Guide updated 7 October 2026

Climate Hazard Engine / Data and methodology / v1

Physical climate
hazard analysis

The engine samples observed and modelled hazard data for a location. This guide explains the geographic resolution, calculations and limitations of each measure.

Sources, scenarios and years

From a place to a physical measure

The engine matches a location to a point or boundary, samples a named climate dataset, and applies a calculation specific to that measure. It returns the value with its unit, geographic coverage, source, period and any scenario. Prepared readings are reused when available.

Data sources

Flood models: JRC river-flood depth maps and Environment Agency flood zones describe modelled present-day conditions. The London example uses the JRC model without flood defences.
Climate projections: NASA NEX-GDDP-CMIP6 supplies daily temperature and humidity; IPCC AR6 supplies coastal sea-level projections. The scenario and horizon belong to each reading.
Historical records: NOAA IBTrACS provides storm tracks; NASA FIRMS provides satellite thermal detections; Hydro-JULES provides monthly water-balance anomalies.
Coastal and soil data: Copernicus GTSM provides modelled extreme water levels. SoilGrids predicts clay content, and Copernicus SWI records a soil-water indicator.
Place boundaries: GeoNames, GHS-UCDB, ONS and Natural Earth locate places and define their spatial scope. An urban footprint and an administrative area can produce different results. Boundary sources and licences.

How the calculation works

Resolve the location
Identify a point or boundary and its supported resolution. Ambiguous places need review before a reading is attached.
Select the input
Choose the source layer, its observation period or projection scenario, and the cells or recorded points relevant to that place.
Apply the measure’s rule
Count wet flood cells, calculate heat-threshold days, or summarise another source in its own units. Each rule is explained before its example below.
Retain the evidence
Keep the value, unit, boundary, source, period, uncertainty and coverage limits together. Missing inputs remain gaps.

Units and aggregation. These measures have no common 0–100 scale and are not weighted into an overall hazard score. Some calculations divide by area or time; SPEI is already standardised against each cell’s own history. Heat reports the median and spread across available model members. These operations do not turn the measures into comparable financial risks.

Hazard, exposure and risk. Hazard describes physical conditions at a place. A bank’s disclosed activities, assets or lending establish its exposure context. Estimating financial risk would also require asset locations, values and vulnerability; this engine does not calculate losses. Any separate workbench screening bands are provisional Greenophy presentation rules, not the source providers’ classifications or a combined risk score.

Missing is not zero. Unavailable coverage, an unresolved place or an unsupported scenario produces a gap, not a zero. A valid zero has the meaning of its measure: no counted observations in that record, no threshold exceedance, or—in SPEI—a local historical norm. Modelled and derived values remain labelled; an illustrative diagram is never a measured reading.

Place

London, UK. A qualified name identifies the place.

Geometry

London’s urban footprint, from GHS-UCDB.

Source & sampling

Source data are sampled within the boundary. The grid is schematic.

Value + unitGeometry + resolutionSource + scenario + periodCaveat + provenance

Reading

The result retains its geometry, source and limitations.

Ambiguous place? Candidates for review. No hazard readings until resolved.

Real UK and London outlines, simplified for display. The grid is illustrative; no hazard values are plotted. London’s urban footprint differs from its administrative boundary. Boundary sources.

Physical hazard only. No asset value, vulnerability, loss estimate or composite risk score.

01 / Precision

Spatial resolution

A location can be assessed at different spatial resolutions. The comparison below uses the same London point with flood-zone and climate data.

Compare source resolution

Point retained · EA declares 5 m positional resolution.

Real London outline and coordinate. EA flood extents are not plotted. Heat mode shows the native 0.25° grid; the highlighted cell indicates spatial support, not temperature. Map sources.

P0effective tier

EA flood zones declare 5 m positional resolution. This London coordinate retains effective tier P0; the map does not assert a flood-zone result.

Point query · fine-scale source geometry

Why precision depends on the source

The same address can be compared with fine flood-zone outlines and a coarse climate grid. A precise query cannot make its source more detailed. The result reports the scale the data can support.

The maps use the same London location. EA flood zones are vector outlines; no flood extent is plotted here. The climate grid uses native 0.25° cells, whose east–west spacing varies with latitude.

P0 · Point

Central London

Coordinate · 51.50853° N, 0.12574° W

A local value or flag

P1 · City

London

Urban footprint · GHS-UCDB

A spatial distribution

P2 · Region

England

Administrative boundary · ONS

A spatial distribution

P3 · Country

United Kingdom

National boundary · Natural Earth

An inventory of hazards

Each outline is fitted independently, not drawn at a shared scale. England is a P2 boundary within the UK; the P3 UK query returns a hazard inventory, never a national hazard number. Boundary sources.

How we identify a place

A coordinate, postcode or qualified place name establishes the query location. Where a name is ambiguous, such as “Newcastle”, the choice must be resolved before a hazard reading is attached. A similar-looking name is not silently substituted.

Place names and boundaries come from GeoNames, GHS-UCDB, Natural Earth and ONS. Urban footprints and administrative boundaries answer different questions; the geometry used is stated with the result. Map sources and licences.

Source details: EA flood zones · Warming delta

02 / Hazard methods

Hazard measures

The examples below show how each measure is calculated and interpreted.

