Your nature-risk report can tell you that a supplier's site sits in a water-stressed basin with declining ecosystem integrity. That is real work, and it is worth having.
It cannot tell you which forest makes the rain that grows the crop on that site, how much of the rain comes from there, or what happens to the harvest — and to your loss ratio — if that forest thins.
Those are different questions. The first one is about a location. The second one is about a pathway. Almost every tool in nature finance today answers the first, and the second is the one an underwriting decision would actually need.
what nature-related financial risk actually is
Nature-related financial risk is the risk that a business, a portfolio, or an insured asset loses value because an ecosystem service it depends on — water, rainfall, pollination, soil fertility, flood buffering, disease regulation — degrades, becomes unreliable, or stops arriving.
Read that definition again and notice where the risk actually sits. It is not in the land. It is in the arrival. A forest two thousand kilometers upwind can be in perfect condition and still be irrelevant to your position. A modest wetland six kilometers upstream can be the only reason your plant has a water permit. Condition and consequence are separate variables, and only one of them shows up on a map of where your assets are.
So there are two questions a risk desk can ask about nature, and they are not interchangeable:
- What condition is this place in? Answerable today, at global scale, by several competent vendors.
- Who depends on this place, along what mechanism, and by how much? Answerable regionally in the scientific literature, rarely answerable at the resolution of a named source and a named beneficiary, and rarely asked by a risk desk — which is inconvenient, because it is the one your cashflow actually rides on.
There is a cheap test in that pair. Ask whoever built your nature-risk score which of the two questions it answers. If the answer is the first, you have an inventory, not a risk.
the site score and the pathway are not the same object
The Taskforce on Nature-related Financial Disclosures gave the market its working method: LEAP — locate your interface with nature, evaluate dependencies and impacts, assess risks and opportunities, prepare to respond and report. It is a good method, and it asks the right question. Evaluate is explicitly about dependencies: which ecosystem services does this business rely on, and how much.
The gap is not in what LEAP asks. It is in the resolution of what a filer can answer it with. In practice, Evaluate is discharged with a sector-level dependency rating — this business line depends materially on surface water, or on flood protection — attached to the asset's location. That is a type of dependency pinned to a place. It is not a named source, a named mechanism, and a named beneficiary, with a confidence on each link. The commercial layer inherited the same resolution: map millions of asset locations against biodiversity and ecosystem datasets, score the surrounding condition, flag the hotspots. Genuinely useful. Also silent about direction and distance.
The Green Finance Institute's UK assessment is the cleanest illustration of why that silence is expensive. It models nature degradation as reducing UK GDP by 6% under its domestic and international scenarios and up to 12% under its most severe health scenario by the 2030s — and it finds that roughly half of the UK's nature-related financial risk originates overseas. Half the risk is not where the assets are. A tool anchored to asset locations will, by construction, look straight past it.
| stock metric / site score | dependency graph | |
|---|---|---|
| the question it answers | What condition is this place in? | Who depends on this place, along what mechanism, by how much? |
| the object | A location, scored | A directed edge from a source to a beneficiary |
| unit of analysis | Hectare, parcel, portfolio average | Source → mechanism → flow → named beneficiary |
| what "important" means | Large, rare, degraded, or nearby | Load-bearing — the position other nodes lean on |
| the failure it can see | The place gets worse | The service stops arriving reliably |
| what it can support | Disclosure, screening, exclusion | Pricing, triage, funding a specific source |
| honest weakness | Blind to direction and distance | Model-dependent attribution; links unvalidated until backtested; has not yet beaten a simple baseline everywhere it has been tried |
the biggest patch is rarely the load-bearing one
In any network, importance is a position, not a mass. Jay Gutierrez puts it plainly on Fluvion: a source forest is not a patch of green in the picture, it is a position, and two forests that look identical from the road can sit in completely different places in the wiring. The wiring, not the canopy, sets how much a downstream economy leans on either one. The useful question stops being where does the water start and becomes which positions carry the load.
Two recent papers show what happens when someone actually follows the edge instead of scoring the node.
Baker and colleagues (Communications Earth & Environment, 2026) estimated that each square meter of Amazon forest contributes about 300 liters of rainfall to the surrounding region each year. Applying a constant water unit price — a simplification the authors flag themselves — they put Amazon rainfall generation at roughly US$59 (± 22) per hectare per year, and the Brazilian Legal Amazon's rainfall generation service at about US$20 ± 7 billion annually.
Pranindita and colleagues, with Lan Wang-Erlandsson (Nature Water, 2025), traced moisture from forests to farmland worldwide. Forest moisture supports about 18% of global crop production, and agricultural areas in 155 countries depend on forests in other countries for as much as 40% of their annual precipitation.
Neither of those is a site score. Both are statements about pathways — about where a service is produced, where it lands, and who is holding the downside if it stops. That is the object nature finance is missing, and it is the object this series is about.
why stock metrics keep coming up short
A credit desk does not rate a borrower by counting the buildings in the borrower's zip code. It reconstructs the obligation: who owes what to whom, on what schedule, secured by what. Nature-risk practice today is closer to the zip-code method than anyone in the industry would like to admit.
The reason is not laziness. It is that stock inventories are tractable and dependency graphs are hard. Satellites make condition cheap to observe. Nobody's satellite sees an obligation. Reconstructing one means naming a source, naming the mechanism that carries value away from it, naming the beneficiary where that value lands as a cashflow, and then being honest about how confident you are in each link.
