IAPB · Methodology
A method for calculating the return a company can expect from paying to restore the catchment its site depends on, and whether buying nature credits gives a better return than funding the same work directly.
This methodology covers three of the tool's five use cases: nuclear plants, hydropower stations and data centres. They share one methodology because all three depend on the site's water supply.
This document sets out a method for working out the return a company can expect from paying to restore the catchment its site depends on for water, and from buying nature credits for that work. It explains the question the method answers, the data it uses, how a change in the catchment becomes a change in what the site is worth, and which of its inputs are evidence and which are judgement.
The returns from protecting and restoring nature are documented for whole sectors and countries. There is little public analysis of the return to a single company, and none that measures what nature credits add to it. The method is meant for a company or investor to apply to their own business or client.
The document can be read without knowing the tool. Parts 1 to 6 describe the method. Part 7 says how far it can be trusted and what it leaves out. Part 8 describes what the programme does for nature. The appendices give what is needed to build the model again.
A company runs a site that needs water. It may be a nuclear plant cooled by a river or lake, a data centre that cools its halls with water, or a hydropower station that turns its turbines with the river. When the catchment, the land that drains to the site's water source (§2.4), runs short, the site loses money. That affects the company's finances and valuation.
The method answers two questions in turn. First, what financial value does a better-managed catchment, where more of the rain reaches the streams, offer the company and its business model? Second, does buying nature credits give a better return than funding the same work directly?
The answers are given as company value and cash flows, the measures a company uses for capital decisions.
| To | Read |
|---|---|
| Understand what the method does | Part 1 |
| Use the tool and read its results | Parts 1 and 6 |
| See how reliable the results are | Parts 1 and 7 |
| Build the model again | Parts 2 to 5 and the appendices |
| See what the programme does for nature | Part 8 |
The model values one site under two futures and measures the difference.
Both futures run through the same model, on the same data and assumptions. They differ only in whether the company pays for the programme. So the gap between the two answers comes from the work alone.
The model then compares two ways to pay. It values the same work a third time, paid for by buying credits instead of funding it yourself. So each site is valued three times: doing nothing, funding it yourself, and buying credits. The same work happens in the catchment on both routes that pay for it.
Figure 1 — Doing nothing and the programme, in the same model.
The return has a benefit side and a cost side.
The benefit has three parts. Each acts on a different line of the accounts, so they add up without counting anything twice.
The cost has two parts.
When the company buys credits, it pays for the work and for MRV and verification through the credit price. It pays a developer, who does the work. The price is set on the share of the work delivered. It adds a charge for the risk on the rest, and the developer's financing charge where payment waits for delivery (§5.3, §5.4). The whole credit price, MRV and verification included, is scaled by the payment factor. So at Reference a credit buyer pays 72% of the cost of MRV and verification, USD 0.36m against USD 0.50m at Santiago. The difference is part of risk transfer (§5.4), so MRV and verification are not a separate difference between the routes.
The tool works out each benefit part by switching that one thing off and running the model again. The last bar, "Everything else", shows any part of value created that the named parts do not explain, such as the effect of the parts on each other. At every site at Reference it is zero. Tax and timing are included in each bar. The tool shows this bar but does not present it as a separate benefit.
Figure 2 — How the return is made up.
The model reports the return in two ways. Both are divided by what you commit. They differ in how they measure the gain.
Value created = company value with the programme − company value doing nothing
(both modelled at the start of 2026)
Extra cash = Σ free cash flow with the programme − Σ free cash flow doing nothing
(2026 to 2060, undiscounted)
What you commit = Σ the company's own programme payments, 2026 to 2060, undiscounted
Times what you put in = value created ÷ what you commit (and extra cash ÷ what you commit)
The two measures answer different questions.
Neither measure is always the larger. They can rank the two routes differently, so both are reported. Each multiple compares a total with what you commit over 35 years. It is the value created, or the extra cash, for each USD 1 committed. It is not a yearly rate of return.
Figure 3 — Value created and extra cash.
How the tool shows these figures, and the rules it uses to round and word them, are in Appendix E.
A third figure, what doing nothing costs, answers a different question:
What doing nothing costs = present value, after tax, of the yearly loss to water doing nothing,
to the last model year (2060), discounted from the first year
at the site's cost of capital before any programme; the years after it are not counted
At a data centre or a nuclear plant, this figure compares the site with the same site with all the water it needs. A hydropower station has no demand to fall short of. There, the figure compares the station with the same station making its published output every year. Each figure is labelled with the comparison it uses. Value created compares the programme with the site as it is, so the two figures answer different questions. They cannot be added together. Company value also counts the years after 2060. This figure leaves them out, and every place that shows it says so.
The company does not own the catchment. The work happens on land held by others, so the company can pay for it in two ways.
Both routes pay for the same work and get the same physical benefit. So the operating benefit is the same on both routes. Cheaper borrowing is not, because buying credits also earns the credit premiums. What remains is cost, timing and the named differences in the table.
| Difference | Favours | How it works |
|---|---|---|
| Duration premium | Buy credits | A contracted, multi-year delivery, against spend renegotiated each year (§4.6) |
| Verification premium | Buy credits | A verified instrument, against a self-reported claim (§4.6) |
| Lower cost of debt | Buy credits | Included in the cost of capital when buying credits, weighted by the site's debt (§4.6) |
| Coordination | Buy credits | Funding it yourself pays once to organise the companies sharing the work (§5.5) |
| Risk transfer | Buy credits | Credits are priced on the share of the work delivered, plus a charge for the risk on the rest. Funding it yourself commits the spending whatever is delivered (§5.4) |
| Tax timing | Fund it yourself | An expense takes its tax relief at once. A credit recorded as an asset spreads it over 10 years (§4.3) |
| Developer's financing charge | Fund it yourself | Where payment waits for delivery, the developer borrows to pay for the work, and the cost of that borrowing is in the price (§5.3) |
A test checks the comparison. It switches every named difference off at once and sets the developer's rate equal to the site's cost of capital. The difference between the routes must then be exactly zero. If it were not, the result would include an advantage that is not listed (§7.4). The neutral developer's rate is one equal to the cost of capital. A rate of zero is not neutral, because an interest-free loan is itself an advantage.
The three credit premiums, the duration premium, the verification premium and the lower cost of debt, are judgements with no published source. At some sites they make up most of the difference between the routes. So the credits line in the tool's summary has a Judgement badge that links here. The difference shows which route is better, but not reliably by how much. At step 10 the tool splits the difference three ways, each worked out by running the model again: the cash the company actually saves, how the spend is treated in the accounts, and what banks and investors give it.
Figure 4 — Buying credits or funding it yourself.
Everything after this part works from one physical measure. For data centres and nuclear plants, it is the shortfall, the share of the water a site needs that the catchment cannot supply (§3.1). For a hydropower station, it is how much less the station generates than in a normal year, the median year of today's climate (§3.5). Each is measured with and without the programme.
The physical numbers do not come from this tool. They come from a separate model of the hydrology, the movement of water through the catchment, which this document calls the hydrology model. It combines river flow (GloFAS), rainfall (CHIRPS), a model of how much water the land yields (InVEST) and the World Resources Institute's Aqueduct 4.0 water-risk projections, over catchments drawn from HydroSHEDS. It runs a daily water balance for the catchment modelled for each site over 200 simulated weather years. The tool receives the results of our water calculations (Appendix A, A1 to A23). These re-run the daily water balance once, outside the tool, with our own groundwater rule and climate adjustments. The tool itself does not recalculate them.
At data centres and nuclear plants, each day, demand is the water used by the catchment's farms, industry and towns, plus the water the site itself consumes (§3.1). Supply is a share of the river's flow, plus groundwater. At the three data centres it also includes the recorded share of the city's water that comes from outside the catchment, such as by desalination or a transfer from another basin. That share is 90% at Las Vegas, 60% at Santiago and 50% at Barcelona. What supply cannot meet is the shortfall (§3.1).
For groundwater, we replaced the estimate in the hydrology results we received with one tied to recharge, the water that soaks into the ground. In our rule, groundwater is a share of the water the land yields each year. It is held in a store the size of 90 days of natural recharge. Each of the 200 simulated years starts with the store full. It refills by the same small amount each day, up to its size, with no daily limit on what is drawn. So groundwater can run short within a dry year. The simulated years are separate weather years, not a sequence, so the model cannot show groundwater drawn down over several dry years in a row (§7.6). The store applies at data centres and nuclear plants. This departs from the hydrology results we received, and at some sites it changes the answer a great deal (§7.5).
Each site has ten sets of water figures: today's climate, and three scenarios in each of 2030, 2050 and 2080. Between those years the tool draws straight lines: from today to 2030, from 2030 to 2050, and from 2050 on along the line towards the 2080 state. Each point is fixed to a calendar year, not a programme year (Appendix A, step 12).
The loss is the average across the 200 weather years. The shortfall that the tool shows is the median across those years, so half the years are worse and half are better. So is the number of shortfall days shown at data centres. A few very dry years can lift the average loss above what the median shortfall suggests.
The model has four kinds of work, each on 5%, 15% or 30% of the area it can treat, plus combinations of them. The tool values one of them, clearing invasive trees on 30% of the catchment. The model assumes invasive trees could be cleared on 30% of each catchment. It does not check how much of each catchment they actually cover.
| Work | What it does to the water | Area it can treat |
|---|---|---|
| Clearing invasive trees | 1,300 m³ more river flow per hectare a year | the whole catchment |
| Recharge basins that soak wet-season water into the ground | about 250 m³ a hectare a year into groundwater (330 m³, less a quarter lost to evaporation) | the whole catchment |
| Contour banks and gully repair on slopes | about 200 m³ a hectare a year into groundwater (400 m³, less half) | the whole catchment |
| Paying farmers to keep the ground covered | 10% less farm water demand on treated land, plus about 150 m³ a hectare a year of infiltration | cropland only |
The effects come from a literature review of each kind of work. What each kind costs is in §5.1.
The clearing coefficient is the water that clearing adds for each hectare: 1,300 m³ a year. Much of the result depends on it. It is at the top of the plausible range. It is reasonable for dense stands along rivers and optimistic for average land. The flat coefficient is not limited by the land's own water yield. At Almaraz it adds 2.3 times that yield. It is stated as a limit (§7.5).
Hydropower uses a different rule. At a hydropower station the model has only invasive clearing, on 5%, 15% or 30% of the catchment. The water added is a share of the catchment's own runoff, the rain that flows off the land into the streams. Treated land gives 20% more runoff, a judgement. So the water added can never exceed what the land yields, and it shrinks in a drier scenario. The rule applies at all four hydropower stations. The other types of site use the flat coefficient (§7.5).
We assume the added water arrives on days below the median flow. Scott, Le Maitre and Fairbanks (1998) found that trees reduce low flows by a larger share than annual flow, about 7.8% against 3.2%, which supports the direction. But by their totals only about 7% of the water trees remove is in the low flows. So putting all the added water on dry days probably overstates the water added when the river is low, and with it the gain at hydropower stations (§7.6). Their study is of commercial plantations, not the clearing of invasive trees.
Two points follow.