01 / Flood

Flood extent and depth

Calculation. Sample the JRC flood-depth cells within the selected boundary at a stated return period. Divide the number of inundated cells by the number of boundary cells; calculate median and 90th-percentile depth from the inundated cells only. The share is a cell-count fraction, without weighting by each cell’s ground area.

Meaning. The headline uses the present-day 100-year river-flood model, corresponding to a 1% annual chance in the model. Seven return periods are available. Flood defences, coastal surge and surface-water flooding are excluded. This is a measure of land within the boundary, not a share of collateral or a building’s flood depth.

Worked example: London’s GHS-UCDB urban footprint · present-day JRC river flooding · 100-year return period
Stage	Evidence and result
Source and scope	London urban-centre boundary 1912; native JRC cells of approximately 90 m. Boundary source · JRC tile 106 · JRC tile 113.
Selected cells	347,789 cells within the boundary; 44,140 have finite, non-nodata inundation depths. All of those selected depth values are positive.
Intermediate calculation	44,140 ÷ 347,789 × 100 = 12.6916%. Order the 44,140 wet-cell depths and use their nearest-rank percentiles.
Reported result	12.69% wet-cell share; median wet-cell depth 1.758 m; 90th percentile 4.251 m. These are model outputs, not an observed flood. The map below shows the same source cells.

Modelled inundationUrban footprint

London · GHS urban footprint, not an administrative boundary. Native ~90 m flood cells; blue shows extent, not depth.

A Native cell detail

20 columns × 14 rows
84 wet cells / 280 shown

Counted as wetOther boundary cells

Actual ~90 m source cells, enlarged. The totals on this page use the whole London boundary.

Wet-cell share

12.69%

44,140 wet cells out of
347,789 boundary cells

Share across the whole boundary

Depth in wet cells

Median1.76 m

90th percentile4.25 m

Wet cells only. Dry cells are excluded from these depth statistics.

Actual model output, not an observed flood. JRC CEMS-GLOFAS present-day, undefended river flooding over London’s GHS-UCDB urban footprint. Source read 20 September 2026. The Thames Barrier’s protection is not represented. Data and calculation.

Reading limit Undefended river flooding only. Wet area is not a share of collateral, or a building’s flood depth.

Source details: JRC flood · JRC return periods

Technical detail: London source cells and depth statistics

The source-window statistics use cell centres inside London’s GHS-UCDB urban footprint. The wet-cell depth range is 0.100–6.302 m. Nearest-rank percentiles describe the wet cells only; the boundary share uses all 347,789 selected boundary cells.

The magnified detail shows 280 cells from the same source raster, including 84 counted as wet. It is a display crop; the reported 12.69% uses the whole boundary. The outline is simplified for display, while the sampling uses the source geometry.

JRC source and CC BY 4.0 licence · London boundary source.

Measured example: Udaipur’s non-monotonic pair

Udaipur · two measured points

200-year return period19.74%

500-year return period4.06%

A smaller share at a rarer period. JRC publishes separate model outputs. The engine reports and flags this pair; it does not smooth or reorder it.

Measured JRC model outputs, 17 September 2026. Bars share a 0–25% scale. Only these two Udaipur points are shown; they are not a complete curve.

02 / Heat

Warming and heat thresholds

Calculation. NASA NEX-GDDP-CMIP6 daily maximum temperatures are sampled on the selected grid cells. Warming is the projection mean minus the baseline mean. For threshold days, count valid cell-days strictly above 35 °C, divide by valid cell-days sampled, and multiply by the model’s mean days per year. Report each model’s result, then the ensemble median and spread.

Inputs and meaning. The example uses SSP2-4.5, 2050–2060 against 1991–2010, with five temperature models on a 0.25° grid (nominally 25 km). Wet-bulb days use a 31 °C threshold and temperature, humidity and estimated pressure from four models; the approximation can understate the daily wet-bulb maximum. The diagrams below illustrate the rules, not measured values for a place. The model-selection rule is in the technical detail below.

Warming delta

ΔT · kelvin

Projection mean − baseline mean.

Five models · ensemble median + spread

Dry-bulb exceedance

Days / year

Count only days strictly above 35 °C.

Five models · baseline, projection + change

Wet-bulb exceedance

Days / year

Count only days strictly above 31 °C wet bulb.

Four models · MIROC6 lacks humidity

Illustrative sequences, not place-specific or annual readings. A day exactly at either threshold is excluded.

Reading limit Wet bulb uses daily-mean humidity at maximum temperature and can understate the true daily wet-bulb maximum.

Source details: Warming delta · Threshold days · Wet-bulb days

Ensemble selection and acquired members

Default ensemble filter

TCR 1.4–2.2 K

AR6 likely range

Five acquired members

GFDL-ESM4
MIROC6 no humidity file
MPI-ESM1-2-HR
MRI-ESM2-0
NorESM2-LM

Five for dry bulb; four for wet bulb. The member spread is not a probability distribution.

The default ensemble screens transient climate response (TCR) to the AR6 likely range of 1.4–2.2 K, following Hausfather et al. (2022), as documented in the reference. This addresses the CMIP6 hot-model problem.

The acquired subset contains five named members. A full-ensemble mode and a single-model mode are available and must be named when used.

The five-member median and spread are not a probability distribution. Eligibility under the screening rule does not mean every eligible model is in this acquired subset.

