We've priced rivers for centuries. Water rights, riparian law, hydro concessions, barge tariffs, the compacts that split the Colorado down to the acre-foot. Every river with a bank has a ledger somewhere.
The rivers in the sky have neither. A flying river — a stream of moisture a forest breathes out, carried by wind, falling as rain a continent away — feeds some of the world's largest harvests and appears on no one's balance sheet. Not the farmer's, not the crusher's, not the bank's, not the reinsurer's. It has no address, so nobody can fund it, watch it, or thank it.
This post is about giving it one. Not how to make rain — that is a different post. This is about following the river back to its source, reading the exposure honestly, and being clear about what a price on that exposure is and is not.
what flying rivers are
Flying rivers are the popular name for large, persistent flows of atmospheric moisture: water that plants transpire, that prevailing winds carry, and that falls as rain hundreds or thousands of kilometres downwind. You cannot stand on one's bank. On a clear day you can fly through it and never know.
The scientific object underneath the name is the precipitationshed, a term Keys and colleagues introduced in 2012. A watershed is the land that drains to a point after rain falls. A precipitationshed is the land that supplied the rain before it fell — the watershed idea, run backwards through the air.
A flying river is a precipitationshed you can follow: the upwind land whose evaporation becomes a downwind region's rain.
The mechanics are not exotic. Roots lift water from soil. Leaves release it as vapor. In the Amazon that pump runs at continental scale: Baker and colleagues estimate each square metre of Amazon forest adds about 300 litres a year to regional rainfall. Trade winds push that moisture west until the Andes turn it south, where it falls on the La Plata basin — the soy and maize belt of southern Brazil, Paraguay, and northern Argentina. Zemp and colleagues mapped that network in 2014: in the wet season, 18 to 23 percent of the La Plata basin's rainfall originates in the Amazon, rising to 24 to 29 percent once moisture that re-evaporates along the way is counted. Shares move with season, method, and year, which is why the literature is usually quoted as a band rather than a point.
Two things follow. The forest is a supplier — of water, delivered by air, to farms that have never contracted with it. And a supplier has a location. That is what makes the river priceable at all, and what makes the honesty around the price matter so much.
how much rides on it
Two recent papers put scale on this, each with its own method. Neither is Fluvion's, and their numbers do not add.
Pranindita, Teuling, Fetzer, and Wang-Erlandsson (Nature Water, 2025) paired global moisture tracking with crop production and trade data. Agricultural areas in 155 countries rely on forests across a border for up to 40 percent of their annual precipitation. Moisture from forests supports about 18 percent of the crop production they studied and about 30 percent of crop exports. Their point is structural: importers are exposed to upwind forests that appear in no supplier list.
Baker and colleagues (Communications Earth & Environment, 2026) asked a narrower question: what is Amazon rainfall generation worth if you price the water at what Brazilian agriculture pays for it? Using the IBGE and ANA agricultural water price of about US$0.0198 per cubic metre and their 300 litres per square metre, they get about US$59.40 per hectare per year — roughly US$20 billion, give or take seven, across the 330 million hectares of the Brazilian Legal Amazon. They call it a simple approach with a constant unit price, and they are right to. It is a rainfall-generation value, not a loss model, and not a market price.
Hold those numbers as what they are: evidence of scale. What none of them does is tell one desk which forest its harvest leans on, by how much, and how sure to be.
why your nature-risk score cannot see it
Nature-risk platforms do a real job. TNFD's LEAP process asks you to locate your interfaces with nature. GIST Impact and NatureAlpha map millions of asset locations against biodiversity and ecosystem data. If you hold South American soy — as a crusher, a feed importer, an ag lender, or the reinsurer behind the crop policy — those tools will tell you about water stress and habitat where the asset sits.
They will not tell you that a fifth of the rain on that asset is made two thousand kilometres upwind, in a forest that appears in none of your location data. Location is not pathway. The site score reads the sink. The dependency lives along the edge that connects sink to source.
| question | site-score nature-risk tool | upwind moisture pathway |
|---|---|---|
| what it locates | where your asset sits | where your asset's rain is made |
| unit of analysis | site, footprint, biodiversity or water-stress score | source cell → moisture flow → downwind harvest |
| direction of the question | outward from the company | backward from the harvest to the forest |
| what it can miss | little that is local | the supplier you never contracted |
| what an underwriter gets | exposure, by location | a dependency you can size, and a source you can watch |
| what a TNFD lead gets | the interface, located | the mechanism behind the interface |
| what the forest gets | nothing | a named position in the supply line |
This is the gap the rest of this series is about: stock metrics tell you how much nature remains; the graph tells you whether the service keeps arriving. The pillar post is the missing object is the graph. Here, the graph is a river of air.
fluvion: following the water back, then pricing the dependency
Fluvion is Jay Gutierrez's public engine for exactly this — "flying rivers, priced" — built as teleconnected water value-at-risk. We collaborate with Jay; Fluvion is his, and everything below is from his public page, with his caveats kept intact.