The three climate scenarios come from Aqueduct 4.0. Aqueduct's optimistic, business-as-usual and pessimistic scenarios each combine a climate path with a path for population and economic growth. The tool labels them better, middle and worse, with Aqueduct's own names in small print. It does not label them by emissions pathway, because each Aqueduct scenario also combines a path for population and economic growth.
We take Aqueduct's change in supply, demand and year-to-year variability, as a share of today's level, and apply it to our own figures for today. Then we run the water balance again. We never use Aqueduct's own amounts, which are drawn at a coarser scale.
The climate setting is not a scale from mild to severe. A worse scenario does not always mean less water at a given site. Aqueduct's own projections do not keep the three scenarios in order basin by basin, and the way water moves in a region can make the middle scenario the driest. At several sites in this set the path also turns after 2050. We report the scenarios as they are, without ranking them. Where it happens, the step says so on the page.
The climate is the same with or without the programme. It is outside the company's control. Single floods, storms and heatwaves are not modelled, except as part of the spread of dry and wet years.
The rule for choosing sites has three conditions. First, the site's model company is solvent, which means it can pay its debts. Second, doing nothing, the site loses at least USD 1m from water shortages in 2026. Third, buying credits creates value and extra cash with the reward from banks and investors set to none. The third condition means the list of sites does not depend on the most disputed judgement (§4.6). Funding it yourself is reported but not used to choose sites. The rule is applied to every modelled site, and selects 10 of the 29. One, a data centre in Chennai, is not shown because its water data is incomplete, so 9 are shown. Of the 29, 24 are solvent. All nine sites meet the second and third conditions with the model described here.
Some sites are not modelled at all. Catchment work cannot help a site whose own water comes wholly from sources the catchment does not control. That rules out sites that use sea water, piped recycled water, or closed-loop or air cooling, and sites supplied only by transfer from another basin.
The catchment is the area modelled for the site. For a hydropower station it is everything upstream of its intake, which is why the largest catchments belong to hydropower stations. A catchment is much larger than its site. Across the nine sites it runs from 50 km² to 26,700 km². Cost grows with the area treated (§5.1), while the benefit grows with the site's sales. So the size of the catchment has a large effect on the result (§5.2).
Cropland is a small share of these catchments, so paying farmers to keep the ground covered does little here. That comes from this set of sites, which are cooling and industrial catchments. It says nothing about regenerative farming in general.
This part builds the two operating parts of the return, costs avoided and sales kept.
For data centres and nuclear plants, the shortfall is the share of the day's demand that the catchment cannot supply. A day 40% short is a day when 40% of the water is missing. The model applies that share to the site. The site is taken to be short by the same share as the whole catchment. So this share measures how scarce water is in the catchment. It is not a measurement at the site's intake. A site with a stronger legal right to water would be less short than this. A site with a weaker one could be more short. Legal priority cannot be observed site by site, so it is left out and stated as a limit (§7.6).
The site's own demand is the water it consumes, not the water it draws and returns. For a data centre it is its capacity times a measured rate of water use per megawatt. For a nuclear plant it is generation times a consumption rate for its cooling type.
Data centres and nuclear plants turn the day's shortfall into a share of the day's value lost, each with its own curve (§3.3, §3.4). The year's loss is the average of that share over 365 days, times the site's sales or power bill. It is then averaged across the 200 weather years (§2.1). A hydropower station has no demand and no daily loss curve. Its loss is how far a year's generation falls below the station's normal year (§3.5).
The model treats the three types of site differently, because a shortfall costs each of them money in a different way.
| Type | What a shortfall does | What it costs | Running costs | Sales |
|---|---|---|---|---|
| Nuclear plant | runs below full power, then shuts down (§3.3) | the share of the day's power it cannot make, at its sales price | none | the whole loss |
| Data centre | cools with more electricity, then cuts back what it runs (§3.4) | first extra power for cooling, then lost sales | the whole loss at the default setting | none at the default setting |
| Hydropower station | generates less than in a normal year (§3.5) | the power it does not make, at a published price | none | the whole loss |
A hydropower station does not consume water. It passes the river through its turbines and gives it back. Its loss is a shortfall in generation, not unmet demand. So the demand approach of the other two types does not apply, and it has no running-cost part at all.
The types cannot share a formula. The nuclear formula counts its whole loss as lost sales. For a data centre, that loss holds only the first stage, the extra power for cooling. So a data centre valued on the nuclear formula would drop its lost sales without any sign, and its cooling cost would be measured against sales, not the power bill. A hydropower station has no demand, so the nuclear formula would show it no loss.
The losses are also shares of different amounts. Nuclear and hydropower losses, and a data centre's second stage, are shares of sales. A data centre's first stage is a share of its power bill.
A nuclear plant needs cooling water. Its permit usually limits how much it may warm the river or lake it uses. So the less water there is, the less power it may produce.
share of the day's output lost = 0 short by less than the margin
= (shortfall − margin) ÷ (1 − margin) from the margin to the shutdown line
= the whole day at the shutdown line or beyond
margin = 40% short; shutdown line = 60% short
after each shutdown day, the plant's restart days are also lost in full;
the count starts again on any later shutdown day
Figure 5 — How a shortfall costs a nuclear plant, day by day.
A nuclear operator confirmed the shape of the curve: no loss up to a margin, then reduced output, then shutdown. The numbers on the curve are our estimate. Once there is a shutdown step, the slope of the derating barely matters. The restart days matter at both plants. A smaller margin raises the loss a little at Almaraz and a great deal at Embalse, whose short days are close to the shutdown line.
Less water does not raise a nuclear plant's running costs. It cuts how much power the plant can make, and power not made in a dry month cannot be made up later.
The size of the programme matters more here than at the other types of site. Because of the shutdown step, a programme that leaves the short days past the shutdown line is worth little, and one that brings them back below it is worth a lot.
At both nuclear sites the simulated weather years are almost all the same year. The modelled loss is therefore one representative year, not a range. The tool shows the nuclear loss only as a share, in one representative year. It never shows it as a count of short or shutdown days. The days a plant takes to restart after each shutdown are a fixed input to the curve above, not a count over the year.
Every nuclear page states two limits.
When water runs short, a data centre first cools with more electricity, which costs more. It cuts back what it runs only when extra power can no longer keep it cool enough.
first stage, each day: extra power for cooling = cooling cap × min(1, shortfall ÷ threshold) of the power bill
second stage, each day: lost sales = sales cap × max(0, (shortfall − threshold) ÷ (1 − threshold))^1.5
of sales
cooling cap = 13%; sales cap = 60%
threshold = 60% short by default
Figure 6 — How a shortfall costs a data centre, day by day.
What operators do supports the two-stage shape. Some switch their cooling from air to water in hot weather, and some sites have powered down part of a hall when cooling failed. No peer-reviewed curve links a data centre's output to water shortage, and that is a gap in the evidence. The two caps and the threshold are modelling choices, not cited figures. Real cut-backs so far were caused by heat or the power grid, not by water. So the second stage is treated as a capped loss on the worst days only.
With every setting at Reference and the climate on the middle scenario, none of the three data centres loses sales at the default threshold. All of their loss is extra power for cooling. Under the better and worse scenarios, Barcelona loses a little: USD 0.07m and USD 0.20m a year. That is why a data centre's loss shows in the running-cost box, and the sales box reads "none".
A data centre's sales and power bill are estimated from published sector figures applied to its real capacity (§4.5).
Most of these stations hold little water back, so changes in flow pass straight into generation. Two draw on reservoirs, the Nera at Terni (Lake Piediluco) and the Mucone at Acri (Lake Cecita). The model has no storage term at any station. Storage enters only through the design flow, which is fitted to each station's published output (§7.6). The Jhelum at Karot holds about two days of flow behind its dam and is treated as run-of-river. A station is financed and valued against a normal year, and the loss is the shortfall against that year.
generation in a simulated year = Σ over days of min(river flow after the environmental flow, design flow) normal year = the median simulated year of today's climate loss in a year = max(0, normal year − that year) ÷ normal year expected loss = the average loss across the simulated years
Figure 7 — How a dry year costs a hydropower station.
The hydropower loss is tested against observed events. The model predicts that a station's output responds less to flow the more of the year it spends at full capacity. Measured data confirm that pattern, within plants and across them. But at stations that run near full capacity, the model's response falls off more slowly than observed. So the model overstates the avoided loss there. The benefit from the water is therefore more likely too high than too low. The exception is the Mucone at Acri. An irrigation scheme draws on the same water in dry periods, which the model leaves out, so its benefit is more likely too low. The model is not adjusted for this, because there are few observed data at the high end and some are distorted by limits on exporting power.
Some things are not captured. The model leaves out losses that speed up as a reservoir empties. So at a station with storage, losses in a deep drought would be too low. Two stations in this set draw on reservoirs. Each simulated year starts fresh, so the model does not show a reservoir drawn down across several dry years. The price is fixed. Where a regulated tariff rises in a dry year, holding it fixed overstates the owner's loss. Where a tariff or contract shields the owner from low flow, the physical loss is unchanged but it falls on the power buyer. The tool values it as if the site's company bears it (Appendix E).
Protection is the share of the physical effect of the work that the site gets in a year. It is the ramp times effectiveness.
loss with the programme, year t = loss doing nothing
− ramp(t) × effectiveness × (loss doing nothing − loss with the work fully delivered)
The benefit is not shared. Each company that co-funds the work keeps all of its own avoided loss. Only the cost is shared (§5.2).
The buyer pays for a reduction in how often and how badly the site runs short, which shows up as a lower yearly loss. A credit pays for that reduction once it is delivered and verified.
This part builds the cheaper-borrowing part of the return and says where each effect shows in the accounts.
The parts of the return show in different places in the accounts. Lost sales reduce revenue. Extra cooling power raises running costs. A credit purchase is recorded as an asset on the balance sheet and written off over time. A single "loss" number that mixes these together cannot show a tax effect or a financing effect correctly.
So the model builds an income statement, a balance sheet and a cash flow statement for every year and every future. It works out company value from free cash flow. Only a set of statements that must balance records each effect exactly once, and shows that none is counted twice or left out. The balance sheet balances in every year of every future (§7.4).
| Future | Water | How the programme is paid for and booked |
|---|---|---|
| Doing nothing | the catchment as it is. The site bears the shortfall | none |
| Funding it yourself | the programme's water, with effectiveness applied | paid as costs fall due and treated as an expense, with MRV, verification and coordination |
| Buying credits | the same water, the same protection | credit purchases, recorded as an asset and written off over 10 years |
The accounting rules are the same in every run and cannot be changed. Any difference between the two routes that pay for the work comes from how much cash is paid, when tax relief arrives, or one of the named differences in §1.4. A test confirms there is no other source of difference (§7.4).
Interest depends on debt, debt depends on cash, and cash depends on interest. The model breaks this loop by working out each year's interest on the previous year's debt:
interest(t) = cost of debt × net debt(t − 1)
So the model has no circular references and needs no repeated calculation. The model company pays no dividend, so all its cash goes to reducing its debt.