03 / Tropical cyclone

Storm frequency and closest approach

Calculation. From NOAA IBTrACS main-track positions since 1980, retain each storm’s closest recorded position within the declared search radius, normally 100 km. Count each qualifying storm once. Divide that count by the record length in decades to report frequency; report the strongest retained pass separately.

Meaning. Nearby track positions do not establish landfall, local wind or damage. Extratropical stages are included in the count and identified separately; wind sources and averaging periods are retained. The Manila example below follows one storm through the selection step, not the city’s full storm history.

Recorded trackRetained positionSearch boundary

Retained observation · 11 Nov 2020, 18:00 UTC · 64.9 km

01 / Follow the observations

Recorded storm positions

Follow the recorded positions westward towards Manila.

Markers are recorded observations; no positions are interpolated.

02 / Apply the search radius

Observations within 100 km

Two records lie within 100 km: 64.9 km and 77.9 km away.

Both observations belong to Vamco, the same storm.

03 / Retain the closest pass

One qualifying storm

2
track positions

1
storm pass

Two positions belong to Vamco. Count the storm once, at its closest recorded approach.

Closest approach: 64.9 km
Retained timestamp: 11 Nov 2020 · 18:00 UTC
Wind at retained position: 70 knots · WMO

Wind at the storm’s position, not the wind experienced in Manila.

FrequencyCount distinct qualifying storms across the complete declared record, then express passes per decade.

IntensityReport the strongest retained pass separately. A nearby track does not establish landfall or damage.

Real IBTrACS track, GeoNames Manila point and Natural Earth coastline. This is one storm’s worked example, not the city’s full storm history. Only recorded main-track positions are tested; connecting lines do not add observations.

Reading limit Track proximity does not establish landfall, local wind or a damage footprint. Extratropical passes are counted separately.

Source details: IBTrACS

About this storm example

The query is Manila (14.6042° N, 120.9822° E). The observations describe Vamco in November 2020; the closest retained position is 15.1° N, 121.3° E. The diagram uses recorded track positions, not an interpolated local wind footprint. NOAA IBTrACS source.

04 / Wildfire

Vegetation-fire detection density

Calculation. Select NASA FIRMS VIIRS detections within the boundary over a declared window, excluding overlapping records. Keep type 0, presumed vegetation-fire detections, for the headline. Density = retained detections ÷ area in km² ÷ days in the window × 1,000; other thermal detections are counted separately.

Meaning. The England example uses 1 January–31 December 2025 and approximately 375 m satellite detections. These are not distinct fires, burned area or a future probability. Cloud and overpass timing limit detection; a valid zero means no detections in that record, not proof that no fire occurred. Comparisons require matching observation windows.

Type 0 · presumed vegetation fireTypes 1–3 · other thermal detections

3,979 detections shown; repeated locations may overlap.

01 / Select vegetation-fire detections

3,979
all detections

2,017
type 0 retained

02 / Account for area and time

2,017 retained detections130,541.3 km² × 365 days

× 1,000

0.0423

detections per 1,000 km² per day

The density always uses type 0, whichever map view is selected.

Actual locations and counts from the archived FIRMS files, clipped to the ONS England boundary. 1,962 detections of other types are excluded from density. Detections are not distinct fires, burned area or a future probability. Cloud and overpass limits still apply.

Reading limit Detections are not distinct fires or burned area. The 2025 observation window is not a climatology.

Source details: FIRMS

Monthly counts, observation window and source checks

When were detections recorded?

The example covers 1 January–31 December 2025, excluding overlapping records. Counts describe satellite detections, not unique fires; the chart separates presumed vegetation-fire detections from other thermal sources.

England · 2025 · counts of detections
Month	Type 0	Other types	All
Jan	13	111	124
Feb	21	65	86
Mar	263	176	439
Apr	260	206	466
May	229	246	475
Jun	113	224	337
Jul	191	190	381
Aug	670	199	869
Sep	109	265	374
Oct	38	118	156
Nov	102	114	216
Dec	8	48	56

Other thermal detections: type 1, 0; type 2, 1,893; type 3, 69. FIRMS source details · Boundary source.

05 / Drought

Change in local water balance

Calculation. Hydro-JULES supplies monthly SPEI on 0.05° cells over 1981–2022. For an area, take the median across sampled cells each month; for a point, use the selected cell and state any nearest-cell fallback. Calculate the last ten years’ mean minus the first ten years’ mean, excluding unavailable months. A negative change means the recent window was drier.

Meaning. SPEI is a water-balance anomaly already standardised against each cell’s own history. Three- and twelve-month versions describe different durations. Zero means local normal, not a wet climate; the change is not absolute aridity or water stress. The central London example below is one nearby cell, not a London-wide average, and the 1981–2022 reference already includes warming.

Scroll the chart sideways to see the full record →

Below local normalAbove local normalDecades compared

December 2022 · SPEI-12: -0.288

Last decade − first decade

+0.348 SPEI

A shift towards wetter conditions in this cell.

1981–1990 mean: -0.012
2013–2022 mean: +0.335

London query → nearest cell
51.50°N, 0.15°W · 1.93 km away

Zero is local normal.Two places at SPEI 0 can have very different climates. This series does not measure absolute aridity or water stress.

Unavailable months are excluded from the means.11 initial months have no SPEI-12 value. The means use 109 available months in the first decade and 120 in the last.

