a páramo above Quito, or an upwind forest whose share of your rain someone has computed with the uncertainty attached, that source is a natural asset with a location, a condition, and people already stewarding it. That is something a utility, a municipality or a reinsurer can fund now, as a dependency it has, rather than as philanthropy. In ensurance, a named natural asset can hold an onchain account and issue a certificate that funds its present condition, so a beneficiary can pay the source directly while the treaty is still being drafted. That is the whole gloss. Our stage is early: live instruments, small volumes, real places. The graph that says which upwind land your rain leans on is Jay's work and others', not ours.
A treaty is the widest version of the same idea. The Amazon Conservation paper, the Nile proposal and Jay's one-line implication all point at the same object: the upwind land that makes the rain, put on the map and given parties. Whether the next water treaty is written around it is not ours to decide. That it will need to be written around something more than the basin is getting harder to argue with.
If you run water for a city, a basin or a book of agricultural risk and are working out how to fund an upwind source you already depend on, talk to someone who can help →. We can connect you with the public science and the people doing the attribution, and show what funding a named source looks like in practice. If you want the ground-level version first, start with the water cycle, broken, and how to put it back.