The model company is the simplified company the method values. Each type of site has one, built from the usual figures for its sector, not from any company's accounts. It is scaled to the site. Its fixed assets are the site's capacity times a cost per megawatt, and everything else is a ratio. So the company grows with the site, and only its ratios depend on its type.
The figures for each type are in Appendix B.
Tax is 25% and long-run growth after 2060 is 2% for all three.
Where the financial figures come from differs by type.
In every case the cost and effect of the work are best estimates from the literature review. The shares and rankings in the results, and whether each result is positive or negative, are more reliable than the amounts.
The model company's cash flows run to 2060, with a value for the years after, whatever a plant's licence says. Almaraz's licence runs to 8 June 2030. Whether it will run longer is not known. If the plant closed then, most of its value created would go, because most of it comes after 2030. The tool states this limit on the plant's pages (Appendix E).
Company value in a year is the site's free cash flow from that year on, including that year's own flow, discounted back to that year. It includes a terminal value at 2060. This is the value of the years after 2060, assuming the cash keeps growing at a steady rate for ever:
free cash flow = operating profit × (1 − tax) + credit write-off − change in stock − credit purchase value(2060) = ( FCF + FCF × (1 + g) ÷ (w − g) ) ÷ (1 + w) value(year) = ( FCF(year) + value(year + 1) ) ÷ (1 + w)
Depreciation and the capital spending that keeps the site running are set equal, so they cancel out. The site keeps its assets up but does not add to them. The terminal value uses the final year's cost of capital, when the programme has ended, so it includes no benefit after 2060.
One small effect also applies to the years after 2060, on the buy-credits route only. Each credit purchase is written off over 10 years, so the MRV and verification paid for in the programme's last years, up to 2055, is still being written off in 2060. The value of the years after 2060 is worked out from 2060's cash flow, which includes the tax saving from that write-off. So it counts the saving every year for ever, although the write-off ends in 2064. This adds less than USD 5,000 to value created by buying credits at Santiago, and USD 0.1m at Karot. Every buy-credits figure in this document includes it.
Company value is modelled at the start of 2026. It includes the first year's cash flow, so a payment made in 2026 counts against the value created. The agriculture methodology models value a year later. This methodology follows forestry and models it at the start of the first year. This is because invasive clearing puts all of its set-up cost in the first year. Modelling it later would leave that cost out of value created but keep it in what you commit.
The cost of capital is the rate at which the site's future cash is discounted. It differs by future.
doing nothing: w = base rate
funding it yourself: w(t) = base rate − cut × f(t)
buying credits: w(t) = base rate − ( cut + duration premium + verification premium
+ cost-of-debt benefit × (1 − tax) × net debt at the start
÷ company value doing nothing, start of 2026 ) × f(t)
f(t) = ramp(t) × effectiveness × share of the loss the work removes when fully delivered, in year t
(the product is the share of the loss actually removed)
Cheaper borrowing is the least certain part of the model. The cut and the premiums are assumed. None of them is measured. No published figure measures a lower cost of capital from investing in a supplier's or a site's water resilience. The green-bond evidence points to a premium that is small, and by some measures close to nothing. That is why the setting runs down to none, and why the tool's summary shows value created with and without this part of the return (Appendix E). At the worked-example site, 68% of the value created when funding it yourself comes from it (§6.4). Where the share is above half, the page says so in words. This applies at the three data centres on both routes, and at Karot when funding it yourself. At the nuclear plants and the other hydropower stations, cheaper borrowing is less than a quarter of value created. Without it, every site still creates value by buying credits, and every site except Barcelona still creates value by funding it yourself.
The result without cheaper borrowing is always shown. The cash card leaves it out entirely, and the "From the water" row shows value created without it. No figure in the tool's summary treats the cut or the premiums as measured.
This part builds the cost side of the return.
hectares treated = share treated × the area the work can treat set-up cost = hectares treated × cost per hectare re-treatment = half the set-up cost, in the programme's fifteenth year the company's share of each = cost ÷ (1 + number of co-funders)
| Work | Cost |
|---|---|
| Clearing invasive trees | USD 300 a hectare, and half that again in the programme's fifteenth year |
| Recharge basins | USD 3,000 a hectare |
| Contour banks and gully repair | USD 250 a hectare |
| Paying farmers to keep the ground covered | USD 100 a hectare, every year |
The tool states the programme cost without discounting, before the bill is shared. It shows the set-up cost in the first year and the re-treatment in its own calendar year. The same step gives the programme's total over its life, so the re-treatment is always shown. At the worked-example site, set-up is USD 8.4m in 2026, re-treatment USD 4.2m in 2040, and the life total USD 13m. What the company commits when funding it yourself is USD 3.4m. That is USD 1.7m set-up, USD 0.8m re-treatment, USD 0.5m MRV and verification and USD 0.4m coordination (§5.5). Buying credits, it commits USD 2.2m.
Other kinds of work have other lifetimes. Recharge basins need yearly maintenance. Contour banks and combinations build up over 10 years. Paying farmers is a yearly payment, not a set-up cost. The tool does not show them.
The costs come from the literature review and are applied the same way at every site (Appendix D). The cost of invasive clearing per hectare is the same everywhere. What differs is how many hectares a catchment has. Labour and access costs differ between countries, so the real cost will be lower at some sites and higher at others. The Programme cost setting (30% cheaper to 50% dearer) can be used to show the effect.
One site rarely funds a whole catchment programme alone. Co-funders are other parties that share the cost of the programme, such as other companies, utilities or public bodies. By default there are 4, so the company pays 20% of the cost. This is a judgement, and the number is a slider on the page. The split follows the forestry use case. The cost of the work, and the hectares on which MRV and verification are charged, are divided. The physical benefit, coordination and the reward from banks and investors are not. The same programme is delivered whoever pays for it.
A large catchment needs co-funders. Cost grows with the catchment, and the benefit grows with the site's sales. At a hydropower station the two can be far apart. The largest catchment in the set is the Karot hydropower station's. There the programme treats 801,000 hectares and its set-up alone costs USD 240m, against sales of USD 240m a year. So the number of co-funders decides whether the result is positive or negative. The slider marks where each route changes from losing value to creating it. Karot needs at least 4 co-funders to create value funding it yourself, and at least 3 buying credits. At the default of 4 it creates value on both routes: +USD 4.7m funding it yourself and +USD 16m buying credits (§6.4). The slider's highest setting is 4, so funding it yourself at Karot creates value only at that setting. With 3 co-funders it loses USD 6.7m, and buying credits creates USD 7.5m. At 4 co-funders each of the five funders pays USD 72m for the work over the programme's life. With MRV and verification and coordination, the company commits USD 87m. Of the USD 4.7m created by funding it yourself, USD 4.1m is cheaper borrowing. From the water alone it is USD 0.6m. More co-funders always lower the company's cost, but more catchment work does not always improve the answer (§2.2).
The company pays at these times.
The programme lasts 30 years.
When the company buys credits, the developer pays for the work and is repaid as it delivers. Where payment waits for delivery, the developer borrows to pay for the work before the buyer pays for it, and has to repay that borrowing. The cost of that borrowing is the developer's financing charge, and it is included in the credit price:
credit purchase(year) = the company's share of (programme spending + MRV)(year − H) × (1 + developer's rate)^H
× payment factor (§5.4)
H = the delay, in years, between paying for the work and its delivery
For invasive clearing there is no delay (H is zero). The work acts from the first year, so there is no wait and no financing charge. Credits for invasive clearing are paid in the same years as the work. They are priced on the share of the work delivered, plus a charge for the risk on the rest. Kinds of work that build up over several years have a delay, and the financing charge then applies.
The developer's rate is 7%. It is a judgement with no published source. It does not change the result for invasive clearing, because there is no delay.
Credit payments rise and fall with delivery. When the company funds it itself, the spending is committed whatever is delivered. A credit buyer does not pay for credits that never arrive. The payment factor is the share of the work a credit buyer pays for:
payment factor = effectiveness × (1 + (1 − effectiveness) × 50%)
At the Reference effectiveness of 60%, the buyer pays for the 60% of the work that is delivered, plus a charge for the developer's risk on the 40% that fails. The charge raises the price by half of that 40%, so by 20%. The factor is 60% × (1 + 40% × 50%) = 72%.
Effectiveness works on both routes. It scales the physical benefit for everyone (§3.6), and it scales the credit price. A less effective programme lowers the benefit on both routes. It also lowers what a credit buyer pays, so buying credits gains ground as effectiveness falls. At Karot, at 35% effectiveness, buying credits adds USD 24m to the value created by funding it yourself, against USD 12m at Reference.
The way the model is built understates the benefit of passing risk to the developer. A single central set of results captures only the payment the buyer avoids on average. Passing on the risk also spares the buyer the ups and downs around that average, and a calculation on the central set of results cannot show that. So the figure reported is a minimum.
MRV and verification is monitoring, reporting and verifying the work. It costs USD 3 a hectare a year, charged on the company's own hectares (hectares treated ÷ (1 + co-funders)), every year the programme runs. The rate is the agriculture use case's. At the default rate, MRV and verification cost about a fifth of the clearing work itself, close to the forestry use case's share. MRV and verification are paid on both routes. If the company funds the work itself, it pays for them each year. If it buys credits, the developer pays for them and recovers the cost in the credit price. That part of the price is scaled by the payment factor like the rest of it, so at Reference a credit buyer pays 72% of the cost, USD 0.36m against USD 0.50m at Santiago (§5.3, §5.4). The difference is part of risk transfer, so MRV and verification are not a separate difference between the routes. That part of the price is written off over 10 years with the rest of the purchase, so the MRV and verification paid for up to 2055 is still being written off in 2060. The value of the years after 2060 counts that tax saving every year for ever, although the write-off ends in 2064 (§4.6). A programme that treats no hectares has no MRV and verification cost and creates no value (§7.4).
Coordination is the one-off cost of organising the co-funders. It is USD 0.4m, paid in 2026. Only funding it yourself pays it, and it is not shared. It is the same in every use case. USD 0.4m is a low estimate. Organising several companies to fund one programme needs contracts, governance and reporting, so the real cost is likely to be much higher, and part of it recurs each year. A higher cost would favour buying credits, because only the fund-it-yourself route pays it. At small sites it is a large part of the cost of funding it yourself: USD 0.4m of the USD 0.56m committed at Barcelona, and of the USD 0.9m at Mucone at Acri. There, a higher coordination cost would make funding it yourself worse still.
Each site's page reads left to right in four lanes: the catchment and the programme; what happens to the water; what it does to the site; what the site is worth. Each box shows both futures side by side, without and with the programme. Clicking a box opens its step:
| Step | Box | Group |
|---|---|---|
| 0 | How the catchment is managed | |
| 1 | Future climate | Doing nothing |
| 2 | Water in the catchment (nuclear: water from the catchment; hydropower: water reaching the turbines) | Doing nothing |
| 3 | Water the site goes without (nuclear: output the plant loses; hydropower: generation shortfall) | Doing nothing |
| 4 | Impact on running costs | Doing nothing |
| 5 | Impact on revenues | Doing nothing |
| 6 | Profit | Doing nothing |
| 7 | What the site is worth, doing nothing | Doing nothing |
| 8 | The nature programme | The programme |
| 9 | What the site is worth, with the programme | What you gain |
| 10 | Buy credits, or fund it yourself? | How you pay |
Each step opens with a one-line answer and three points, then a chart, the settings that move it, a detail section that opens on a click, and a link to the section of this document that explains it. Cheaper borrowing has no box of its own, as in the forestry use case. It appears in steps 9 and 10 and in the tool's summary, always marked as judgement.