The corridor. Three standing forests in western Amazonas, 324,542 hectares between them, upwind of the La Plata soy belt. Fluvion takes moisture flow from RECON, a published moisture-tracking dataset, reconciles it to ERA5 reanalysis, and puts the Amazon's annual share of La Plata rainfall at 21.5 percent — inside the 12 to 35 percent band he benchmarks against Zemp and colleagues. His own seasonal reading, separate from Zemp's: the Amazon delivers far more water to the belt in the wet season, but its share of the belt's rain peaks near 27 percent in the dry season, when little else falls. Running his own tracking model, WAM2layers, is his stated next milestone, not something he claims today.
The backtest. Twenty-two harvest-years of IBGE crop statistics. The 2021–22 La Niña drought cut Rio Grande do Sul's soy by more than half while the cerrado held; the engine reproduces that split, and repeats it on 2024, a year held out in advance, at a correlation of r = 0.45. His own words: the skill is real but modest — it recovers the sign and rank of which regions fall, not the exact loss — and it is a backtest, not a forecast.
The price. As a rain machine, a hectare of the corridor forest is worth about $350 — his indicative value of the soy harvests it protects across the whole downwind belt, counted over thirty years — against about $295 a hectare that the land itself sells for in Brazil's registry. Every one of his qualifiers travels with that number: one corridor, one crop, water only, one place the rain lands, Phase 1B (his label for an early-stage, indicative build), not prudential grade.
The graph, not the lane. Underneath the corridor the same moisture flows form a directed, weighted network of thousands of cells. Pointed at the soy belt, the top tenth of upwind source cells supply about 48 percent of its imported moisture — and the cells that carry the load are not simply the biggest evaporators. Position, not mass. He then ran that result against null models and reported that most of the concentration is the plain geometry of a corridor between source and sink: decision-useful, he says, not Nature-grade. That sentence is worth more than the map.
Grade beside the price. Forest condition — read from evaporation volatility and published resilience indicators — moves the spread and the drought tail. It never touches the $350. It sits beside the price, not inside it, because he treats it as a lagging correlate still being validated, not a proven driver.
What an underwriter should hear in all that is not poetry. It is a dependency with a source, a magnitude with a stated confidence, a backtest with a published skill, and a model that tells you where it stops.
what the $350 is, and what it is not
It is Fluvion's indicative avoided-loss value for one corridor, one crop, one moisture service, over thirty years, at Phase 1B. It is what a hectare of that forest's rain is worth to the whole soy belt downwind of it — shared across every farm, crusher, and lender in that belt, so no single desk's dependency is that number. It is an order-of-magnitude anchor, not a payable, and nobody is invoiced on the strength of an r of 0.45.
It is not the forest's worth. It is not Baker's $59.40 per hectare, and the two neither add nor rank: Baker prices one year of total rainfall volume at a flat agricultural water price across the whole Legal Amazon; Fluvion counts thirty years of avoided loss on one crop, in one place the rain lands, for water only. Different method, different scale, different question — not a bigger and a smaller estimate of the same thing. It is not a figure from RealValue, our own natural capital accounting engine, which values a parcel's full set of ecosystem services against its measured condition rather than one corridor's rain. It is not the price of any ensurance instrument, and Fluvion is not wired into how any of our instruments are priced.
The number is an anchor for what the corridor's rain is worth to the harvest downwind, not a ceiling on the forest's worth.
The conservationist in the room is right to bristle at "$350 a hectare." The forest was there before the number, and it holds far more than soy rain: carbon, species, the Andes' own water towers, the people who live in it and have never been asked. A price is a bridge. It gives a soy desk a reason to care about a forest in Amazonas it will never visit. It does not give the forest a ceiling. The moment a model says a forest is worth $350, the model has stopped doing its job and started doing something else.
from an address to a funded source
Pricing the desk's downside and funding the living source are two different jobs, and the second only becomes possible once the first has named the pathway.
The beneficiary is not innocent, either: soy expansion is among the leading drivers of the clearing in the Amazon and Cerrado that threatens the rain machine, which is exactly why a priced dependency could bite where a disclosure does not.
Once you can say this forest → this moisture flow → this harvest → these balance sheets, the beneficiaries have a supplier with an address. That is where ensurance sits: a way for the people who depend on a named source to fund its present condition upfront and hold that as an asset, rather than disclosing the exposure and waiting. Certificates tie to a named natural asset; coins fund protection across many. Proactive, not a claim after the drought.
Our stage is honest too: live agents, coins, and certificates at small volumes. What Fluvion does — price the dependency for the desks that carry the downside, grade the source for the capital that would protect it — and what ensurance does — let that capital fund and hold the source — are complementary, and separate. He follows the river back. We give the people at the end of it a way to pay for the beginning.
where to go next
- Read the pillar: the missing object is the graph — why nature-related financial risk is a pathway, not a site score.
- If you came here to make it rain: you can't seed a cloud that isn't there — moisture recycling, and why seeding is the wrong lever.
- See how a named source becomes something you can fund: natural assets.
- Read the engine on its own terms: Fluvion — Jay Gutierrez's public corridor, backtest, and caveats.