Real monthly data from the nearest cell to the page’s central London coordinate; not an average over the London boundary. Dark mean lines compare the first and last ten calendar years. The 1981–2022 standardisation period already contains warming.

Reading limit This is not absolute aridity or water stress. The 1981–2022 baseline already includes warming.

Source details: SPEI

The seven source-literature categories

Seven source-literature categories, without severity colours
Extremely dry	≤ −2.0
Severely dry	−2.0 to −1.5
Moderately dry	−1.5 to −1.0
Near normal	−1.0 to 1.0
Moderately wet	1.0 to 1.5
Severely wet	1.5 to 2.0
Extremely wet	≥ 2.0

McKee et al. (1993), carried into SPEI by Vicente-Serrano et al. (2010). SPEI ≤ −2 is about the 2.3rd percentile of the standard normal.

How the drought point is sampled

The example uses the nearest source-grid cell when the query contains no grid-cell centre. The location and spatial resolution belong to the result; they do not describe a site-specific measurement. Hydro-JULES source and licence.

06 / Sea level

Sea-level projections by workflow

Calculation. Select the nearest eligible coastal location in IPCC AR6, measuring distance from the query geometry and applying the declared search radius. Read the projection’s median, 17th–83rd and 5th–95th percentiles for the selected scenario and year; convert millimetres to metres. These are uncertainty quantiles at that coastal reference, not spatial percentiles across the city.

Meaning. The Hull example uses SSP2-4.5 in 2100, relative to 1995–2014 and including vertical land motion. Medium- and low-confidence workflows remain separate distributions. A missing scenario is not substituted, and a landlocked place has no coastal reading. Mean sea-level change is not storm surge or inundation extent.

Scroll sideways to compare the full ranges →

Median17th–83rd percentiles5th–95th percentiles

The medians and ranges are close in this example. Low confidence adds processes; it does not prescribe a fixed extra rise.

Hull coastal location

Both workflows select AR6 location 286, a tide-gauge location 11.19 km from Hull’s urban boundary.

Distance is measured from the boundary, not its centroid.

Medium confidenceIncludes processes assessed at medium confidence or better.Archive through 2150.

Low confidenceAdds ice-sheet expert judgement and marine ice-cliff instability.Archive through 2300 · no SSP3-7.0 file.

Actual archived AR6 quantiles for the same Hull geometry, scenario and year. Each workflow has its own distribution; low confidence is not an error bar around medium confidence. These are changes in mean relative sea level, not inundation depths.

Reading limit Mean sea level is not storm surge or inundation extent. Coastal eligibility is checked before choosing a nearby location.

Source details: AR6 medium · AR6 low

Exact quantiles and source selection

Hull · SSP2-4.5 · 2100 · metres relative to 1995–2014
Workflow	5th	17th	Median	83rd	95th
Medium	0.270	0.384	0.586	0.844	1.055
Low	0.224	0.364	0.573	0.844	1.063

06 continued / Extreme water level

Extreme water-level recurrence

Calculation. Combine the Copernicus GTSM present-day water-level curve with a selected AR6 sea-level-rise quantile. Hold today’s 100-year water level fixed, subtract the projected rise, then interpolate its revised return period on the present-day curve. Divide 100 by that revised period to describe how often the same level recurs. Do not extrapolate beyond the published 1–100-year curve.

Meaning. This is a conditional frequency calculation with storm climate held fixed, not a forecast or loss estimate. The Hull example keeps the input locations and uncertainties visible. GTSM under-resolves tropical cyclones; flood defences are not modelled, and the offset between GTSM’s 1985–2014 and AR6’s 1995–2014 baselines is not removed.

Scroll the chart sideways to see the full curve →

Present-day central curveCurve + selected risePresent-day 5th–95th fit bounds

The same water level, reached more frequently under the selected rise.

01 / Fix the reference height

Present-day 100-year water level

Fix the reference at 5.735 m above the GTSM baseline.

The pale band describes uncertainty in the present-day curve fit.

02 / Add the selected sea-level rise

Add projected mean sea-level rise

The dashed curve uses the selected AR6 quantile. Storm climate is held fixed.

03 / Calculate the return period

Calculate the revised return period

100 → 16.2 years

Return period of the same water level

6.2× as often

At the median rise of 0.586 m

Read the intersection on the return-period axis. Frequency comparison only; no loss estimate.

The source location matters

GTSM station 39352 is 10.8 km from Hull’s urban boundary and 17.7 km from its centroid. AR6 location 286 is 11.19 km from the boundary. These coastal levels do not model flood defences.

GTSM stationAR6 location

Real GTSM and AR6 inputs; a conditional calculation, not a forecast. The 1985–2014 GTSM baseline differs from AR6’s 1995–2014 baseline; that offset is unremoved. Storm climate is held fixed, and ERA5 under-resolves tropical cyclones. The pale fit band and the AR6 quantiles describe different uncertainties.

Reading limit Storm climate is held fixed. GTSM under-resolves tropical cyclones, and its 1985–2014 baseline differs from AR6’s 1995–2014 baseline; the offset is unremoved.

Source details: GTSM · Amplification

Published curve, quantile results and calculation limits

Subtract each AR6 rise from the 5.735 m present-day 100-year level, then interpolate linearly in log(return period) on the published central curve. No extrapolation beyond 1–100 years: if the shifted level falls below the 1-year endpoint, report a frequency bound of at least 100×, not an exact recurrence. The selected Hull quantiles remain within that range.