Choices are decisions the buyer makes: the site and the number of co-funders, set on a slider (§5.2). The catchment work is fixed. It is clearing invasive trees on 30% of the catchment, for 30 years. The user does not choose between buying credits and funding it yourself. The tool always calculates both.
Scenario settings are conditions the buyer does not control. There are six (§6.2). Five are set at Favourable, Reference or Adverse. Climate offers three named scenarios instead. The tool opens with every setting at Reference and the climate on the middle scenario.
Each setting can be opened to show the inputs it controls. Changing one of those inputs by hand marks the setting Custom, so a setting never shows a level that its inputs no longer match. More inputs appear beneath the settings: the developer's rate, how many years credits are written off over, how much of the risk of failed work is added to the credit price, the MRV and verification rate, the coordination cost, the most the power bill can rise, and the most output a data centre can lose.
Each scenario setting has three levels. Favourable is the easier condition for the site and Adverse the harder. An easier condition can make the programme worth less, because there is less loss to avoid, as the How dry row in §6.3 shows. Reference is the level the others are compared with. The climate has no order (§2.3), so it is not in the table. Three settings describe the world and three describe the programme.
The climate setting belongs to the world group and chooses the Aqueduct scenario. It offers three named scenarios, better, middle and worse, with Aqueduct's names in small print. They are not ranked. The tool opens on the middle one.
| Setting | Group | What it moves | Favourable | Reference | Adverse |
|---|---|---|---|---|---|
| Risk to the price the site is paid | world | sales, every year | prices 20% higher | prices as today | prices 20% lower |
| How dry the site can run | world | the data-centre threshold (§3.4) | 70% short | 60% short | 50% short |
| Programme cost | programme | cost per hectare | 30% cheaper | as costed | 50% dearer |
| Programme effectiveness | programme | share of the work that succeeds (§3.6) | 85% | 60% | 35% |
| Reward from banks and investors | programme | the cut and the three credit premiums (§4.6) | full | half | none |
Each row moves one setting from Favourable to Adverse with every other setting at Reference, at the worked-example site. The climate row moves from the better scenario to the worse one, with the other settings at Reference. This does not mean the scenarios are in order (§2.3). Value created is shown for each route.
| Setting | Fund it yourself, Favourable → Adverse | Buy credits, Favourable → Adverse |
|---|---|---|
| All Reference | +USD 11m | +USD 14m |
| Climate: the better scenario → the worse scenario (not an ordering) | +USD 12m → +USD 11m | +USD 15m → +USD 14m |
| Risk to the price the site is paid | +USD 13m → +USD 9.2m | +USD 16m → +USD 11m |
| How dry the site can run | +USD 10m → +USD 13m | +USD 13m → +USD 16m |
| Programme cost | +USD 12m → +USD 10m | +USD 14m → +USD 13m |
| Programme effectiveness | +USD 17m → +USD 5.8m | +USD 20m → +USD 7.9m |
| Reward from banks and investors | +USD 19m → +USD 3.5m | +USD 23m → +USD 4.2m |
| All Favourable → all Adverse (the climate stays on the middle scenario) | +USD 32m → +USD 1.5m | +USD 40m → +USD 2.9m |
At Santiago the middle climate scenario gives the lowest value created of the three.
The multiples move differently from the values. Programme cost barely changes the value created, but it changes the multiple a lot, because it changes what you commit, which the value is divided by. Funding it yourself, the multiple falls from 4.4 times at Favourable to 2.2 times at Adverse.
The Santiago data centre, paying to clear invasive trees on 30% of its catchment, every setting at Reference. Every figure comes from the model at these settings.
| Input | Value |
|---|---|
| Site | A data centre of 75 MW of IT capacity in Santiago, Chile, which the model treats as cooled with water |
| Catchment | 931 km², of which 14.0 km² is cropland |
| Climate | The middle scenario |
| Co-funders | 4 other companies share the bill; the company pays 20% |
| Programme effectiveness | 60% |
| Reward from banks and investors | Half |
| MRV and verification | USD 3 a hectare a year |
| Share of the water the site needs that the catchment cannot supply | Doing nothing | With the programme |
|---|---|---|
| Today | 22% | |
| 2050, with the work fully delivered | 38% | 27% |
| Average 2026–2060, with effectiveness applied | 33% | 28% |
| Doing nothing | With the programme | Against | |
|---|---|---|---|
| Extra power for cooling (first stage) | 3.1 | 2.6 | a power bill of 48 |
| Sales lost to cutting back (second stage) | none | none | sales of 600 |
In 2026 the site loses USD 2.0m from water shortages; the programme avoids USD 0.5m of it.
The programme treats 27,930 ha. Set-up costs USD 8.4m in 2026 and re-treatment USD 4.2m in 2040, before the bill is shared.
| Doing nothing | Fund it yourself | Buy credits | |
|---|---|---|---|
| Company value, start of 2026 | 2,187 | 2,198 | 2,201 |
| Value created | +11 | +14 | |
| of which from the water | +3.5 | +4.2 | |
| of which from cheaper borrowing | +7.6 | +9.6 | |
| What you commit, 2026–2060 | 3.4 | 2.2 | |
| Value created, times what you put in | 3.3 times | 6.3 times | |
| Extra cash before tax and timing | +15 | +16 | |
| Extra cash | +11 | +12 | |
| Extra cash, times what you put in | 3.3 times | 5.6 times |
What you commit when buying credits includes MRV and verification, at the payment factor: USD 0.36m of the USD 2.2m (§5.5).
Buying credits adds USD 2.5m to the value created by funding it yourself, and needs 36% less cash. Most of the value created on both routes is the reward assumed from banks and investors, not the water (§4.6). The site passes the rule for choosing sites only because of the clearing coefficient (§7.5).
Every input has one of four labels.
| Label | Meaning |
|---|---|
| Sourced | a real external figure, dataset or published study, listed in §7.3 |
| Derived | computed from other inputs by a stated formula |
| Judgement | a stated choice with no external source |
| Structural | a modelling convention, not a quantity with a right value |
Part 8 uses a fifth, weaker label, Indicative, for figures estimated by reasoning instead of calculation.
The label says where a figure comes from, not how much it matters. Some judgements move the result more than some sourced figures do, and §7.5 says which.
| Input | Label | Basis |
|---|---|---|
| Catchment water balance and climate states | Sourced | published datasets and the hydrology model (§2.1) |
| How groundwater is estimated | Derived | our recharge rule, which departs from the hydrology results we received (§2.1) |
| Effect of invasive clearing, flat per hectare | Judgement, from an unpublished literature review | literature; optimistic on average land |
| Effect of invasive clearing at hydropower stations | Judgement | 20% more runoff from treated land; the timing of the added water is an assumption that probably overstates the gain (§2.2) |
| Nuclear loss curve | Judgement | shape confirmed by an operator; numbers our best estimate |
| Data-centre loss curve | Judgement | shape supported by operator practice; no published curve |
| Hydropower loss | Derived | from simulated flow; validated against observed events; more likely too high than too low |
| Sales and power bill, data centres and nuclear | Derived | sector figures applied to real capacity |
| Sales, hydropower | Sourced | published generation and price |
| Cost of the work | Judgement, from an unpublished literature review | literature review (Appendix D) |
| Effectiveness | Judgement | taken from the agriculture use case |
| Reward from banks and investors | Judgement | assumed, with no direct source |
| Number of co-funders | Judgement | a judgement, shown as a slider |
| MRV and verification rate | Judgement, taken from the agriculture use case | the agriculture use case's rate |
| Coordination, developer's rate | Judgement | no published source |
| Share of risk repriced into the credit | Judgement | no published source; it sets the payment factor (§5.4) |
| Environmental flow at hydropower stations, 10% | Judgement | a stated choice (§3.5) |
| Nuclear restart days, 2 at Almaraz and 3 at Embalse | Judgement | part of the nuclear loss curve; set by the reactor's design (§3.3) |
| Share of a city's water from outside the catchment, data centres | Sourced, weak | each city's published share: 90% Las Vegas, 50% Barcelona, 60% Santiago (§2.1); Santiago's published 70% is reduced for the long drought, a judgement (§7.3) |
| Shortfall shared in proportion | Structural | the same share for every user |
| Model companies | Structural | sector conventions per type |
| Base cost of capital, by type | Judgement | sector conventions for each type of site; no cited source (§4.5) |
The water balance is based on published data. The effect of the work on it is a judgement: from an unpublished literature review at data centres and nuclear plants, and a share of runoff at hydropower stations. The step from water to financial value is weaker for data centres and nuclear plants, whose sales and costs are estimated. Cheaper borrowing is based almost entirely on judgement. That is why both return measures are shown, and why the tool's summary shows value created with and without the reward from banks and investors. Extra cash leaves cheaper borrowing out entirely.
Hydrology and climate
Nuclear loss curve
Loss shapes, all types
Water brought in from outside the catchment (data centres)
Catchment work
Hydropower prices
Cost of capital
Nature outcome (Part 8)
Each rebuild of the tool runs the first six checks below, and all of them pass. The last four belong to the water calculations. They pass.
Ranked by how much each could change the result, based on our analysis.
This table lists what the model leaves out, why, and which way each would move the answer. Things that cannot be observed are left out and listed here. The model does not estimate them.
| Left out | Why | Effect on the result |
|---|---|---|
| A plant's own water storage: reservoir, cooling lake or private buffer | usable volume depends on a minimum operating level that is rarely published | the loss, and so the benefit of the programme, can be too high at plants cooled from a reservoir, which includes both nuclear plants: Almaraz (Arrocampo reservoir) and Embalse (Río Tercero reservoir) |
| Water temperature | not in the hydrology; the model follows volume | hot water causes most real nuclear output losses, so where heat limits output, the loss can be too low; the effect on the benefit is unclear, because the work does not change the water's temperature |
| Legal priority to water | allocation rules in a shortage are not observable site by site | either way, for the loss and the benefit: a stronger right means less shortfall, a weaker one more |
| Each plant's actual permit margin | how much of its permitted warming a plant uses is not published | either way |
| Sustained triggers and unit-by-unit steps | plant operating rules are not observed | either way: the curve works day by day, for the plant as a whole |
| The cost of buying replacement power | what a plant or station would pay to cover lost output depends on its contracts, which are not published | the loss, and so the benefit, can be too low, because nuclear and hydropower losses are lost sales only |
| Water held over from one year to the next in a reservoir | each simulated year starts fresh; storage operating rules are not published | either way: drawdown across dry years is missed, but so is storage that eases a single dry year |
| Groundwater drawn down over several dry years | the 200 simulated years are separate weather years, not a sequence, and the groundwater store starts each one full | the loss in a run of dry years can be too low at data centres and nuclear plants, and so can the benefit |
| When in the year clearing adds water, at hydropower stations | the model puts all the added water on days below the median flow; the published evidence supports the direction, but puts only about 7% of the change in the low flows (§2.2) | the water added on dry days, and so the benefit at hydropower stations, is probably too high |
| Single floods, storms and heatwaves | the hydrology works with the spread of dry and wet years, not single events | the loss can be too low where single events cause outages; the effect on the benefit is unclear, because the work does not prevent floods or heatwaves |
The tool states each of these limits on the screens where it applies. The wording it uses is in Appendix E.
