Post date 5 Oct 2026 (original LinkedIn date)
Original LinkedIn version· 2,437 words
‘Hypnotized’ by Fleetwood Mac, written and sung by Bob Welch on their 1973 album, ‘Mystery to me’. This was during their transitional era, post Peter Green and pre Buckingham and Nicks, with the McVies, Mick Fleetwood, and the two Bobs (Welch and Weston).
The New Forest is lovely at almost any time of year, but I think it is particularly beautiful and rich on a sunny autumn day. Supporting an ultra run this weekend left me with plenty of time to reflect on an excellent challenge/provocation posed by Henry Leveson-Gower FRSA (you can find Henry’s reflections here:
https://lnkd.in/p/e-jvWdcF
with some interesting BTL comments too). There is an obvious attraction to the idea of paying for water outcomes, in the sense that if we can create contracts around carbon sequestration, why not around reduced flood risk, improved drought resilience, cleaner water or some other measurable change in the way a catchment behaves? We all get that land managers can restore peat, reconnect rivers with their floodplains, create wetlands, slow flows and change agricultural practices. And we can reasonably surmise that somebody downstream stands to benefit from those interventions, and a contract would provide the mechanism through which value is transferred back upstream. That, at least in broad terms, has been part of the thinking behind Resilience Purchase Agreements. I still firmly believe that there is something incredibly important in the idea, but Henry’s recent challenge to the concept made me wonder whether we have been too casual about what, exactly, the buyer is purchasing, and perhaps even about what we mean by an outcome in the first place.
The problem is relatively easy to describe if I’ve understood the challenge correctly. All agree that it is perfectly possible to contract for actions: restore this area of peat, create this wetland, maintain these leaky barriers, change this land-management practice. But it is much harder to contract with the same confidence for a defined downstream result, because the route from intervention to outcome runs through a system shaped by rainfall, soil moisture, drainage, other land-use decisions, grey infrastructure, catchment condition and, increasingly, a climate that is making the historical record a less reliable guide to the future. Let’s call it complexity!
Hydrological modelling helps, of course, and in many river restoration projects it already provides the basis for estimating the likely effect of interventions, often with some allowance for uncertainty. But a buffer does not make the model right; rather it simply allows us to be more conservative about how much of the modelled benefit we are prepared to rely upon. Let us imagine that a suite of interventions is modelled to reduce peak flow at a downstream location by 12%, but that the realised effect, insofar as we could ever establish it cleanly, is closer to 9%. Has the land manager failed, even though every intervention was delivered and maintained exactly as agreed? Has the modeller failed or has the project developer failed? Or have we simply encountered the entirely predictable difference between a modelled expectation and the behaviour of a complex natural system?
It seems unreasonable to push that risk back to the farmer or land manager on their own, because doing so would effectively ask them to underwrite not only the interventions they control but also the hydrology of the catchment, the future climate and the actions of everybody else within it. Yet the opposite position is not especially satisfactory either. A downstream buyer is unlikely to want to pay indefinitely for a resilience benefit that it cannot have reasonable confidence has been delivered. The danger is that this leaves nature-based solutions in a sadly familiar place: everybody agrees that the intervention is sensible, everybody acknowledges that it is likely to provide value, but nobody can get comfortable enough with the uncertainty to turn that value into a contract.
I am not convinced that the answer is to abandon outcome-based thinking. Perhaps the issue is not uncertainty itself, but who carries it. There is a useful parallel here in a paper by Michel Loreau and colleagues on the idea of biodiversity as insurance. Their argument is not really about financial contracts at all. It is about the way more diverse ecological systems can behave under variable conditions. Different species and functional groups do not respond to environmental change in exactly the same way or at exactly the same time, so losses in one part of the system can be partly offset by gains elsewhere. The result can be lower variability in aggregate ecosystem functioning.
