The NEER Flood Score — Methodology

Version: score_v1 · Layer: India Flood Atlas (1% and multi-return-period modeled flood depth, ~30 m) · Status: public

The NEER Flood Score is a 0–100 flood hazard score for any location in India. It answers one question: how severe is the flood hazard here, accounting for both how deep the water gets and how often?

Hazard, not full risk. The score measures the physical flood hazard at a location. It does not include the value, construction, or vulnerability of a specific building — a lender or insurer combines the NEER Flood Score with those to reach a full risk view. Naming it a hazard score keeps it honest and portable.

1. What goes in

For every location we model the flood depth at eight return periods — the depth-vs-frequency (“hazard”) curve:

Each is a modeled depth d(p) in metres (0 where the location stays dry at that event). More frequent events are shallower; rarer events are deeper — so d decreases as p increases.

Until all return periods are published, score_v1 is computed from the 100-year depth alone (see §5). The full-curve formula below applies as each return period comes online.

Climate scenario. The curve is modeled per climate scenario. current (today’s climate) is live. Forthcoming 2100 projections under SSP2-4.5 and SSP5-8.5 deepen the same return periods under future climate, producing a separate 2100 NEER Flood Score — so a location can be scored today and at 2100, side by side (useful across a 30-year loan). The scenario is always reported alongside the score.

2. The hazard index

We reduce the curve to one physically meaningful number — the Expected Annual Inundation Depth (EAID): the area under the depth-vs-probability curve.

H = ∫₀¹ d(p) dp   ≈   Σ  ½ · [d(pᵢ) + d(pᵢ₊₁)] · (pᵢ₊₁ − pᵢ)

(trapezoidal integration across the eight modeled points). Tail conventions, stated explicitly:

• Rarer than 1000-year (p < 0.001): not counted (bounded, negligible).
• More frequent than 2-year (p > 0.50): if the location still floods at the 2-year event, we carry d(0.50) flat out to p = 1 (it floods in even more frequent events we don’t separately model); otherwise 0.

H has units of metres·(1/yr) — read it as the average depth of water this location sees per year, integrated over all flood frequencies. It rewards both depth (deeper curve) and frequency (floods at low return periods).

This is deliberately close to how “expected annual damage” is derived in flood economics — but without a damage function, so it stays a pure hazard measure.

3. Mapping to 0–100

NEER Flood Score = round( 100 · ( 1 − e^(−H / H₀) ) )

A saturating curve, naturally bounded to [0, 100]: H = 0 → 0; large H → 100. H₀ is a single published calibration constant, fixed so the national distribution of modeled locations spreads sensibly across the bands below. H₀ is versioned with the score.

Categories

4. What the API returns (transparency by design)

The score is never a black box. Every point/report response carries its components:

{
  "neer_flood_score": 72,
  "category": "Major",
  "depth_by_rp_m": { "2yr": 0.0, "5yr": 0.2, "10yr": 0.6, "25yr": 1.1,
                     "50yr": 1.6, "100yr": 2.1, "500yr": 3.0, "1000yr": 3.6 },
  "floods_from_rp": "5yr",
  "depth_100yr_m": 2.1,
  "prob_30yr_100yr": 0.26,
  "score_version": "v1",
  "model_version": "india-atlas-2026.08"
}

• floods_from_rp — the single most intuitive fact: “this location is expected to flood as often as 1-in-5-years.”
• prob_30yr — cumulative probability over an n-year horizon: P = 1 − (1 − p)ⁿ. For the 100-year event over a 30-year mortgage, 1 − 0.99³⁰ ≈ 0.26.
• Anyone can reconstruct the score from the components — or build their own.

5. Versioning & phasing

• score_v1 (now): only the 100-year layer is public, so H is approximated from the 100-year depth banded onto the same 0–100 scale. floods_from_rp and the full curve are reported as they become available.
• score_v2 (as return periods land): full EAID integration across all eight return periods. Same field name, better math. The score_version field always tells a consumer which applied, so historical reports remain auditable.

6. Limitations (read these)

• Modeled, ~30 m. Regional hazard screening — not a parcel-level or street-level guarantee, and least reliable in steep terrain.
• Fluvial and pluvial flooding. Models both riverine (fluvial) and rainfall-driven surface-water (pluvial) flooding. It does not model coastal cyclone storm surge — on exposed coasts the true hazard can exceed what the map shows. As a ~30 m product, very fine sub-grid drainage detail may not be fully resolved.
• Hazard only. It does not know a building’s plinth height, construction, or value.
• Not an official government flood certificate. It is a modeled, dated, versioned assessment for awareness and diligence. Administrative boundaries are indicative — for external boundaries refer to the Survey of India.

The NEER Flood Score is produced by NEER — The AI for Water — from the India Flood Atlas.