Three other simplifications apply to every result.
Parts 1 to 7 describe a financial method. None of it relies on measuring nature. The model uses catchment water, and the financial value follows from that.
This part covers a separate question, which is what the programme does for nature in the catchment. It is descriptive. No figure in it enters the valuation, and changing it changes no result. The model does not put biodiversity into company value, because there is no evidence of that link here. The site gains only through water.
The tool's nature card shows two kinds of figure, and they have different labels.
The tool reports ecosystem integrity, from the Ecosystem Integrity Index. The index scores a landscape by how far it is from what it could naturally support, across three pillars.
| Pillar | What it measures | Data |
|---|---|---|
| Functional | How much the land grows, against what this climate and soil could support | Copernicus Sentinel-3 productivity against a modelled natural potential, 300 m |
| Structural | How intact and connected the habitat is across the surrounding landscape | Global Human Modification v3, Theobald et al. 2025, over a 10 km neighbourhood, 100 m |
| Compositional | How much of the originally present native community is still there | Biodiversity Intactness Index, Impact Observatory, on the PREDICTS database, 100 m |
The compositional pillar is measured against the community that was originally present, not against a count of what is there now. A species that arrives because the land has been disturbed does not raise it; a native species lost does lower it.
Sources are in §7.3.
The index takes the weakest of the three pillars, not their average, and then lowers the result further where the other two are also poor. A landscape that is productive and well connected but has lost its native community is not a healthy one, and an average would hide that.
index = weakest pillar × ( a + b − a × b ) a, b = the other two pillars, each from 0 to 1
The bracket is close to 1 when either of the other two pillars is close to intact, and it falls as both fall. Condition and change are worked out from this index (§8.4).
As a result, a programme that lifts a pillar which is not the weakest moves the score less than it moves the land.
Clearing invasive trees replaces a non-native stand with the community that belongs along the watercourses, so it works mainly on the compositional pillar. Where composition is the weakest pillar, the clearing improves it, and the score moves more. Where structure is the weakest, the clearing improves the land but the score moves less, because settlement, roads and farmland around the watercourses stay as they are.
| Site | Condition today | Weakest pillar | Change with the programme |
|---|---|---|---|
| Data centre, Barcelona | Heavily modified | Structural | Slight |
| Data centre, Las Vegas | Heavily modified | Structural | Slight |
| Data centre, Santiago | Heavily modified | Structural | Slight |
| Nuclear, Almaraz | Partly intact | Structural | Modest |
| Nuclear, Embalse | Substantially modified | Compositional | Modest |
| Hydropower, Nera at Terni | Partly intact | Structural | Modest |
| Hydropower, Jhelum at Karot | Substantially modified | Compositional | Moderate |
| Hydropower, Rímac at Surco | Substantially modified | Compositional | Modest |
| Hydropower, Mucone at Acri | Largely intact | Structural | Modest |
Mucone at Acri starts closest to its natural potential, so the same small gain closes a larger share of the gap than at the other sites where structure is weakest (§8.4). The change is still modest.
The three data centres are in city catchments, where structure is the weakest pillar and the change is slight. The rule also shows where further gains would come from. Where structure is the weakest pillar, the next gain would come from joining up cleared strips along the watercourses. Where composition is the weakest pillar, the next gain would come from native vegetation planted or allowed to return along the banks after the clearing.
The tool reports condition and change in words, not numbers, because the values behind them are reasoned rather than computed (§8.5). Each word matches a fixed range of values, so each means the same thing on every landscape.
Change is measured as a share of the remaining gap, not as a percentage of the index value. A percentage of the value would reward degradation. The same action on a degraded landscape would score several times higher than on an intact one, because it starts lower. A share of the gap does not depend on where a landscape starts.
The calculation runs in two parts. Our water calculations turn the hydrology results we received into the share of each year's sales or power bill that a site loses from water shortages, for every site, climate state and kind of work (steps A1 to A23). They run once, and their results are stored in the tool's data file. The model turns those shares into three sets of financial statements, company value and the figures in the tool's summary (steps 0 to 50). Each rebuild runs the model for every combination of the settings and stores the results in one file per site. The tool reads those files and works out no figure itself.
The code uses different names for some terms. In it, S1 is doing nothing, S2a is funding it yourself and S2b is buying credits. Delivery rate is effectiveness, capital recognition is the reward from banks and investors, invest value is value created and cash out is what you commit.
Symbols: y is the calendar year, 2026 to 2060. k is the programme year, k = y − 2025, so 2026 is k = 1 and 2055 is k = 30. Amounts are in USD m. Where a step differs between the two routes, it says so.
A1. delivery for each site, from the hydrology results: 200 simulated years × 365 days of river flow,
the land's water yield and rainfall; the catchment's daily demand, in total and for
farms; the river abstraction rate and the site's own daily demand
A2. site facts catchment area; the land's yearly water yield (InVEST, mm); depth to the water table
(cm); the Aqueduct basin and today's Aqueduct variability
A3. climate states for each future f and year h (2030, 2050, 2080), from Aqueduct 4.0 for the site's basin:
2020 level = ba(middle, 2030) × √(ba(middle, 2030) ÷ ba(middle, 2050))
supply multiplier = ba(f, h) ÷ 2020 level
demand multiplier = ww(f, h) ÷ its own 2020 level, built the same way
variability ratio = iv(f, h) ÷ today's variability
today: all three are 1
A4. year scale scale(s) = flow plus water yield over year s ÷ its average over the 200 years
A5. climate on flow m(s) = supply multiplier × max(0, 1 + (scale(s) − 1) × variability ratio) ÷ scale(s)
A6. supply river(d, s) = (flow + water yield)(d, s) × m(s) × abstraction rate
A7. demand demand(d) = max(0, catchment demand(d) × demand multiplier
− farm demand(d) × demand multiplier × farm cut
+ the site's own demand)
A8. the work treated ha = catchment area (km²) × 100 × share treated (5, 15 or 30%)
clearing invasive trees: added(d) = treated ha × 1,300 m³ ÷ 365, to supply, every day
recharge basins: 247.5 m³ a ha a year, contour banks: 200, to the groundwater inflow
paying farmers (irrigated cropland only): farm cut = 10% × share treated, and 150 m³
a ha a year to the groundwater inflow
A9. groundwater G = 0.4 × water yield (m) × catchment area (m²) × max(0.3, 1 − water-table depth ÷ 300 cm)
a store of G × 90 ÷ 365, full on the first day of each simulated year; it refills by
(G + added recharge) ÷ 365 a day, up to its size
A10. transfer data centres: T = the city's share supplied from outside × (catchment demand + the
site's own demand), a year; a store of T, full on the first day, refilled by T ÷ 365 a
day and drawn by at most T ÷ 365 a day. Shares: Las Vegas 0.90, Santiago 0.60,
Barcelona 0.50. Nuclear plants: none
A11. shortfall gap = max(0, demand − river − added); draw from groundwater, then from transfer;
shortfall(d, s) = what is left; z(d, s) = shortfall ÷ demand (0 when demand is 0)
A12. nuclear loss L(z) = 0 below 0.40; (z − 0.40) ÷ 0.60 from 0.40 up to 0.60; 1 at 0.60 or more
each day at 0.60 or more is followed by R days lost in full: R = 2 at Almaraz (a
pressurised-water reactor), 3 at Embalse (a CANDU); a new such day starts the count again
Gmean = average over the 200 years of (Σ over the days of L) ÷ 365
A13. data-centre for each threshold t = 0.40, 0.50, 0.60, 0.70, 0.80:
curves eK(t) = average over the 200 years of (Σ over the days of min(1, z ÷ t)) ÷ 365
rK(t) = average over the 200 years of (Σ over the days with z above t
of ((z − t) ÷ (1 − t))^1.5) ÷ 365
A14. step figures share unmet = median over the 200 years of Σ shortfall ÷ Σ demand
short days = median over the 200 years of the days with any shortfall
ba is the water available, ww the water withdrawn and iv the year-to-year variability, as Aqueduct reports them. The 2020 level is worked back from the middle scenario's 2030 and 2050 values, so only Aqueduct's relative change is used (§2.3). z is the share of the catchment's demand left unmet that day. The site is taken to be short by the same share (§3.1). Losses use the average over the 200 years. The share unmet and the short days use the median.
A15. flow q(d, s) = delivered flow(d, s) × published mean flow at the intake ÷ the delivered mean
(Nera at Terni 63.81, Jhelum at Karot 819, Rímac at Surco 12.13, Mucone at Acri 4.08 m³/s)
A16. climate on flow scale(s) = mean flow of year s ÷ mean over the 200 years
sp(s) = max(0, 1 + (scale(s) − 1) × variability ratio)
m(s) = supply multiplier × (sp(s) ÷ average of sp) ÷ scale(s)
here the variability ratio = iv(f, h) ÷ its own 2020 level
A17. design flow Qd = the flow at which average generation today, with no work, equals the published
capacity factor, GWh ÷ (MW × 8.76); found by halving
A18. generation g(s) = Σ over the days of min(q(d, s) × m(s) × (1 − 0.10) + added(d, s), Qd) ÷ (365 × Qd)
0.10 is the environmental flow
A19. the work treated ha = published catchment area (km²) × 100 × share treated (5, 15 or 30%)
runoff (m³ a ha a year) = mean flow at the intake (m³ a day) × 365 ÷ catchment area (ha)
added a year = treated ha × runoff × 0.20 × supply multiplier
added(d, s) = added a year, spread evenly over the days of year s whose flow is below
today's median daily flow
A20. normal year g_normal = median over the 200 years of g(s), today, with no work; the same in every state
A21. loss lossFrac = average over the 200 years of max(0, g_normal − g(s)) ÷ g_normal
doing nothing: added = 0; with the work: added as in A19
A22. data file for each site, for today and for 2030, 2050 and 2080 in each future, and for each kind
of work: Gmean; eK and rK at the five thresholds; lossFrac; treated ha;
programme cost = treated ha × USD 300 ÷ 1,000,000 (invasive clearing: the set-up of the
whole programme, before sharing); share unmet; short days
for each site: sales; power bill (data centres: energy use × tariff); capacity; cost per MW
sales: data centres and nuclear plants, estimated from sector figures on the real
capacity; hydropower stations, published GWh × published price ÷ 1,000
A23. sites the rule of §2.4 selects ten sites; nine are shown
The data file rounds the programme cost to USD 1,000 and the hectares to whole hectares. So the cost a hectare that the model reads back is not always exactly USD 300. It is USD 299.73 to USD 300.03 across the nine sites.