That distinction between average performance and stability turns out, I think, to be quite an important one. Biodiversity can affect the mean level of ecosystem functioning, but it can also reduce its variability through time, and those are not the same thing. I think that is highly relevant to the way we talk about water resilience. We have tended to ask how much flood reduction an intervention produces, as though the only useful number is an average change in peak flow, expected annual loss or some similar central estimate. But perhaps part of the value sits elsewhere. A more diverse, better functioning catchment may also be less volatile, less brittle and less prone to extreme deterioration when conditions move away from the norm. That sounds much more like resilience and it means that the objective of an intervention might not simply be to produce a larger expected outcome, but to improve the distribution of possible outcomes.
This is, of course, very familiar territory in finance. Diversification is valuable not because every asset performs better, but because different assets respond differently to the same conditions. The resulting portfolio may have lower volatility even if its expected return is not dramatically higher. Loreau and colleagues make the same connection, while being careful not to pretend that ecological and financial portfolios are identical. That helps us with the catchment problem because a catchment-scale nature intervention should probably not be thought of as a single asset producing a single output anyway - arguably it is much closer to a portfolio situation.
Wetlands, peatlands, riparian woodland, floodplain reconnection, soil restoration, leaky barriers and land-management changes will respond differently to different events and will matter differently in different places. Some will be more valuable in smaller, frequent rainfall events. Others may matter more during prolonged wet periods or drought. Some will contribute disproportionately to water quality or biodiversity even where their marginal flood benefit is smaller. Their value lies partly (significantly?) in combination.
The paper goes further with the idea of “spatial insurance”, where ecological stability emerges not only through time but across landscapes because different communities and species respond differently in different places. That seems particularly relevant to catchment finance. We may be making a mistake when we try to value each individual intervention too precisely in isolation, when much of the resilience benefit may emerge from the collection as a whole. Maybe that should also change how we think about failure. A flood is not evidence that a nature-based intervention has failed, any more than a market downturn proves that diversification has no value. The relevant question is whether the system behaved differently from the way it would otherwise have behaved. Was the peak lower or was it delayed? Was expected annual loss reduced? Was water retained for longer in the right part of the catchment? Was there greater stability across a range of conditions?
Conventional infrastructure already works with this sort of uncertainty. A reservoir is not financed because somebody knows precisely what demand will look like thirty years from now, nor is a flood defence built because an engineer can guarantee that rainfall will remain within a particular envelope. Infrastructure is designed around models, tolerances, contingencies, safety factors, reserves, warranties and insurance, with different risks allocated to the parties best able to manage or absorb them. There is no obvious reason why nature infrastructure should be expected to meet a fundamentally different standard. Once we look at the problem in those terms, is it reasonable to separate out the risks?
There is delivery risk ie was the intervention actually created and maintained as specified? There is performance or model risk ie did the intervention affect the system in the way anticipated? And there is hazard risk, namely did an event occur that sat materially outside the conditions for which the system had been designed? it doesn’t seem defensible that those risks should necessarily sit in the same place. A land manager can reasonably be held responsible for whether the agreed intervention was delivered and maintained. It is much harder to argue that they should also carry the risk that a model overestimated the resulting attenuation, or that an extreme rainfall event overwhelmed an otherwise effective intervention. We would not regard a conventional flood defence as defective simply because an event exceeded its design standard, yet there is sometimes a tendency to judge nature-based solutions against precisely that kind of binary test. That feels increasingly worrisome and wrong to me.
Could it then be that the relevant question is not whether an intervention removes risk altogether - very few interventions of any kind do that. The true question should be whether it changes the distribution of risk. If a model suggests that a package of interventions is expected to reduce a particular risk measure by 12%, there is no requirement to contract on the basis that 12% is an absolute truth. The contract might recognise only 8% of that benefit, with the difference providing a buffer against ordinary model uncertainty. The numbers would obviously vary from project to project, but the broader principle is familiar enough: do not monetise the full central estimate where uncertainty is material. That is how finance deals with uncertainty all the time. Banks haircut collateral values, and infra projects carry contingencies. Insurers hold capital against adverse outcomes, while carbon projects retain buffers against particular forms of non-delivery risk. None of those devices removes uncertainty, they just make it manageable.
