Nature Stress Test

Nature Stress Test

Purpose

The Nature Stress Test estimates the financial loss a company or portfolio could face when the ecosystems its sites depend on degrade under forward-looking climate and land-use scenarios. It answers a concrete question: if nature follows an optimistic versus a pessimistic pathway to 2035, 2050 or 2080, how much production — and therefore revenue — is at risk at each site?

It is a scenario-based stress test, in the same family as the ECB, NGFS and Banque de France / ACPR nature exercises: it reports a loss estimate under each named scenario, not a probability-weighted Value at Risk (VaR) or Average Annual Loss. A true Nature VaR would require a probability distribution over loss outcomes, which nature risk does not yet support in settled practice — see Scope & limitations.

At a glance

How it works

The assessment chains four steps, from a spatial ecosystem shock at a location to a financial loss at a site:

1. Ecosystem-component shocks. For each of the five ecosystem components — Atmosphere, Structural & Biotic Integrity (Habitats), Soils & sediments, Species, Water — a spatial shock layer gives the projected change in the component's state at the site, per scenario and horizon, relative to a present-day baseline. Where the ecological literature supports it, the change is passed through a non-linear threshold function so that a state crossing a known tipping point produces a disproportionate shock. Full sources, proxies and threshold forms are on the Shock Layers page.

2. Projection onto ecosystem services. Each ecosystem service is supplied by one or more components. Using the ENCORE delivery matrix (which component supplies which service, and how strongly), the component shocks are projected into a service shock S_e:

S_e = AGG_component[ delivery(component, e) × shock(component) ]

A subset of services is instead modelled directly from its own supply layer and bypasses this projection — see Direct vs component-projected services below.

3. Site dependency → production shock. How much a service shock actually hurts a site depends on how much the site's activity depends on that service. The dependency is derived from the site's Scope-1 activity (product or monetary intensity, materiality-weighted per ENCORE) or, absent activity data, from its sector and site typology:

production_shock_e = S_e × dependency(site, e)

4. Financial loss. The production shock is applied to the site's exposure (its financial base — revenue by default, with other bases selectable as labels):

loss_e = production_shock_e × exposure       SiteLoss = AGG_service[ loss_e ]

Dependency is applied exactly once, and both the direct and projected service paths feed the same financial step.

Direct vs component-projected services

A service shock S_e can be produced two ways, and both meet at the same dependency × exposure step, so the two kinds of service are directly comparable in the results.

• Component-projected (default). Most services have no global supply model, so their shock is derived from the ecosystem components that supply them, through the ENCORE delivery matrix (step 2 above). The driving component — the one that most limits the service — is surfaced in the drill-down.
• Directly modelled. A small set of services is scored from its own supply layer (relative change in supply capacity vs the baseline), bypassing the component projection. Four services are covered to date:

How the product tells them apart:

• A directly-modelled service carries a "direct" badge in the per-site breakdown and appears as its own selectable layer on the map.
• Because a direct service has no supplying component, it cannot be attributed to one of the five components in the component view; it is collected in a dedicated "directly-scored services" bucket, so the component balance still reconciles to the site total (Σ 5 components + direct bucket = total).
• In the drill-down, a projected service opens its component → service decomposition (biome split + driving component); a direct service shows its own supply shock, with no component projection.

Aggregation modes

Two computation toggles set the figures, and one visualization toggle only regroups the same figures without changing them.

• Component → Service (Max / Sum): per service, Max reads the single limiting component; Sum (capped at 100 %) adds the components' contributions.
• Service → Site (Max / Sum): per site, Max is the stress-test reading — the single worst service drives the site loss (a limiting-factor / value-at-risk framing); Sum adds every service's loss.
• Group by service / component (visualization only): switching the view leaves the totals invariant — total loss, per-site loss and per-service loss are identical either way.

In the component view the loss is shown as an additive balance (Σ components = site total, in every mode): there is no single "max component", because the worst service is chosen at the service level and, across a portfolio, different sites' worst services can fall on different components. Services scored directly (no supplying component) form a dedicated "directly-scored services" bucket so the component balance still reconciles to the total.