A run is one site at one combination of the settings. The model first sets the run's constants.
0. settings six scenario settings at Favourable / Reference / Adverse (§6.2):
climate the better / middle / worse scenario
risk to the price the site is paid price level 1.2 / 1.0 / 0.8
how dry the site can run threshold 0.70 / 0.60 / 0.50 (data centres)
programme cost cost multiplier 0.7 / 1.0 / 1.5
effectiveness e = 0.85 / 0.60 / 0.35
reward from banks and investors scale 1 / 0.5 / 0
and co-funders N (the slider runs from 0 to 4; default 4)
1. sales S = sales (A22) × price level
2. power bill P = power bill (A22); the price level does not change it
3. your share a = 1 ÷ (1 + N)
4. fixed assets FA = capacity (MW) × cost per MW; depreciation = FA ÷ depreciation life
5. model company base profit = S × profit margin
other cash costs = S × (1 − profit margin) − P
net debt at the start = (net debt ÷ profit) × base profit
stock = other cash costs × stock days ÷ 365 (the model company: Appendix B)
6. programme cost C = programme cost (A22) × cost multiplier; your set-up = C × a
7. payment factor pf = e × (1 + (1 − e) × 50%)
8. delay H = the years the work takes to deliver (invasive clearing: 0)
financing factor = (1 + developer's rate 7%)^H
the developer's financing charge = financing factor − 1
9. MRV your hectares = treated ha × a; K = USD 3 × your hectares ÷ 1,000,000, a year
10. reward cost-of-capital cut = scale × 0.25 percentage points
duration premium = scale × 0.03; verification premium = scale × 0.01;
cost-of-debt benefit = scale × 0.05 (all in percentage points)
11. loss shares at the four anchors (today, and 2030, 2050 and 2080 of the chosen future), doing
nothing and with the work fully delivered:
nuclear plant sales share = Gmean power share = 0
data centre sales share = 60% × rK(threshold) power share = 13% × eK(threshold)
hydropower station sales share = lossFrac power share = 0
eK and rK at the threshold by straight lines between the five thresholds of A13
12. path each share by straight lines: today (2026) to 2030, 2030 to 2050, then along the
line from 2050 towards 2080 (2060 is a third of the way)
13. spending your share of the programme in programme year k:
set-up your set-up, in k = 1
re-treatment your set-up × 50%, in k = 15
upkeep your set-up × upkeep rate, each year k ≤ 30 (invasive clearing: 0)
spending(k) = set-up + re-treatment + upkeep
The path ties the climate points to calendar years. The 2030 climate point is the year 2030, whatever the programme year. The other kinds of work, which the tool does not show, have other lifetimes. Recharge basins have upkeep of 2.5% a year. Contour banks build up over 10 years, with upkeep of 1%. Paying farmers is a yearly payment that builds up over 6 years.
For each year from 2026 to 2060 the model runs the three futures side by side: doing nothing, funding it yourself and buying credits. They share steps 0 to 13.
14. programme on on = 1 while k ≤ 30 (2026 to 2055), else 0
ramp = on (invasive clearing works from its first year; work that builds up over
n years has ramp = on × min(1, k ÷ n)); doing nothing: ramp = 0
15. protection p = ramp × e
16. loss shares sales lost share(y) = nothing(y) − p × (nothing(y) − delivered(y)); power likewise
(nothing and delivered are the two paths of step 12)
17. share removed removed = 1 − (S × delivered sales share + P × delivered power share)
÷ (S × nothing sales share + P × nothing power share)
kept between 0 and 1; 0 when the loss doing nothing is 0
18. reward factor f = ramp × e × removed doing nothing: f = 0
19. cost of capital doing nothing: w = base rate
fund it yourself: w = base rate − cut × f
buy credits: w = base rate − (cut + duration premium + verification premium
+ cost-of-debt benefit × (1 − 25%) × net debt at the start
÷ company value doing nothing) × f
20. fund it yourself cash(y) = spending(k) + coordination USD 0.4m (k = 1) + K (k ≤ 30)
21. buy credits credit purchase(y) = (spending + MRV)(k − H) × financing factor × pf
MRV(k − H) = K while 1 ≤ k − H ≤ 30; no purchase while k − H is below 1
invasive clearing at Reference (H = 0): (spending(k) + K) × 72%
22. sales lost sales = S × sales lost share; sales = S − lost sales
23. power bill P × (1 + power lost share)
24. profit sales − other cash costs − power bill − (fund it yourself only: its cash, step 20)
25. write-off buy credits: (credit purchases of this year and the nine before) ÷ 10
26. operating profit profit − depreciation − write-off
27. interest (cost of debt − (buy credits only: cost-of-debt benefit × f)) × net debt(y − 1)
28. tax 25% × (operating profit − interest); a year with a loss gets a full tax credit
29. net income operating profit − interest − tax
30. balance sheet credit asset = previous + credit purchase − write-off
net debt = previous − (net income + write-off − change in stock − credit purchase)
equity = previous + net income (no dividend)
check: stock + FA + credit asset − net debt − equity = 0, every year, every future
31. free cash flow FCF = operating profit × (1 − 25%) + write-off − change in stock − credit purchase
Step 24 puts the payments for funding it yourself into profit as an expense, so their tax relief comes at once. The credit purchases of step 21 go on the balance sheet and affect profit through the write-off, over 10 years. The site spends on upkeep exactly its depreciation, so the two drop out of free cash flow. The stock does not change, so its change is 0. Interest moves net income and net debt but not free cash flow, so it does not affect company value. The same physical benefit (steps 14 to 18) applies to both routes.
32. after 2060 V(2060) = (FCF(2060) + FCF(2060) × (1 + 2%) ÷ (w(2060) − 2%)) ÷ (1 + w(2060))
33. backwards V(y) = (FCF(y) + V(y + 1)) ÷ (1 + w(y)), y = 2059 down to 2026
34. company value V(2026) for each future: V0 doing nothing, VD fund it yourself, VC buy credits
35. value created fund it yourself: VD − V0; buy credits: VC − V0
36. from the water value created in the same run with the reward at none (scale 0)
37. cheaper borrowing value created − from the water
38. what you commit fund it yourself: Σ step 20; buy credits: Σ step 21; 2026 to 2060, not discounted
39. extra cash Σ FCF(the route) − Σ FCF(doing nothing), 2026 to 2060, not discounted
40. extra cash before Σ (lost sales doing nothing − lost sales with the programme)
tax and timing + Σ (power bill doing nothing − power bill with the programme) − what you commit
41. tax and timing extra cash − extra cash before tax and timing
42. value created, value created ÷ what you commit
times what you put in
43. extra cash, extra cash ÷ what you commit
times what you put in
44. shown as "N.N times what you put in" when what you commit is at least USD 1m, the amount is
not below zero and N is from 0.1 to 20; otherwise the two amounts (Appendix E)
45. credits line buying credits adds VC − VD; cash saved = what you commit, fund it yourself
− what you commit, buy credits; less cash = cash saved ÷ what you commit, fund it
yourself
46. what doing nothing Σ over 2026 to 2060 of (lost sales + P × power lost share, doing nothing)(y)
costs (step 7) × (1 − 25%) ÷ (1 + base rate)^(y − 2025); the years after 2060 are left out
47. the bridge (step 9) sales bar = from the water − the same with lost sales off
power bar = from the water − the same with the 13% cap at 0 (data centres only)
cost bar = from the water − the same with programme cost, coordination and
MRV at 0, on both routes
other = value created − the four bars (sales, power, cost, cheaper borrowing)
48. the step-10 split gap = VC − VD
banks and investors = gap − the gap with the three credit premiums at 0
accounts = that gap − the same with the write-off over 1 year
cash = what is left
49. fewest co-funders the smallest N from 0 to 10 at which value created is at least USD 5,000
50. MRV break-even fund it yourself only: USD 3 × c0 ÷ (c0 − c1), where c0 is value created with
MRV at 0 and c1 with MRV at USD 3 a hectare a year
Each step from 36 to 50 that says "the same with" is a separate run of steps 0 to 34 with that one input changed. Each year's cash counts at the end of the year, and value is modelled at the start of the year. The value of the years after 2060 uses 2060's cost of capital. The programme ended in 2055, so 2060 has no benefit and no reward. On the buy-credits route, 2060 still has the write-off of the MRV and verification paid for from 2051 to 2055. The value of the years after 2060 is worked out from 2060's cash flow, so it counts that write-off's tax saving every year for ever, although the write-off ends in 2064. This adds less than USD 5,000 to value created at Santiago, and USD 0.1m at Karot (§4.6).
Inputs that are switched off. The model has a step for the cost of buying replacement power. It is set to 0, and §7.6 lists that cost as not modelled. Dividends are 0, so all cash goes to reducing debt. Two optional changes are off. One would stop the year-15 re-treatment once a programme has ended, and changes no result at the Reference settings. The other would take the write-off that continues past 2060 out of the value of the years after 2060 (§4.6).
The model at Santiago with every setting at Reference, to check a rebuild step by step.
| Step | What | Value |
|---|---|---|
| 1, 2 | sales S; power bill P | 600.0000; 48.0270 |
| 4 | fixed assets FA; depreciation | 825.0000; 55.0000 |
| 5 | base profit; other cash costs | 270.0000; 281.9730 |
| 5 | net debt at the start | 1,350.0000 |
| 6 | programme cost C; your set-up | 8.3790; 1.6758 |
| 7 | payment factor pf | 0.7200 |
| 9 | treated ha; your hectares; MRV and verification a year K | 27,930; 5,586; 0.016758 |
| 19 | net debt at the start ÷ company value doing nothing | 0.617219 |
| 21 | credit purchase in 2026; of which MRV and verification | 1.218642; 0.012066 |
| 34 | company value: doing nothing; fund it yourself; buy credits | 2,187.2298; 2,198.4002; 2,200.9476 |
| 35 | value created: fund it yourself; buy credits | 11.1704; 13.7178 |
| 38 | what you commit: fund it yourself; buy credits | 3.4164; 2.1718 |
| 39 | extra cash: fund it yourself; buy credits | 11.2745; 12.2049 |
| 40 | extra cash before tax and timing: fund it yourself; buy credits | 15.0327; 16.2773 |
The model company for each type of site (§4.5). Tax is 25% and long-run growth after 2060 is 2% for all three.