GTSM station 39352 · metres above its 1985–2014 mean sea level
Return period · yr	Central level · m	5th–95th fit bounds · m
1	4.257	4.139–4.331
2	4.479	4.284–4.599
5	4.772	4.444–4.970
10	4.995	4.553–5.282
25	5.289	4.675–5.675
50	5.512	4.753–6.006
75	5.643	4.798–6.221
100	5.735	4.826–6.372

All archived AR6 quantiles applied to the central GTSM curve; not a combined uncertainty interval
Rise quantile	Rise · m	Recurrence · yr	Frequency factor
5th	0.270	43.20	2.3×
17th	0.384	30.31	3.3×
50th	0.586	16.16	6.2×
83rd	0.844	7.24	13.8×
95th	1.055	3.75	26.7×

08 / Subsoil clay

Clay content and prediction coverage

Calculation. Sample SoilGrids predictions of clay content at 60–100 cm, with 100–200 cm as a companion depth, using the 1 km aggregation of the 250 m product. Summarise the cells with predictions; keep masked cells in the coverage denominator. Publish coverage separately from median clay content.

Meaning. Missing predictions beneath built-up land are not zero clay. Shading requires predictions for at least half the boundary’s cells; this is a coverage rule, not a susceptibility threshold. Clay alone does not establish subsidence risk: foundations, tree influence and soil plasticity are not measured. Clay and soil-moisture readings remain separate.

01 / Choose the depth

Clay at 60–100 cm

A fixed depth below the surface. Foundation depth is not measured.

02 / Check the coverage

Prediction coverage

418 predicted1,446 unpredicted

One square = one counted cell.
Sorted by status; not a geographic map.

03 / Report the result

Median clay and coverage

35.3%

Median predicted clay
60–100 cm · predicted cells only

90th percentile 40.5%

22.4% of boundary cells predicted

Map colour withheld

The median is reported; coverage is too low for map shading.

Unpredicted does not mean zero clay. Built-up land and other masked surfaces have no prediction. The 50% rule governs map shading, not susceptibility.

London urban footprint · GHS-UCDB 1912. Counts and percentiles are measured from the source; the depth section is schematic. Clay content is a susceptibility proxy, not a subsidence hazard. Mineralogy, plasticity, foundations, trees and the drying cycle are not measured here. SoilGrids source details · Boundary source.

Compare five measured places · coverage and clay content

Measured over the gazetteer’s own boundaries, headline interval 60–100 cm
Place	Cells predicted / in boundary	Share predicted	Median clay	90th percentile	Map colour
London · ghs_ucdb:1912	418 / 1,864	22.4 %	35.3 %	40.5 %	withheld — under half predicted
Bristol · ghs_ucdb:1771	40 / 163	24.5 %	29.2 %	32.6 %	withheld
Manchester · ghs_ucdb:1780	232 / 671	34.6 %	20.8 %	23.7 %	withheld
Houston · ghs_ucdb:315	559 / 3,399	16.4 %	36.0 %	49.1 %	withheld
United Kingdom · adm0:GB	222,938 / 243,344	91.6 %	21.8 %	33.4 %	shaded

Coverage and missing predictions

Built-up land, water and other masked areas have no clay prediction. A nearby predicted cell may be used for an otherwise unsampled location, with the distance stated; an area with no usable predictions has no result. SoilGrids source and licence.

09 / Soil-moisture deficit

Soil-water shortfall

Calculation. Use Copernicus SWI_100, the deep-response soil-water indicator, over 2007–2025. Exclude frozen or melting conditions. At each source point, find the fifth percentile of its own valid record, sum daily shortfalls below that threshold within each year, then average across years with sufficient observations. Area summaries use the median of available point indices; a nearest-point fallback is identified.

Meaning. The result is a relative historical shortfall in SWI·days. It is not absolute dryness, water stress, a return period or a projection. Prolonged events can dominate the mean. The Bristol example below uses a source point near the boundary; it is shown alongside clay evidence without combining them into a score.

Bristol’s nearest land point (6.7 km from the boundary): annual shortfall below its own 5th percentile, threshold 29.5
Year	Days below threshold	Magnitude (SWI·days)	Jul–Sep mean SWI_100
2010	243 of the record’s, most of them	460.5	25.7
2013	—	74.0	—
2018	0	0.0	33.2
2022	0	0.0	33.1
2025	—	47.5	—

Bristol: 32.7 SWI·days a year · mean over 18 usable years, 2007 excluded as too short

Reading limit The threshold is defined from the point’s own record, so the index measures relative shortfall rather than absolute dryness. Most below-threshold days occurred during the prolonged 2010 event. The 2018 and 2022 observations reached the threshold without falling below it. Unfrozen conditions and a 100% quality flag support the validity of the July–November 2010 observations. Source details: Copernicus SWI.

Source and observation period

Copernicus Land Monitoring Service, Soil Water Index, deep-response indicator SWI_100. This example uses daily observations from 2 January 2007 to 31 December 2025, read on 20 September 2026. The selected land point is 6.7 km from Bristol’s boundary. Copernicus soil-moisture products.

03 / Joining evidence

Linking disclosures
to hazard data

A disclosed location can be matched to hazard data. The interpretation depends on whether the disclosure describes a portfolio or an individual site.

