AquaStateWater & climate intelligence
Updated 13:50 UTC · 249 countries
↑ ↓ moveEnter openEsc close

Method

The arithmetic behind the ledger.

Three scales, one deterministic core. Every number on the site carries its formula, inputs, source, confidence and contribution.

Three scales, one question

AquaState answers “are we covered until the next rain?” at three resolutions. The scales are deliberately separate: a national score never silently rewrites a dam score, and a forecast never rewrites an observed score.

ScaleIndexQuestion
WorldWCRI-2.0How resilient is this country’s water-climate system, and which pressure is driving it?
CountryWRI-2.0Is stored water covered until the next rain, and what does the horizon say?
DamDRI-2.0What is this reservoir’s hydrological resilience now, and what would the forecast change?

WCRI-2.0 — the country index

0–100, higher is better. Six pillars, each with a published formula and confidence. Weights are expert-judgment screening weights; they are visible and versioned.

PillarWeightFormula (plain language)
Water availability0.220.55 × freshwater-per-capita score + 0.45 × aridity score. Per-capita is log-scaled between 500 m³ (0) and 10,000 m³ (100).
Drought & soil0.200.40 × SPI-3 + 0.25 × SPI-12 + 0.20 × soil-moisture percentile + 0.15 × dry-spell score.
Groundwater0.140.60 × root-zone percentile + 0.40 × root-zone trend score. Labelled a proxy when no aquifer inventory exists.
Flood resilience0.14100 − [0.75 × forecast-discharge pressure + 0.25 × active-event pressure].
Wildfire resilience0.14100 − [0.65 × fire-weather pressure + 0.20 × drought coupling + 0.15 × active-fire pressure].
Desertification resilience0.16100 − [0.40 × aridity + 0.30 × rainfall trend + 0.20 × soil trend + 0.10 × forest change].

Aggregation and the cascade

base = 0.22·W + 0.20·D + 0.14·G + 0.14·F + 0.14·R + 0.16·L

penalty = 0.12·max(0, 30 − min(pillars))
        + 0.05·max(0, 45 − D)·max(0, 45 − R)/45
        + 0.05·max(0, 45 − D)·max(0, 45 − L)/45

WCRI = clamp(base − penalty, 0, 100)

The first penalty is a bottleneck guard (max 3.6 points). The other two are the explicit drought–fire and drought–desertification cascade terms (max 2.25 each). The score is never hidden behind a multiplier; the formula is printed on the country page.

Bands: Resilient ≥70 · Stable 55–70 · Watch 40–55 · Stressed 25–40 · Critical <25. A country with less than 40% of the index weight available is shown as NO_DATA, not given a false score.

The four pressures

PressureWhat is measuredWhat it is not
Drought / الجفافEmpirical SPI-3 and SPI-12, soil-moisture percentile, days since >2 mm rain.Not an official drought declaration.
Flood / الفيضانGloFAS forecast peak discharge vs the same point’s 60-day baseline, plus active EONET floods.Not a local flood warning or a hydraulic model.
Wildfire / حرائق الغاباتVPD, dry spell, wind and soil percentile, coupled to drought, plus active EONET fires.Not an ignition forecast or a fuel model.
Desertification / التصحرAridity (P/PET), Sen rainfall trend, soil trend, forest-cover change.Not a land-cover map or a soil-degradation survey.

All four pressures are reported separately as 0–100 pressure (higher is worse) and inverted into the resilience pillars. This keeps the map readable: the index layer is green=resilient, the hazard layers are green=low pressure to red=extreme.

WRI-2.0 — the water system

The original water-state engine is preserved. It computes five pillars — storage buffer, soil moisture, underground reserve, forecast opportunity and storage trend — and a separate Smart Horizon verdict.

  • Mid-rank empirical percentiles replace ad-hoc ranks; the current year is never in its own baseline.
  • SPI-3 and SPI-12 are added to the drought signal.
  • Weights are effective weights: nominal weight × confidence, renormalised over available pillars.
  • The Φ buffer bridge and the continuous low-storage guard come from the space app; there is no hard step at a threshold.
  • The score is the current-state score. “OK until rain” is a separate horizon verdict, not a rewritten index.
WRI = Σ wᵢ·Sᵢ − 0.15·max(0, 35 − min(Sᵢ))
Φ    = rain_gate · σ((D_autonomy − D_until_rain)/7) · σ(ln((V_in14 + ε_V)/(14·Q_out + ε_V))/0.5)

DRI-2.0 — the dam deep index

The space app’s signed HRI-1.0 mathematics is ported unchanged. It is the only index that uses reservoir bathymetry, basin climate windows and event-scale SCS-CN runoff.

Sub-indexWhat it readsNominal weight
SSISurface storage: usable fill against the same-week baseline.0.30
CMICatchment meteorology: 30/90-day rain and dry spell.0.20
SRRISoil and runoff readiness: SCS-CN with antecedent soil.0.15
SAISubsurface/aquifer; NO_DATA for pilot basins smaller than GRACE’s footprint.0.20
EESIEvaporative and ecological stress.0.15

The signed fixtures are preserved: Beni Haroun’s storm case reproduces HRI_now=38.616, HRI_proj=83.293, Φ=0.94977, HRI=81.05, RECHARGING. The dry case reproduces 24.9 CRITICAL; the low-autonomy case reproduces 19.9. These are regression tests, not marketing numbers.

Confidence and coverage

Confidence never changes the score. It is reported beside it, decomposed per pillar, and used to renormalise the effective weights. Coverage is the share of nominal weight with a real source; under 40% the country is NO_DATA.

  • Live — a fetch succeeded and a timestamp was stored.
  • Cached — served from the last successful fetch.
  • Modeled — a mass-balance, trend or screening calculation.
  • Reference — a public inventory or annual statistic.
  • Proxy — a labelled substitute, e.g. root-zone wetness for groundwater.
  • Unavailable — neutral 50 with low confidence, never a hidden zero.

What this is not

  • Not an operational flood, fire or drought warning system.
  • Not official water accounting; reservoir storage is modeled/blended, not telemetry.
  • Not a groundwater model; GRACE is not ingested in this MVP.
  • Not a complete dam census; the atlas is notable dams with public records.
  • Not a forecast of conflict, migration or food security.

Reproduce the world layer with python manage.py bootstrap_world; recompute the dam layer with python manage.py compute_indices --all. The source ledger at /sources/ lists every fetch and timestamp.