| Nuclear plant | Hydropower station | Data centre | |
|---|---|---|---|
| Fixed assets per MW (USD m) | 7.8 | 2.9 | 11.0 |
| Profit margin (before interest, tax, depreciation) | 55% | 65% | 45% |
| Depreciation life (years) | 60 | 60 | 15 |
| Net debt ÷ profit | 3.0 | 4.5 | 5.0 |
| Cost of debt | 5.0% | 5.0% | 5.5% |
| Cost of capital | 7.0% | 6.5% | 7.5% |
| Fuel stock (days of cash costs) | 120 | 0 | 0 |
| Site | Capacity, MW | Catchment, km² | Hectares treated | Set-up cost | Sales a year | Company value, doing nothing |
|---|---|---|---|---|---|---|
| Data centre, Barcelona | 24 | 50 | 1,488 | USD 0.4m | USD 163m | USD 563m |
| Data centre, Las Vegas | 90 | 716 | 21,477 | USD 6.4m | USD 648m | USD 2.3bn |
| Data centre, Santiago | 75 | 931 | 27,930 | USD 8.4m | USD 600m | USD 2.2bn |
| Nuclear, Almaraz | 1,800 | 630 | 18,885 | USD 5.7m | USD 1.4bn | USD 4.8bn |
| Nuclear, Embalse | 608 | 376 | 11,277 | USD 3.4m | USD 295m | USD 801m |
| Hydropower, Nera at Terni | 210 | 2,338 | 70,140 | USD 21m | USD 55m | USD 204m |
| Hydropower, Jhelum at Karot | 720 | 26,700 | 801,000 | USD 240m | USD 240m | USD 1.2bn |
| Hydropower, Rímac at Surco | 137 | 592 | 17,760 | USD 5.3m | USD 59m | USD 350m |
| Hydropower, Mucone at Acri | 106 | 154 | 4,635 | USD 1.4m | USD 22m | USD 71m |
Labels: capacity and catchment are Sourced, except that a data centre's capacity is the model's record, not a published figure. Hectares treated and set-up cost are Derived (30% of the catchment, at USD 300 a hectare). Sales are Sourced for hydropower stations (published output at a published price: Nera 465 GWh and Mucone 185 GWh at USD 117/MWh, Karot 3,174 GWh at USD 76/MWh, Rímac 874 GWh at USD 68/MWh) and Derived for data centres and nuclear plants (§4.5). Company value is Derived. Almaraz restarts in 2 days after a shutdown and Embalse in 3. Almaraz's licence runs to 8 June 2030 (§4.5).
Every figure in this table comes from the model with every setting at Reference.
| Input | Value | Label |
|---|---|---|
| Share of the catchment treated | 30% | Structural |
| How long the work is kept up | 30 years, 2026 to 2055 | Structural |
| Clearing cost | USD 300 a hectare; re-treatment at half that in the programme's fifteenth year | Judgement, from an unpublished literature review |
| Water added by clearing, data centres and nuclear plants | 1,300 m³ a hectare a year | Judgement, top of the plausible range |
| Water added by clearing, hydropower stations | 20% more runoff from treated land | Judgement |
| Co-funders | 4 others; the company pays 20% | Judgement, user choice |
| Programme effectiveness | 60% at Reference (85% Favourable, 35% Adverse) | Judgement, taken from the agriculture use case |
| Share of risk repriced into the credit | 50% | Judgement |
| Payment factor | 72% at Reference | Derived |
| Developer's rate | 7% | Judgement |
| Credit write-off period | 10 years | Structural |
| MRV and verification | USD 3 a hectare a year, on both routes | Judgement, taken from the agriculture use case |
| Coordination | USD 0.4m, once, fund it yourself only | Judgement |
| Nuclear loss curve | No loss below 40% short; shutdown at 60% short | Judgement; shape confirmed by an operator |
| Data-centre loss curve | Cooling cap 13% of the power bill; sales cap 60%; threshold 60% short at Reference | Judgement |
| Hydropower environmental flow | 10% of the river | Judgement |
| Hydropower normal year | The median simulated year of today's climate | Structural |
| Reward from banks and investors | Half at Reference: 0.125, 0.015, 0.005 and 0.025 percentage points | Judgement |
| Tax, long-run growth | 25%; 2% after 2060 | Structural |
| Horizon and valuation date | 2026 to 2060; value modelled at the start of 2026 | Structural |
The six scenario settings and the value of each at its three levels are in §6.2. Each one multiplies or replaces one input of the model (Appendix A, step 0). The climate setting picks one of Aqueduct's three scenarios. With today's climate, that gives ten water states for each site: today, and 2030, 2050 and 2080 in each scenario, joined by straight lines (§2.1).
A file of target figures is saved and re-checked on every rebuild, which stops if any figure has changed. A rebuild must reproduce the same targets. At each of the nine sites they include:
That is 16 configurations at a data centre and 14 at a nuclear plant or hydropower station, because how dry the site can run moves only data centres: 132 in all. Every configuration starts from the Reference inputs of Appendix B, with MRV and verification inside the credit price.
Each configuration records 15 figures. They are your share of the programme cost; company value doing nothing and on each route; value created on both routes; buying credits less funding it yourself; the loss and the loss avoided in 2026; and what you commit, extra cash and extra cash before tax and timing, on both routes. That is 1,980 numbers. A rebuild must match each within USD 0.01m. The table gives five of them, in USD m, rounded to two decimals. The targets file holds all 15, to six decimals.
| Configuration | Company value, doing nothing | Value created, fund it yourself | Value created, buy credits | What you commit, fund it yourself | What you commit, buy credits |
|---|---|---|---|---|---|
| Data centre, Barcelona (16 configurations) | |||||
| Every setting at Reference | 562.65 | 0.15 | 0.50 | 0.56 | 0.12 |
| Climate: the better scenario | 560.70 | 0.35 | 0.73 | 0.56 | 0.12 |
| Climate: the worse scenario | 559.47 | 0.59 | 0.98 | 0.56 | 0.12 |
| Price: Favourable | 714.10 | 0.21 | 0.58 | 0.56 | 0.12 |
| Price: Adverse | 411.20 | 0.09 | 0.43 | 0.56 | 0.12 |
| How dry the site can run: Favourable | 564.52 | 0.11 | 0.46 | 0.56 | 0.12 |
| How dry the site can run: Adverse | 550.90 | 2.94 | 3.47 | 0.56 | 0.12 |
| Programme cost: Favourable | 562.65 | 0.17 | 0.52 | 0.52 | 0.09 |
| Programme cost: Adverse | 562.65 | 0.11 | 0.47 | 0.63 | 0.16 |
| Effectiveness: Favourable | 562.65 | 0.36 | 0.72 | 0.56 | 0.15 |
| Effectiveness: Adverse | 562.65 | −0.06 | 0.29 | 0.56 | 0.07 |
| Reward from banks and investors: Favourable | 562.65 | 0.37 | 0.78 | 0.56 | 0.12 |
| Reward from banks and investors: Adverse | 562.65 | −0.07 | 0.23 | 0.56 | 0.12 |
| Every setting Favourable | 715.97 | 0.81 | 1.29 | 0.52 | 0.11 |
| Every setting Adverse | 401.38 | 0.75 | 1.09 | 0.63 | 0.11 |
| Check run: reward none, lost sales off | 562.65 | −0.07 | 0.23 | 0.56 | 0.12 |
| Data centre, Las Vegas (16 configurations) | |||||
| Every setting at Reference | 2,313.79 | 6.19 | 8.22 | 2.72 | 1.67 |
| Climate: the better scenario | 2,304.08 | 5.13 | 6.82 | 2.72 | 1.67 |
| Climate: the worse scenario | 2,302.68 | 4.95 | 6.59 | 2.72 | 1.67 |
| Price: Favourable | 2,915.13 | 7.71 | 10.10 | 2.72 | 1.67 |
| Price: Adverse | 1,712.45 | 4.67 | 6.34 | 2.72 | 1.67 |
| How dry the site can run: Favourable | 2,315.56 | 5.94 | 7.97 | 2.72 | 1.67 |
| How dry the site can run: Adverse | 2,311.31 | 6.54 | 8.57 | 2.72 | 1.67 |
| Programme cost: Favourable | 2,313.79 | 6.51 | 8.47 | 2.14 | 1.25 |
| Programme cost: Adverse | 2,313.79 | 5.66 | 7.81 | 3.69 | 2.37 |
| Effectiveness: Favourable | 2,313.79 | 9.39 | 11.80 | 2.72 | 2.12 |
| Effectiveness: Adverse | 2,313.79 | 3.00 | 4.74 | 2.72 | 1.08 |
| Reward from banks and investors: Favourable | 2,313.79 | 12.08 | 15.62 | 2.72 | 1.67 |
| Reward from banks and investors: Adverse | 2,313.79 | 0.33 | 0.86 | 2.72 | 1.67 |
| Every setting Favourable | 2,916.90 | 22.06 | 27.64 | 2.14 | 1.59 |
| Every setting Adverse | 1,709.98 | −0.74 | 0.39 | 3.69 | 1.52 |
| Check run: reward none, lost sales off | 2,313.79 | 0.33 | 0.86 | 2.72 | 1.67 |
| Data centre, Santiago (16 configurations) | |||||
| Every setting at Reference | 2,187.23 | 11.17 | 13.72 | 3.42 | 2.17 |
| Climate: the better scenario | 2,191.44 | 11.92 | 14.66 | 3.42 | 2.17 |
| Climate: the worse scenario | 2,188.28 | 11.34 | 13.93 | 3.42 | 2.17 |
| Price: Favourable | 2,744.02 | 13.11 | 16.11 | 3.42 | 2.17 |
| Price: Adverse | 1,630.44 | 9.23 | 11.32 | 3.42 | 2.17 |
| How dry the site can run: Favourable | 2,191.46 | 10.42 | 12.97 | 3.42 | 2.17 |
| How dry the site can run: Adverse | 2,178.57 | 13.20 | 15.84 | 3.42 | 2.17 |
| Programme cost: Favourable | 2,187.23 | 11.59 | 14.04 | 2.66 | 1.63 |
| Programme cost: Adverse | 2,187.23 | 10.48 | 13.18 | 4.67 | 3.08 |
| Effectiveness: Favourable | 2,187.23 | 16.60 | 19.64 | 3.42 | 2.76 |
| Effectiveness: Adverse | 2,187.23 | 5.75 | 7.92 | 3.42 | 1.40 |
| Reward from banks and investors: Favourable | 2,187.23 | 18.84 | 23.34 | 3.42 | 2.17 |
| Reward from banks and investors: Adverse | 2,187.23 | 3.54 | 4.16 | 3.42 | 2.17 |
| Every setting Favourable | 2,748.26 | 32.41 | 39.55 | 2.66 | 2.07 |
| Every setting Adverse | 1,622.35 | 1.52 | 2.90 | 4.67 | 1.98 |
| Check run: reward none, lost sales off | 2,187.23 | 3.54 | 4.16 | 3.42 | 2.17 |
| Nuclear, Almaraz (14 configurations) | |||||
| Every setting at Reference | 4,816.61 | 526.79 | 539.15 | 2.44 | 1.47 |
| Climate: the better scenario | 4,850.17 | 510.68 | 523.06 | 2.44 | 1.47 |
| Climate: the worse scenario | 4,832.95 | 516.90 | 529.25 | 2.44 | 1.47 |