Portfolio and site examples

London P1 · urban footprint

Modelled inundation

GHS-UCDB boundary · Geometry source

Disclosure context

Illustrative exposure

Residential mortgages

London, UK

A portfolio distributed across a place.

Disclosure reference and reported amount

Place match

“London, UK” resolves to its GHS urban footprint.

This is not Greater London’s administrative boundary. Ambiguous names require review.

Hazard reading

River flood · 100-year model

12.69% of the boundary is wet

JRC · present-day · ~90 m cells

Undefended model. Thames Barrier protection is excluded.

Reading, source & calculation ↗

What this supports

Portfolio-level geographic screening

The reading describes land within the named area. It does not say that 12.69% of the mortgages or their value are exposed; that needs asset locations.

Illustrative disclosure, real London hazard sample. No firm or amount is assumed. The hazard reading keeps its units, source and caveats; the join produces no loss estimate or composite score. How the link is made.

Disclosure and hazard attributes

The company disclosure retains its amount, reporting scope and source reference. The hazard result retains its physical unit, geometry, scenario, period and limitations. Their shared place links the evidence without multiplying the values into a loss estimate.

Unresolved or missing locations remain gaps. An area statistic describes land within a boundary, not the distribution of a company’s assets.

Hazards selected by exposure type

The table lists the hazards selected for each exposure type. Existing hazard results are linked to the disclosure. Unresolved locations and hazards outside the selected scope are identified in the output.

Seven exposure families · relevance mapping v1, 2026-09-03
Family / spatial semantics	Hazards asked
Residential mortgage · distributed	Flood, sea level, drought (subsidence proxy on clay), heat threshold days.
Commercial real estate · distributed	Flood, sea level, cyclone, heat threshold days, wildfire. Drought is not asked: commercial structures are typically piled.
Project and infrastructure · single site	All eight: flood, warming delta, heat threshold days, wet-bulb days, cyclone, wildfire, drought and sea level.
Trade finance · single site	Cyclone, flood, sea level.
Insurance concentration · distributed	Cyclone, flood, wildfire, wet-bulb days, drought.
Own office · single site	Flood, heat threshold days, cyclone, wildfire, sea level.
Own data centre · single site	Flood, heat threshold days, cyclone, wildfire, sea level and drought.

04 / Sources and verification

Data sources and coverage

The catalogue lists data coverage, scenarios, periods, licences and interpretation limits for each source.

ObservedA recorded historical window

Present-day modelA modelled climate or designation

ProjectedA named scenario and horizon

Data sources, scenarios and years

Follow the original data providers below. Each source has its own coverage, period and interpretation limits. Listed but unused products are explicitly identified.