Is this where the insurance question becomes more interesting, because there may actually be two quite different forms of insurance at work. The first is ecological - a more diverse and functioning landscape can provide a form of biological insurance by reducing the variability of the underlying system. Loreau and colleagues are quite explicit that this is principally a buffering effect, and they also warn against treating ecological “insurance” as though it were identical to a financial insurance contract. The second is financial: once the ecological system has reduced some of the physical risk, there may still be residual model risk, delivery risk or tail risk that can sit within a reserve, guarantee, insurance policy or reinsurance structure. Financial insurance does something different, in that the policyholder pays somebody else to assume defined elements of uncertainty. Those two layers could ultimately sit together, so that we end up with the landscape reducing physical volatility and the financial structure dealing with what remains. That feels rather more convincing than asking either the land manager or the buyer to absorb all of the uncertainty themselves!
There is also something useful in the distinction between being outcome-led and making every payment conditional on a tightly specified outcome. Those are not the same thing and as we do our catchment work, maybe we don’t need them to be. A system in which payments depend entirely upon demonstrating a particular downstream hydrological result years after the intervention may become so tangled in questions of attribution, baselines and counterfactuals that very little ever gets financed. Yet a system based solely upon paying for activities has its own weakness, because we end up knowing exactly how many hectares were restored and how many interventions were installed without necessarily knowing whether they materially altered the risks that motivated the investment.
There must be a middle ground. Outcomes can determine what we are trying to achieve, where interventions should be placed, what should be modelled and monitored, and how the programme should adapt over time, while the contract itself recognises that different parties should carry different elements of the resulting uncertainty. This may also be where we need to stop looking quite so hard for nature’s equivalent of the tonne of carbon. Carbon has given environmental markets a wonderfully convenient abstraction, but nature does not behave that way. The same hectare of restored land can affect flood risk, drought resilience, water quality, biodiversity, carbon and soil condition simultaneously, and its contribution depends heavily on where it sits within the wider system. That complexity is sometimes presented as the reason nature cannot be financed.
I’ve said this before in a previous article, but I can’t help but wonder if we may be trying to standardise the wrong thing. We do not need to standardise the ecology but we can standardise the wrapper around it: how the baseline is established, how uncertainty is represented, how much of modelled performance can legitimately be contracted, how much is retained as a buffer, what is monitored, what happens when the system behaves differently from expectation, and which risks belong with the land manager, the project, the buyer or an insurer. Those elements can all be standardised even where the ecosystems cannot.
And perhaps that is where the Resilience Purchase Agreement becomes most interesting and relevant. What the buyer may ultimately be purchasing is not an “outcome” in the narrow sense at all. It may be a conservatively modelled, independently assessed and continually maintained reduction in the variability and downside risk of a physical system. That seems much closer to green infra than to an environmental credit. It also gets us away from pretending that the river can be made predictable. It cannot, and indeed, the whole point of resilience is that the future is uncertain.
This then leaves us with the job of the ecological intervention being to make the system less fragile in the face of that uncertainty, while the job of the financial structure is to decide who carries what is left. Who should underwrite the river?
References
Loreau, M., Barbier, M., Filotas, E., Gravel, D., Isbell, F., Miller, S.J., Montoya, J.M., Wang, S., Aussenac, R., Germain, R., Thompson, P.L., Gonzalez, A. & Dee, L.E. (2021). Biodiversity as insurance: from concept to measurement and application. Biological Reviews, 96, 2333–2354. https://onlinelibrary.wiley.com/doi/epdf/10.1111/brv.12756
Environment Agency & Flood and Coastal Erosion Risk Management Research and Development Programme (2025). Natural Flood Management evidence. GOV.UK Natural Flood Management evidence
Kelso, M.A., Taylor-Burns, R., Franco, G., Ingram, J.C., Reguero, B.G. et al. (2026). Advancing nature-based insurance to protect people and nature. Communications Earth & Environment, 7, 540. https://www.researchgate.net/publication/407608267_Advancing_nature-based_insurance_to_protect_people_and_nature