Ecosystem components & non-linearity

Full derivations, calibration points and data sources are on the Shock Layers page. Non-linearity is applied to the ecosystem state, not to the production function (elasticity of substitution between services is not yet supported by the literature).

Location awareness (biome dimension)

Delivery is not uniform across the planet: a component supplies a service differently in a tropical forest than in an agricultural or urban landscape. The ENCORE 2024 matrix adds a biome dimension (Ecosystem service × Component × Biome), resolved at each site through the IUCN biome raster using a fixed 10 km buffer and a 10 % covered-area threshold (a biome counts only if it covers at least 10 % of the buffer). This reintroduces location-aware sensitivity into the delivery step.

Scenarios & horizons

The product labels two narrative pathways Optimistic (OPT) and Pessimistic (PES) throughout. OPT pairs an SSP1 narrative with a low-emission pathway — SSP1-RCP2.6, a transition to a sustainable economy; PES pairs a fragmented, high-emission narrative with a high-end pathway — SSP3-RCP7.0 (the pessimistic end was moved off SSP5-RCP8.5, which the IPCC AR6 treats as physically implausible). Each is projected at three horizons — 2035, 2050 and 2080 (short / medium / long-term; 2080 is the long-term anchor — see why these horizons on the Shock Layers page) — with interpolation between a source's available data points where needed. Because every horizon and scenario is an independent projection, impacts are not forced to grow with time, nor is PES always locally worse than OPT — see the Shock Layers page.

What the product shows

The results follow the pipeline above and are fully drillable (an explicit anti-black-box design):

• Headline banner — the average loss across scenarios and horizons, its share of the site's exposure, a min–max range, and a chart of the optimistic, pessimistic and central trajectories across horizons.
• Summary — the contribution rules in force, the top-5 sites and top-5 ecosystem services by average loss across horizons and scenarios (scoped to the selected entity perimeter).
• Explorer — a scenario selector (optimistic / pessimistic / both) and horizon selector, a service-view / component-view toggle, and a results / map toggle.
• Results — the financial loss (€) and production shock (%) per service or component, a site ranking, a lollipop chart of the non-zero categories, and a per-site breakdown table (dependency, component shock, delivery × shock, service shock, loss). In Max mode the loss-driving service is flagged with a "max" badge per scenario (the worst service under the optimistic and the pessimistic pathway need not be the same).
• Map — the five component shock layers and the directly-modelled service layers, site markers (colour = shock intensity, size = share of revenue, dashed grey = no production data), a biome context underlay, and a per-site popup.
• Drill-down — any loss decomposes into component shock → delivery → service shock → dependency → production shock → financial loss, with the biome split and the driving component surfaced.
• Range output — the site loss is a bracket: min = worst single service (MAX), max = Σ services, rather than a single point.

Scope & limitations

• Hybrid pipeline. Four services are directly modelled today; the remainder of the ~25 ENCORE services stay on the component-projection path. Roughly ten are sourceable as direct layers as the set grows — the hybrid design is deliberate.
• Sites only. The shock is assessed at the site; value-chain (upstream/downstream) diffusion is not yet modelled — the current estimate is an assumed lower bound.
• Adaptation & mitigation. V1 offers qualitative adaptation guidance and manual user adjustment of sensitivities; automated quantification of an action's benefit is out of scope.
• Positive shocks. An ecosystem improvement under a scenario is, by default, capped at zero (a conservative, no-benefit reading); allowing a positive contribution is a per-service option.
• Thresholds. Absolute tipping-point thresholds rely on ecosystem-specific literature that is fragmented by ecosystem type; calibrations are documented and under expert review.
• Not a probabilistic VaR. No probability is attached to the scenarios, so no distribution or quantile is produced; a probability-weighted Nature VaR is a roadmap item.

Methodological grounding

The approach builds on the Ethifinance nature stress-test / value-at-risk methodology and is aligned with regulator-grade scenario exercises — ECB research on nature-related financial losses, the NGFS nature pilot, and TNFD scenario-analysis guidance. Data foundations are ENCORE 2024 (service taxonomy, delivery and dependency ratings, biome dimension), the IUCN biome raster, and the per-component and per-service projection sources listed on the Shock Layers page.