Source coverage is not a guarantee that every resolved place has a reading. Unacquired layers remain listed with a reason.
Hazard	Data source	Country / coverage	Scenario	Period / scenario year
River floodModelled present-day	JRC / Copernicus EMS GloFAS (opens in a new tab)	Global except Greenland, Antarctica and basins under 500 km²	None — present-day model	present-day
Resolution, licence and interpretation Native resolution 90 m Return period 100 yr Defence state undefended Licence CC-BY-4.0 Present-day river flooding without flood defences. Excludes coastal surge, surface-water flooding, Greenland, Antarctica and river basins smaller than 500 km². Depth statistics describe inundated cells only; a 100-year return period corresponds to a 1% annual chance in the model.
River flood · curveModelled present-day	JRC / Copernicus EMS GloFAS (opens in a new tab)	Global except Greenland, Antarctica and basins under 500 km²	None — present-day model	present-day
Resolution, licence and interpretation Native resolution 90 m Return period 10 · 20 · 50 · 75 · 100 · 200 · 500 yr Defence state undefended Licence CC-BY-4.0 Seven separately modelled return periods, from 10 to 500 years. These outputs are not guaranteed to increase monotonically at every location; decreases are reported rather than smoothed away. The curve is not a financial loss distribution.
Flood zonesModelled present-day	Environment Agency (opens in a new tab)	England	None — present-day model	Present-day; snapshot 2026-06-09
Resolution, licence and interpretation Native resolution 5 m Return period Z3 ≥1% fluvial / ≥0.5% tidal; Z2 0.1–1% Defence state undefended Licence OGL-UK-3.0 Present-day rivers-and-sea flood-zone outlines for England, without the benefit of flood defences. Excludes surface-water flooding. A zone is an extent classification, not a water-depth estimate; 5 m is the declared positional resolution.
Flood · NINot acquired; unvalidated	Department for Infrastructure NI (opens in a new tab)Source overview · not acquired	Northern Ireland	None declared; layer not acquired	Not established; never fetched
Resolution, licence and interpretation Native resolution 5 m Licence unconfirmed — DfI terms not yet read (blocks live use) Listed for coverage context only. This Northern Ireland product has not been acquired or validated, and its licence has not been confirmed. It is not used in the examples.
Flood risk areasAdministrative designation	Environment Agency (opens in a new tab)Source overview · not acquired	England	None — administrative designation	Not applicable to a hazard statistic; never fetched
Resolution, licence and interpretation Native resolution 1,000 m Licence OGL-UK-3.0 An administrative designation of significant flood-risk areas, not a flood extent or depth model. It has not been acquired for the examples.
Heat · warming deltaProjected	NASA NEX-GDDP-CMIP6 (opens in a new tab)	Global land cells; small islands may lack data	SSP2-4.5	2050–2060 vs 1991–2010
Resolution, licence and interpretation Native resolution 25,000 m Licence public-domain (NASA NEX-GDDP-CMIP6) Change in mean daily maximum temperature, not the absolute temperature or a count of heatwaves. Five climate models are used. A coarse climate grid cannot resolve variation at a specific address.
Heat · days above thresholdProjected	NASA NEX-GDDP-CMIP6 (opens in a new tab)	Global land cells; small islands may lack data	SSP2-4.5	2050–2060 vs 1991–2010
Resolution, licence and interpretation Native resolution 25,000 m Licence public-domain (NASA NEX-GDDP-CMIP6) Days strictly above a dry-bulb temperature threshold; the headline is 35 °C. This is air temperature, not a humidity-adjusted physiological limit. The result uses five models and retains the baseline, projection and change.
Heat · wet-bulb daysProjected	NASA NEX-GDDP-CMIP6 (tasmax + huss) (opens in a new tab)	Global land cells; small islands may lack data	SSP2-4.5	2050–2060 vs 1991–2010
Resolution, licence and interpretation Native resolution 25,000 m Licence public-domain (NASA NEX-GDDP-CMIP6) Days strictly above an approximate wet-bulb threshold; the headline is 31 °C. Humidity is available for four of the five models. The approximation is evaluated near the daily temperature maximum and can understate the true daily wet-bulb maximum.
Heat · observed meanObserved	Met Office HadUK-Grid (opens in a new tab)	United Kingdom	None — observed record	1991–2020
Resolution, licence and interpretation Native resolution 1,000 m Licence Open Government Licence (CEDA registration required) Observed mean daily maximum temperature over 1991–2020. A climatological mean does not describe heatwave severity, duration or threshold exceedances.
Heat · UK projectionProjected	UKCP18 land-rcm 12 km (opens in a new tab)	United Kingdom	RCP8.5	1980–2080
Resolution, licence and interpretation Native resolution 12,000 m Licence UKCP18 terms (CEDA registration required) Annual mean temperature from one regional-model member under RCP8.5. It provides no ensemble spread and must not be read as a central ensemble estimate or a heat-extreme measure.
Tropical cycloneObserved	NOAA NCEI IBTrACS v04r01 (opens in a new tab)	Global, subject to source coverage	None — observed record	1980–present in source description; extract pinned 2026-08-27
Resolution, licence and interpretation Native resolution 100,000 m (declared proximity radius, not grid size) Licence public-domain (NOAA NCEI) Observed storm tracks since 1980. A nearby track is not landfall, local wind speed or a damage footprint. Wind averaging periods depend on the reporting source; storm frequency and strongest-pass intensity remain separate.
WildfireObserved	NASA FIRMS VIIRS SNPP (opens in a new tab)	Global, subject to source coverage	None — observed record	2025-01-01 to 2025-12-31
Resolution, licence and interpretation Native resolution 375 m Licence Public domain · NASA FIRMS Satellite thermal detections, not distinct fires, burned area or future wildfire probability. Only presumed vegetation-fire detections enter the density. Cloud cover and satellite overpass timing limit what can be observed.
DroughtObserved	UKCEH Hydro-JULES SPEI (opens in a new tab)	Global domain: 55°S–85°N	None — observed record	1981–2022 monthly · 3 and 12 months
Resolution, licence and interpretation Native resolution 5,000 m Licence OGL-UK-3.0 A standardised water-balance anomaly relative to each location’s own history. It is not absolute dryness or water stress. The historical baseline already includes warming; a zero anomaly does not mean a wet climate.
Sea level · mediumProjected	IPCC AR6 Sea Level Projections v20210809 (opens in a new tab)	Global coastal	SSP2-4.5 (default). Available: SSP1-2.6, SSP2-4.5, SSP3-7.0, SSP5-8.5	2100 default · 2020–2150 · vs 1995–2014
Resolution, licence and interpretation Native resolution 111,000 m Licence CC-BY-4.0 Relative mean sea-level change against 1995–2014, including vertical land motion. Medium-confidence processes only. Coastal locations must be eligible; a landlocked place has no coastal reading. This is neither storm surge nor inundation extent.
Sea level · lowProjected	IPCC AR6 Sea Level Projections v20210809 (opens in a new tab)	Global coastal	SSP5-8.5 (default). Available: SSP1-2.6, SSP2-4.5, SSP5-8.5	2100 default · 2020–2300 · vs 1995–2014
Resolution, licence and interpretation Native resolution 111,000 m Licence CC-BY-4.0 A separate low-confidence workflow incorporating additional ice-sheet processes. Its own distribution must be read separately from the medium-confidence result. Scenario availability differs, including no SSP3-7.0 in this archive.
Extreme water levelModelled present-day	Copernicus CDS GTSMv3 (opens in a new tab)	Global coastal	None — present-day model	1985–2014 climatology
Resolution, licence and interpretation Native resolution 2,500 m Return period 1 · 2 · 5 · 10 · 25 · 50 · 75 · 100 yr Licence CC-BY-4.0 Total coastal water level combines mean sea level, tide and surge. The model under-resolves tropical cyclones. Published return-period estimates have fit uncertainty; station location matters, especially near flood defences.
Subsoil clayPredicted soil property	ISRIC SoilGrids 2.0 (opens in a new tab)1 km aggregated mean of the 250 m product · Poggio et al. 2021	Global land the source predicts; no cells under built-up land, water, glaciers or bare rock	None — static soil property	No epoch · 60–100 cm headline, 100–200 cm beside it
Resolution, licence and interpretation Native resolution 1,000 m (aggregated from 250 m) Licence CC-BY-4.0 Predicted clay content at 60–100 cm, with 100–200 cm as a companion interval. Built-up land and other masked cells have no prediction, not zero clay. Shading requires at least 50% coverage. Clay content alone is not subsidence probability.
Soil-moisture deficitObserved	Copernicus Land SWI time series v4 (opens in a new tab)Bytes from the Copernicus Data Space Ecosystem · TU Wien / VITO · Metop ASCAT	Global land points on a 12.5 km discrete grid; 308 ten-degree tiles	None — observed record	Daily 2007-01-02 to 2025-12-31 · SWI_100 (T = 100 days)
Resolution, licence and interpretation Native resolution 12,500 m (published grid spacing) Licence Copernicus free, full and open access (Reg. (EU) 1159/2013) Mean annual shortfall below the 5th percentile of each point’s own record. This is a relative historical measure, not absolute dryness or a return period. Prolonged below-threshold events can dominate the mean. Frozen and melting days are excluded. Soil-moisture deficit is reported separately from clay content.
Derivation · amplificationDerived	IPCC AR6 sea level (opens in a new tab) Copernicus GTSM (opens in a new tab)Derived from both inputs	Global coastal; depends on both inputs	Inherits AR6 medium-confidence reading	Inherits AR6 horizon and scenario; GTSM 1985–2014 baseline
Resolution, licence and interpretation Native resolution Not an independent source resolution; inherits input limitations Return period Derived recurrence of today’s 100-year total water level Defence state No independent defence field; inherits the limitations of both inputs Licence CC-BY-4.0 (both inputs) A derived change in the recurrence of today’s 100-year water level, holding the storm climate fixed. No extrapolation is made beyond the published curve. The sea-level and water-level baselines differ and that offset is not removed. This is not inundation extent or financial loss.

