GKN Aerospace — Hazard Zone Simulator

Interactive consequence-model for a methyl methacrylate (MMA) tank emergency, rendered on a live Leaflet / OpenStreetMap map of Garden Grove, CA. Drag the sliders to recompute the blast rings, the toxic-vapor plume, and the evacuation footprint in real time. All math runs client-side — no backend, no key.

Live: https://joevezzani.github.io/hazard-zone-sim/

Real vs modeled (important): the facility location and the official evacuation order (yellow box — N of Trask Ave, S of Ball Rd, E of Valley View St, W of Dale St; ~40,000 residents) are real/authoritative. The blast rings and toxic plume are modeled estimates, screening-level (±factor 2–3), not official guidance.

---

The math

1. Blast — vapor-cloud explosion (VCE)

Fuel mass from the spilled/involved volume:

m  = V_gal × 3.785 L/gal × 0.94 kg/L          (MMA liquid density ≈ 0.94 kg/L)

TNT-equivalent yield (energy basis):

W_TNT = m × ΔH_c × η / E_TNT
        ΔH_c  ≈ 26.7 MJ/kg   (MMA net heat of combustion)
        E_TNT = 4.6  MJ/kg   (TNT)
        η     = VCE yield, 1–10 % of fuel energy coupling to blast

Overpressure radii by cube-root (Hopkinson–Cranz) scaling, R = k · W_TNT^(1/3), for four damage bands: fireball/total destruction, severe (~10 psi), moderate (~3 psi), glass-break/eardrum (~1 psi).

Cascade: each adjacent tank that detonates adds its full inventory to the fuel mass, so W_TNT ∝ V_total and every radius scales with V_total^(1/3).

TNT-equivalent is blast energy only — it is not a measure of radiation or fallout.

2. Toxic plume — dispersion

Health thresholds and the ppm → mass-concentration conversion (MMA, MW = 100):

IDLH   1000 ppm  (lethal)
ERPG-2   75 ppm  (serious health effects)
ERPG-1   10 ppm  (irritation)

C[g/m³] = ppm × MW / 24450

Source rate drives reach. The plume footprint is set by the evaporation/venting rate Q (kg/s), not the tank's total volume. Distance to a given threshold scales sub-linearly with the source and weakly inverse with wind:

d(Q,u) = d₀ · (Q / Q₀)^0.5 · (u₀ / u)^0.3
         Q₀ = 4 kg/s,  u₀ = 8 mph   (anchor conditions for d₀)

The exponent ≈ 0.5 comes from the Gaussian dispersion coefficients growing with downwind distance, so even a very large source only stretches the footprint modestly.

Cascade: leak rate is per tank. N failing tanks give an effective source Q = leak × N, so the plume grows with the cascade alongside the blast.

MMA vapor is ≈ 3.4× as dense as air, so the cloud slumps and resists vertical mixing. Two dense-gas dispersion models are offered:

• Britter–McQuaid (B-M) — empirical dense-gas correlation. The heavy cloud hugs the ground and holds concentration farthest downwind. Largest footprint.

• DEGADIS (two-phase) — US EPA / Coast Guard model. A gravity-driven slumping phase, with vertical mixing suppressed, governed by the bulk Richardson number

  Ri_b = g' · L_v / u*²        g' = g (ρ_c − ρ_a) / ρ_a   (reduced gravity)
  

  hands off to a passive Gaussian phase once Ri_b < ~0.1 (the cloud has entrained enough air that ρ_c → ρ_air), launched from a virtual source matched to the flattened cloud's width. DEGADIS lands between B-M and a neutral plume.

A neutral-Gaussian (passive-gas) solution is the lower bound — it lofts and dilutes freely. It understates a dense gas like MMA, so it is documented in models/ for reference but not exposed in the UI.

3. Wind and direction

The plume travel bearing is wind_from + 180°. Live wind comes from the public Open-Meteo API; a time scrubber replays ~48 h of past hourly wind plus ~48 h of forecast, re-aiming the plume hour by hour.

4. Multi-model comparison

Each distance anchor (d₀) is taken from independent first-principles runs of four LLMs (Claude, Grok 4, Gemini 2.5 Pro, GPT-5). The main map shows one model (default GPT-5); the others overlay in their own colors for side-by-side comparison. The raw numbers, prompts, and per-model methods are in models/.

---

Editing this site

The entire site is one file: index.html (inline CSS + JS). No build step, no framework.

1. Edit index.html.
2. Preview locally: python3 -m http.server 8000 → open http://localhost:8000.
3. Commit and push to main.

Pushing to main deploys automatically. GitHub Pages rebuilds within ~30–60 s. No secrets, no CI config to touch.

Files

File	Purpose
index.html	The whole app — markup, styles, map logic, physics.
models/	First-principles model runs (numbers, prompts, methods) behind the comparison.
og.png	1200×630 social share image (Open Graph / Twitter card).
.nojekyll	Tells GitHub Pages to serve files verbatim (skip Jekyll).

Notes

Educational / situational-awareness visualization. The dispersion and blast figures are first-order engineering approximations (dense-gas screening + cube-root TNT scaling), not a substitute for professional emergency modeling.