Climate risk and nature risk are disclosed separately, modelled separately, and usually owned by different teams. Physically they are one thing. The result is 2 legitimate ways of computing the same loss — and no agreed way of combining them.
One flood, 2 descriptions
Picture a factory beside a river, with a catchment of fields and soil upstream. One winter the river rises and the factory floods.
Was that a climate risk or a nature risk? It was both, and it was one event. The rain fell because of the weather. It reached the factory because the soil upstream no longer held it back. One river, one soil, one winter — but the loss will appear in 2 separate reports, produced by 2 different teams, using 2 different models.
This is not a failure of either model. Both are right about what they describe. The problem is what happens when their 2 outputs meet.
Why the split exists
Climate was framed first: the TCFD published its recommendations in 2017, and a decade of practice, data providers and in-house expertise grew on that frame. Nature came later and borrowed the same architecture — the TNFD follows the TCFD's 4 pillars almost line for line. That was pragmatic, and it produced 2 parallel modelling routes for events the river does not distinguish.
The 2 routes to the same flood
The climate route starts from the peril. Flooding is one item on a list — the European Taxonomy fixes 28 such hazards, from heat to wind to water. For our factory it asks 2 questions: does the flood reach it (exposure), and how badly does it take the blow (vulnerability)?
The nature route starts from the dependency. It asks what the factory needs from the living world in order to run. Among other things, it needs the catchment upstream to hold water. When that service degrades, the loss follows. The 2 questions are the same: does the site depend on the service (dependency), and how badly does it suffer when the service fails (sensitivity)?
Both routes are the same logical machine with different inputs: a shock, a test of whether it reaches this site, a measure of how hard it lands.
The doubling is not specific to floods. For instance, a drought is an acute water-stress peril on the climate side and a failed water-regulation service on the nature side — the same dry riverbed, described twice.
Why the 2 numbers cannot be added — or kept apart
Flood models do not ignore the soil. They carry terrain, land cover and soil properties; the catchment upstream is in the model. What these models do not do is let the soil change. They take terrain, land cover and soil as they are today — usually from recent observations — and keep them fixed for every scenario. Only the climate moves. A 2050 flood scenario is therefore 2050 rainfall falling on 2026 soil: if the catchment degrades between now and then, the model will not see it.
Some methodologies are explicit about this. WRI’s Aqueduct Floods, one of the most widely used global flood datasets, states that floodplain restoration and coastal mangroves are not accounted for, and that only dikes are treated as protection. Others do carry the landscape — one calibrates infiltration and runoff against observed river discharge — but on a snapshot that does not move with the scenario.
This matters because the soil is precisely what sets the size of the loss. Soil that has lost its organic matter absorbs less water; the same rainfall produces more runoff and a deeper flood. The degraded soil does not add a second loss next to the flood. It changes how hard the flood lands. In the diagram above, that is the dashed line: the ecosystem's condition feeding into the climate route's vulnerability. Mangroves and floodplains play the same role against storm surge; tree canopy and soil moisture against heat.
Hence the 2 wrong answers:
- Add the climate loss and the nature loss, and you bill the same flood twice.
- Keep them apart, and you lose the amplification — which is the whole reason the pair is worth modelling.
What we do instead at Darwin. When a hazard and an ecosystem service describe the same event, we do not count 2 losses. The hazard model gives the loss. The ecosystem that absorbs the shock — the soil, the floodplain, the mangrove; call it the buffer — then scales that loss up or down depending on its condition. A healthy buffer reduces the loss, a degraded one increases it. One line, adjusted — not 2 lines, added.
The adjustment depends on how much of the buffer the hazard model already includes, and that varies from one model to another, so we check each one. If the methodology states that the buffer is not accounted for, we add its full effect. If the model already calibrates land cover against observed river flows, the buffer's baseline condition is already in the loss, and we only add the degradation that has occurred since that baseline. The scaling factor is a first cut and deliberately conservative: we would rather understate the interaction than invent a coefficient.
What this unlocks
Making the catchment's condition a variable — degraded further, or restored — and re-running the model gives 2 things the current setup cannot.
Separating the drivers. How much of the factory's 2050 flood loss comes from heavier rain, and how much from poorer soil? Today those arrive as one undifferentiated number. They are different problems with different owners and different remedies. Heavier rain is exogenous: you protect the asset, insure it or move it. A degrading catchment can be acted on directly — often on land you do not own but can influence.
Ranking sites correctly. 2 facilities with identical climate scores can carry very different real exposure, depending on the condition of the land around them. A model that holds that land still cannot tell them apart.
2 boxes for reporting, one model underneath
The reporting split will persist: it is written into regulation, org charts and disclosure calendars, and there is little point fighting it. Disclose against TCFD and TNFD separately where required. But the model underneath should not inherit that split. One river, one soil, one winter — the loss they produce should be computed once.
There is a further consequence. If the catchment's condition is a variable, it is not only a risk factor but a lever: restoring soil organic matter, planting riparian strips or reopening a floodplain upstream change the factory's own exposure, and are within reach of a single company acting on the land around its assets. That is the subject of the next article.
At Darwin, we model climate and nature as one exposure — the hazard gives the loss, the condition of the buffer around your assets scales it, and the split shows up only where regulation asks for it. Book a demo to see how your sites rank once the land around them stops being held still.