Three objections are worth taking head-on, because they are the ones a serious desk raises.
"This is TNFD with extra steps." LEAP already asks the pathway question — Evaluate wants to know which dependencies are material and by how much. The problem is the resolution of the answer. What most filers can produce is a sector typology mapped to a location, and a typology cannot name the specific upstream or upwind node producing the service that crosses your interface, or tell you what is happening to it. That is a defensible disclosure and a weak pricing input. The gap is resolution, not intent, which makes it a modeling problem rather than a disclosure problem.
"No model is good enough to underwrite a forest." Today, mostly correct. Which is why the engines worth trusting publish where they fail. Jay Gutierrez's public Upper Colorado corridor is the model case: the method and the success bar were sealed in a hashed, timestamped record before anyone saw a result, the scores were then checked against real USGS and EPA monitoring data, and the published verdict was that a two-variable baseline of elevation and percent developed beat the custom engine. The engine then recalibrated its own confidence downward. The same page is candid that this first run's inputs are modeled stand-ins rather than live per-watershed pulls. That is what a model you could eventually underwrite with looks like on the way up — and it is the subject of a later post in this series.
"Fine, but I cannot buy a pathway." Not yet, and that is the honest state of the market. But you can fund the source at the head of one, and that is a different transaction from either a disclosure or a payout.
the work being done in public
Both of those examples are Jay Gutierrez's. He is a PhD working at the intersection of ecological network science, graph intelligence, and nature-finance translation, and he has been publishing this layer in the open rather than behind a sales process. The third engine is AI Ecologist, which turns a species list into a map of what holds an ecosystem together and where it breaks first.
His framing of the gap is sharper than ours: capital cannot underwrite a public good it cannot point to. Existing tools can score a site's ecological condition, but almost none can say who downstream actually depends on it, or by how much.
We collaborate with him; the engines are his.
what changes once the pathway has a name
A named pathway converts a diffuse worry into a specific counterparty relationship: this source, this mechanism, this flow, this beneficiary, with this much confidence. Once that exists, three things become possible that were not possible from a heat map.
Triage. You can rank sources by how much load they carry for your specific position, rather than by how green they look or how close they are.
That triage cuts both ways, and the limit is worth stating plainly. A dependency graph inherits whichever beneficiaries you choose to draw into it. Weighted only by insured cashflow, it maps capital's dependencies, not the ecosystem's — the subsistence household, the species downstream of nobody's balance sheet, and the forest's own claim on itself all drop out of the ranking. That is precisely why the transaction this series lands on is funding the source in its present condition rather than pricing the desk's downside. A price is a bridge to the source. It is not a statement of what the source is worth.
Pricing that admits its limits. Condition can move the spread and the tail without pretending to move the headline number. A grade sitting beside a price is more useful to an underwriter than a single confident figure with no error bar.
Funding the source in its present condition. This is where ensurance comes in, and it is a plain idea: instead of only disclosing an exposure and paying out after the loss, a beneficiary funds the protection of the source now. In practice that runs through two ordinary instruments — coins, which fund protection across the protocol, and certificates, which fund one named natural asset directly.
We should be equally plain about our own stage. The instruments are live and the volumes are small. Collaborating with the people building the graph layer is not the same as having their engines in production pricing, and we are not going to claim otherwise.
The sequence, though, is the point. Name the pathway. Grade the source. Then let the beneficiary fund it. You cannot do the third without the first, which is why the missing object matters more than any single number that gets attached to it.
frequently asked questions
what is nature-related financial risk?
Nature-related financial risk is the risk that a business, portfolio, or insured asset loses value because an ecosystem service it depends on — water, rainfall, pollination, soil fertility, flood buffering, disease regulation — degrades, becomes unreliable, or stops arriving. It includes physical risk from ecosystem failure, transition risk from policy and market shifts, and systemic risk where many balance sheets lean on the same natural source. It is distinct from nature impact, which is what a business does to nature rather than what nature failing does to the business.
how is a dependency graph different from a nature-risk score?
A nature-risk score answers "what condition is this place in?" A dependency graph answers "who depends on this place, along which mechanism, and by how much?" The score is anchored to a location and is silent about direction and distance; the graph is anchored to an edge running from a source, through a mechanism, to a named beneficiary. A score supports disclosure and screening. A graph is what you need before you can price, triage, or fund a specific source. The two are complements, not substitutes. Stock metrics tell you how much nature remains. The graph tells you whether the service keeps arriving.
what is graph intelligence for nature?
Graph intelligence for nature is the practice of representing living systems as networks — places, species, and flows as nodes and directed edges — and then computing on that structure to find which positions are load-bearing and where the system breaks first. It borrows from network ecology, where importance was long ago shown to be about position rather than mass. In a finance context it produces something a score cannot: a traceable chain from a source, through a mechanism, to a beneficiary whose cashflow is exposed, with an uncertainty attached to each link.
where to go next
If you came here from a disclosure workstream, read the dependency your map can't see next — it covers why the value produced by a source shows up on the neighbor's balance sheet and nowhere else. If you came from an asset or facilities angle, your place depends on a place walks the same logic through ski resorts, hospitals, and data centers.
Then continue through this series: exposure is not risk takes the argument to the standards desk, and we've priced rivers. not the ones in the sky follows a single dependency two thousand kilometers upwind.
And if you want to see what a named source looks like when someone is actually funding it, open a certificate tied to a specific natural asset. The front door, if you would rather start wider, is ensurance.app.