| Price: Favourable | 6,301.46 | 637.55 | 653.07 | 2.44 | 1.47 |
| Price: Adverse | 3,331.76 | 416.04 | 425.23 | 2.44 | 1.47 |
| Programme cost: Favourable | 4,816.61 | 527.08 | 539.37 | 1.93 | 1.10 |
| Programme cost: Adverse | 4,816.61 | 526.32 | 538.78 | 3.29 | 2.08 |
| Effectiveness: Favourable | 4,816.61 | 749.25 | 766.94 | 2.44 | 1.86 |
| Effectiveness: Adverse | 4,816.61 | 305.76 | 313.22 | 2.44 | 0.95 |
| Reward from banks and investors: Favourable | 4,816.61 | 578.51 | 603.12 | 2.44 | 1.47 |
| Reward from banks and investors: Adverse | 4,816.61 | 475.87 | 476.38 | 2.44 | 1.47 |
| Every setting Favourable | 6,301.46 | 1,005.82 | 1,051.21 | 1.93 | 1.40 |
| Every setting Adverse | 3,331.76 | 220.89 | 221.93 | 3.29 | 1.34 |
| Check run: reward none, lost sales off | 5,771.43 | −1.33 | −0.82 | 2.44 | 1.47 |
| Nuclear, Embalse (14 configurations) | |||||
| Every setting at Reference | 801.12 | 65.05 | 67.55 | 1.62 | 0.88 |
| Climate: the better scenario | 821.93 | 57.85 | 60.38 | 1.62 | 0.88 |
| Climate: the worse scenario | 817.79 | 59.98 | 62.51 | 1.62 | 0.88 |
| Price: Favourable | 1,137.50 | 79.98 | 83.16 | 1.62 | 0.88 |
| Price: Adverse | 464.73 | 50.13 | 51.93 | 1.62 | 0.88 |
| Programme cost: Favourable | 801.12 | 65.22 | 67.68 | 1.31 | 0.66 |
| Programme cost: Adverse | 801.12 | 64.77 | 67.32 | 2.13 | 1.24 |
| Effectiveness: Favourable | 801.12 | 92.87 | 96.19 | 1.62 | 1.11 |
| Effectiveness: Adverse | 801.12 | 37.43 | 39.20 | 1.62 | 0.56 |
| Reward from banks and investors: Favourable | 801.12 | 73.35 | 77.99 | 1.62 | 0.88 |
| Reward from banks and investors: Adverse | 801.12 | 56.88 | 57.30 | 1.62 | 0.88 |
| Every setting Favourable | 1,137.50 | 131.38 | 139.92 | 1.31 | 0.83 |
| Every setting Adverse | 464.73 | 25.78 | 26.51 | 2.13 | 0.80 |
| Check run: reward none, lost sales off | 925.44 | −0.91 | −0.49 | 1.62 | 0.88 |
| Hydropower, Nera at Terni (14 configurations) | |||||
| Every setting at Reference | 203.92 | 14.48 | 15.89 | 7.98 | 5.45 |
| Climate: the better scenario | 212.77 | 14.42 | 15.85 | 7.98 | 5.45 |
| Climate: the worse scenario | 200.54 | 14.63 | 16.04 | 7.98 | 5.45 |
| Price: Favourable | 269.35 | 18.32 | 19.80 | 7.98 | 5.45 |
| Price: Adverse | 138.49 | 10.63 | 11.98 | 7.98 | 5.45 |
| Programme cost: Favourable | 203.92 | 15.55 | 16.72 | 6.08 | 4.09 |
| Programme cost: Adverse | 203.92 | 12.69 | 14.50 | 11.13 | 7.73 |
| Effectiveness: Favourable | 203.92 | 22.32 | 23.01 | 7.98 | 6.92 |
| Effectiveness: Adverse | 203.92 | 6.65 | 9.06 | 7.98 | 3.51 |
| Reward from banks and investors: Favourable | 203.92 | 15.29 | 16.94 | 7.98 | 5.45 |
| Reward from banks and investors: Adverse | 203.92 | 13.67 | 14.85 | 7.98 | 5.45 |
| Every setting Favourable | 269.35 | 30.42 | 31.61 | 6.08 | 5.19 |
| Every setting Adverse | 138.49 | 2.31 | 5.49 | 11.13 | 4.98 |
| Check run: reward none, lost sales off | 307.66 | −4.27 | −3.09 | 7.98 | 5.45 |
| Hydropower, Jhelum at Karot (14 configurations) | |||||
| Every setting at Reference | 1,176.02 | 4.66 | 16.28 | 86.91 | 62.29 |
| Climate: the better scenario | 1,070.73 | 6.91 | 18.26 | 86.91 | 62.29 |
| Climate: the worse scenario | 1,015.47 | 7.62 | 18.87 | 86.91 | 62.29 |
| Price: Favourable | 1,495.72 | 15.06 | 26.94 | 86.91 | 62.29 |
| Price: Adverse | 856.31 | −5.74 | 5.63 | 86.91 | 62.29 |
| Programme cost: Favourable | 1,176.02 | 16.93 | 25.80 | 65.28 | 46.71 |
| Programme cost: Adverse | 1,176.02 | −15.78 | 0.42 | 122.95 | 88.24 |
| Effectiveness: Favourable | 1,176.02 | 25.78 | 28.36 | 86.91 | 79.05 |
| Effectiveness: Adverse | 1,176.02 | −16.41 | 7.32 | 86.91 | 40.12 |
| Reward from banks and investors: Favourable | 1,176.02 | 8.77 | 21.45 | 86.91 | 62.29 |
| Reward from banks and investors: Adverse | 1,176.02 | 0.58 | 11.15 | 86.91 | 62.29 |
| Every setting Favourable | 1,495.72 | 60.33 | 64.97 | 65.28 | 59.29 |
| Every setting Adverse | 856.31 | −44.60 | −11.26 | 122.95 | 56.83 |
| Check run: reward none, lost sales off | 1,474.66 | −45.84 | −35.27 | 86.91 | 62.29 |
| Hydropower, Rímac at Surco (14 configurations) | |||||
| Every setting at Reference | 349.94 | 9.97 | 10.92 | 2.32 | 1.38 |
| Climate: the better scenario | 342.40 | 10.20 | 11.09 | 2.32 | 1.38 |
| Climate: the worse scenario | 342.46 | 10.52 | 11.42 | 2.32 | 1.38 |
| Price: Favourable | 436.01 | 12.30 | 13.36 | 2.32 | 1.38 |
| Price: Adverse | 263.87 | 7.63 | 8.48 | 2.32 | 1.38 |
| Programme cost: Favourable | 349.94 | 10.24 | 11.13 | 1.84 | 1.04 |
| Programme cost: Adverse | 349.94 | 9.51 | 10.57 | 3.12 | 1.96 |
| Effectiveness: Favourable | 349.94 | 14.68 | 15.62 | 2.32 | 1.75 |
| Effectiveness: Adverse | 349.94 | 5.26 | 6.31 | 2.32 | 0.89 |
| Reward from banks and investors: Favourable | 349.94 | 11.87 | 13.28 | 2.32 | 1.38 |
| Reward from banks and investors: Adverse | 349.94 | 8.07 | 8.58 | 2.32 | 1.38 |
| Every setting Favourable | 436.01 | 21.67 | 23.56 | 1.84 | 1.31 |
| Every setting Adverse | 263.87 | 2.63 | 3.64 | 3.12 | 1.26 |
| Check run: reward none, lost sales off | 386.78 | −1.29 | −0.78 | 2.32 | 1.38 |
| Hydropower, Mucone at Acri (14 configurations) | |||||
| Every setting at Reference | 71.01 | 5.58 | 6.00 | 0.90 | 0.36 |
| Climate: the better scenario | 75.76 | 5.48 | 5.91 | 0.90 | 0.36 |
| Climate: the worse scenario | 68.74 | 5.54 | 5.96 | 0.90 | 0.36 |
| Price: Favourable | 97.64 | 6.85 | 7.29 | 0.90 | 0.36 |
| Price: Adverse | 44.38 | 4.31 | 4.71 | 0.90 | 0.36 |
| Programme cost: Favourable | 71.01 | 5.65 | 6.05 | 0.78 | 0.27 |
| Programme cost: Adverse | 71.01 | 5.47 | 5.91 | 1.11 | 0.51 |
| Effectiveness: Favourable | 71.01 | 8.15 | 8.54 | 0.90 | 0.46 |
| Effectiveness: Adverse | 71.01 | 3.03 | 3.48 | 0.90 | 0.23 |
| Reward from banks and investors: Favourable | 71.01 | 5.84 | 6.33 | 0.90 | 0.36 |
| Reward from banks and investors: Adverse | 71.01 | 5.33 | 5.67 | 0.90 | 0.36 |
| Every setting Favourable | 97.64 | 10.52 | 11.08 | 0.78 | 0.34 |
| Every setting Adverse | 44.38 | 2.08 | 2.55 | 1.11 | 0.33 |
| Check run: reward none, lost sales off | 109.11 | −0.55 | −0.20 | 0.90 | 0.36 |
| Rate | Value | Label | Source |
|---|---|---|---|
| Clearing invasive trees | USD 300 a hectare, and half that again in the programme's fifteenth year | Judgement | Unpublished literature review of the effects and costs of catchment work (the project's own) |
| Recharge basins | USD 3,000 a hectare | Judgement | The same review. Not shown in the tool |
| Contour banks and gully repair | USD 250 a hectare | Judgement | The same review. Not shown in the tool |
| Paying farmers to keep the ground covered | USD 100 a hectare, every year | Judgement | The same review. Not shown in the tool |
| MRV and verification | USD 3 a hectare a year | Judgement | The agriculture use case's rate |
| Coordination | USD 0.4m, once | Judgement | No published source |
| Developer's rate | 7% | Judgement | No published source |
The literature review is the project's own and is not published. It is cited as an unpublished literature review (§7.3).
This appendix sets out how the tool presents the figures in this document. None of it changes a result.
The tool's summary shows both routes side by side, with a plus or minus sign on every figure.
| Card | Figure | Definition |
|---|---|---|
| You commit | cash, 2026 to 2060 | Funding it yourself: the company's share of the work, plus MRV, verification and coordination. Buying credits: the credit purchases. Neither is discounted |
| What the site is worth | Doing nothing | company value with no programme, start of 2026 |
| You create | value created, per route, with its plus or minus sign | |
| From the water | value created with cheaper borrowing switched off, by running the model again | |
| Cheaper borrowing | the difference, marked as judgement | |
| times what you put in | value created ÷ what you commit, under the display rule below | |
| Cash | Extra cash before tax and timing | sales kept + power costs avoided − what you commit, summed 2026 to 2060 |
| Extra cash | as defined above | |
| Tax and timing | extra cash − extra cash before tax and timing | |
| times what you put in | extra cash ÷ what you commit | |
| Credits line | Buying credits adds … | value created by credits − value created by funding it yourself, and how much less (or more) cash credits need |
When neither route creates value, the verdict says so. A result is marked marginal when one route creates no value, or when the better route returns under 0.1 times what is put in. Where moving one setting from Reference turns a route's result from positive to negative, or the reverse, the verdict adds a line saying the result depends on that judgement. Where more than half of the value created comes from cheaper borrowing, the page says so in words. The share is worked out by running the model again. The verdict says the comparison is made at 2026, the decision year. For invasive clearing both routes pay in the same years, so the difference between them does not come from paying later. The verdict is therefore not repeated for later start dates.
The tool states these limits on the screens where they apply, in these words (§7.5, §7.6).