Map outlines: geometry sources and licences

These are real source geometries simplified for display. Each map is north-up and independently fitted using a local equirectangular projection with longitude scaled at its mid-latitude. The drawings do not share a scale and must not be used to measure area or distance. Hazard sampling continues to use the source geometry.

London urban footprint: European Commission, Joint Research Centre, GHS-UCDB R2019A V1.2, urban-centre ID 1912. CC BY 4.0. Snapshot 2026-08-27. This is the geometry referenced by the engine’s London entry, not Greater London’s administrative boundary.
Central London coordinate: GeoNames entry 2643743, London, UK: 51.50853° N, 0.12574° W. CC BY 4.0. Snapshot 2026-08-27. It locates an illustrative P0 query, not a sampled asset.
Precision comparison grid: NEX-GDDP-CMIP6 native 0.25° coordinates. The London query selects the cell centred at 51.625° N, 0.125° W. No temperature values are read or shown in the comparison.
England: ONS Countries (December 2025) Boundaries UK BUC, E92000001. Captured 2026-09-17. Contains public sector information licensed under the Open Government Licence v3.0. Contains OS data © Crown copyright and database right 2025.
United Kingdom: Natural Earth, 1:10m Admin 0 Countries, GBR. Public domain. Snapshot 2026-08-27. Northern Ireland and the source’s UK islands are included; small features are simplified at display scale.

GHS-UCDB · GeoNames London · ONS geography · Natural Earth public domain · CC BY 4.0 · OGL v3.0

Coverage and important limits

Hazard is not financial risk. These readings do not include asset values, vulnerability or financial loss. Different hazards are not combined into one score.

Missing is not zero. A source may not cover a place, period or requested scenario. A landlocked place has no coastal reading; it does not have zero sea-level rise.

Coverage differs by source. Global datasets and finer UK products have different footprints and periods. National flood products for Scotland and Wales and the listed Northern Ireland product are not included in these examples. Surface-water flooding and undisclosed supply-chain locations are outside this guide.

Defences matter. The sampled flood sources do not represent the benefit of flood defences. A modelled wet area can coexist with real-world protection. The sea-level examples do not calculate inundation extent.

Observed records describe the past; projections depend on their scenario and model assumptions. Example dates and source versions are stated; coverage may change as sources are updated.

How examples can be checked

Each measured example identifies its place, source, observation or projection period and units. Use the original sources linked above to check those definitions; selecting a different boundary, source version or scenario can change the result.

Illustrative diagrams are labelled and are not measured observations. Sources and limitations are provided alongside each example.

Month	Type 0	Other types	All
Jan	13	111	124
Feb	21	65	86
Mar	263	176	439
Apr	260	206	466
May	229	246	475
Jun	113	224	337
Jul	191	190	381
Aug	670	199	869
Sep	109	265	374
Oct	38	118	156
Nov	102	114	216
Dec	8	48	56

Month	Type 0	Other types	All
Jan	13	111	124
Feb	21	65	86
Mar	263	176	439
Apr	260	206	466
May	229	246	475
Jun	113	224	337
Jul	191	190	381
Aug	670	199	869
Sep	109	265	374
Oct	38	118	156
Nov	102	114	216
Dec	8	48	56

Physical climatehazard analysis