Nature Stress Test — Shock Layers

Nature Stress Test — Shock Layers

The Nature Stress Test (NST) projects how each ecosystem component and ecosystem service at a site is expected to change under climate and land-use scenarios, relative to a present-day baseline. A shock is that projected change, expressed as a signed percentage (or percentage points): negative = loss/degradation, positive = recovery.

The current generation — NST V1 — is horizon-resolved: every layer is produced at three time horizons and two scenarios, rather than the single 2050 / SSP1-2.6-vs-SSP5-8.5 snapshot used by the earlier single-horizon layers.

What is a climate raster?

Each shock layer is stored as a climate raster — a georeferenced grid laid over the globe, where every cell holds the projected shock value (a signed %) at that location for one scenario × horizon. In practice each raster is a cloud-optimised GeoTIFF (COG) at the native resolution of its source model (from ~10 km for the climate/CMIP6-derived layers down to finer grids for the land-use and service models), reprojected to a common web-mercator grid. The stress test never re-runs a model at query time: it simply samples the raster at each site's coordinates to read that site's shock, so one layer serves every site consistently. A raster therefore has three coordinates — where (the cell), which scenario, and which horizon — which is why each family below ships one raster per scenario × horizon combination.

Why SSP3 and not SSP5-8.5? IPCC AR6 treats SSP5-8.5 as physically implausible, so NST V1 standardises the pessimistic end on SSP3. The single-horizon legacy layers that only shipped SSP5-8.5 are superseded by this set (see Legacy layers below).

Why 2035, 2050 and 2080? The three horizons bracket a short, medium and long-term view, mirroring the short/medium/long framing used in TCFD- and TNFD-style scenario analysis. 2080 is the long-term anchor: far enough for climate- and land-use-driven ecosystem shifts to become material and for the optimistic and pessimistic pathways to visibly diverge, while staying within the reliable projection window of the underlying CMIP6 / SSP-RCP model runs (which are published to 2100). It also aligns with the late-century horizons used in long-term climate-scenario exercises (e.g. NGFS). 2035 gives a near-term signal and 2050 the mid-century reference most disclosure frameworks anchor on; together they show the trajectory, not just an endpoint. Where a source model does not publish a 2080 band (e.g. NDR for Water purification), the value is extrapolated from its available slices — flagged per layer.

Ecosystem components

See the dedicated LUH2 natural-habitat shock layer page for the Habitats component in detail.

Ecosystem services

NST V1 produces several ecosystem-service shocks directly (Tier-2 "ES-direct" model slices), each as the relative change in service-supply capacity (%) vs the 2015 baseline:

Services for which no direct model slice exists are projected from the ecosystem components they depend on, via ENCORE dependencies.

The four layers do not share one sign convention. Water purification and Pollination are pressure/state layers: they score the problem (river N concentration; the pollination deficit), so the value rises as the service degrades — the same convention as the PM2.5 and water-stress layers. Soil protection and Pest control are service/abundance layers: they score the service itself, so the value falls as it degrades — like LUH2 and MSA. The stress-test engine normalises both families to a single degradation scale before computing loss, so the results are comparable; the distinction only matters when reading a raw layer value or its map.

Horizon provenance differs by service model — this matters when reading 2035 and 2080. For Water purification, IMAGE-GNM publishes 5-yearly bands: 2035 and 2050 are real model slices, and only 2080 is extrapolated (from the 2050 band). The other three service models publish a single future slice, at 2050. Their 2035 and 2080 layers are linearly scaled off the 2015 baseline — the 2050 shock multiplied by (horizon − 2015) / 35 and clamped to range — so they assume a constant rate of degradation and carry no independent model information beyond 2050. Read those two horizons as a trajectory, not a projection; each layer page states its own provenance.

The full horizon-resolved set

Each component/service family below ships 6 rasters = 2 scenarios × 3 horizons (plus the LUH2 2015 baseline extent layer):

The scenario token differs by source model: climate-driven layers (LUH2, SOC, PM2.5) use ssp126 / ssp370; IAM/biophysical-model layers (MSA and the four services) use ssp1 / ssp3. Both name the same optimistic (SSP1-2.6) and pessimistic (high-end SSP3) pathways. The forcing behind the pessimistic member is not identical across models — the climate-driven layers use SSP3-RCP7.0, while the biophysical service models (IMAGE-GNM, GLOBIO-ES, InVEST) publish their regional-rivalry runs at SSP3-RCP6.0. Same storyline, marginally lower forcing; the application labels the scenario SSP3-7.0 throughout.

Interpreting the shocks: non-monotonic and cross-scenario patterns

Because every horizon and every scenario is an independent projection sampled from its own raster, the NST does not enforce that impacts grow with time, nor that the pessimistic scenario is always worse than the optimistic one. Two patterns that look counter-intuitive are in fact expected consequences of the underlying layers.

1. A 2080 impact can be smaller than the 2050 impact. Several drivers are non-monotonic in time — they worsen towards mid-century and then recede. The clearest is the Habitats (LUH2) layer: the projected land-use trajectory can drive a sharp loss of natural-habitat extent around 2050 and then partially reverse later in the century (land abandonment / regrowth in the scenario), so the shock swings from a strong integrity loss at 2050 to near-zero or even a recovery at 2080. Passed through the Andrén/Fahrig Hill function, a mid-century dip below the ≈30 % inflection followed by a rebound above it produces exactly this V-shape. Atmosphere (PM2.5) and Water (water stress) behave similarly in many regions, where concentrations or demand peak mid-century and then decline. Because a site's headline loss is a limiting-factor MAX over its ecosystem services, when the dominant component recovers by 2080 the whole site's projected loss drops.

2. The optimistic scenario (SSP1-2.6) can show a larger impact than the pessimistic one (SSP3). "Optimistic" and "pessimistic" describe the global climate/emissions pathway, not every local ecosystem outcome. The two scenarios are produced by different integrated-assessment models with different land-use and biophysical trajectories, so at a given location one component can be worse under SSP1 than under SSP3 — for example SSP1 land-use choices (bioenergy-crop expansion, afforestation patterns) degrading habitat in that cell, or higher local water demand — even though the climate is milder. As above, because the headline is a MAX over services, a single component flipping this way flips the site's ordering. This is seen most often on the Habitats (LUH2) and Water components.

Legacy layers

The earlier single-horizon shock pages — ecoregion-shock-{ssp1,ssp5}, msa-shock-{ssp1,ssp5}, pm25-shock-{ssp1,ssp5}, soc-shock-{ssp1,ssp5}, water-stress-shock-{ssp1,ssp5} — describe the 2050-only generation and, for Habitats, the EII Ecoregion proxy. They are retained for reference but are superseded by the horizon-resolved NST V1 set documented here; in particular, the live Habitats component is backed by LUH2, not the Ecoregion shock.