Satellite AI Emission Tracker

Facility & Wind

True emission rate Q 500 kg/hr

Wind speed u 5.0 m/s

Atmospheric stability

AI Retrieval Pipeline

Sensor noise σ 100 µg/m³

Detection threshold 300 µg/m³

Retrieval Readouts

True Q

500 kg/hr

AI-estimated Q̂

— kg/hr

Retrieval error

—

Detection confidence

—

Peak plume concentration (noisy)

—

How it works

The plume follows the classic steady-state Gaussian dispersion model for a stack of height H, with ground reflection:

C(x,y,z) = Q / (2π u σy σz) · exp(-y²/2σy²) ·
           [exp(-(z-H)²/2σz²) + exp(-(z+H)²/2σz²)]

σy(x) and σz(x) are the Briggs rural dispersion coefficients, which grow with downwind distance x and depend on the Pasquill–Gifford stability class (A = strong daytime mixing, F = calm stable night). Because the model conserves mass, the flux of gas through any downwind cross-section equals the true emission rate:

Q = u · ∬ C(y,z) dy dz   (exact, noise-free)

This is the mass-balance / Integrated Mass Enhancement (IME) principle behind real satellite methane retrievals (GHGSat, Sentinel-5P TROPOMI, MethaneSAT): image a downwind cross-section, sum the enhancement above background, and divide out the wind speed to recover an emission rate — without ever seeing the source directly.

Sensor noise σ — Gaussian noise added to every simulated pixel before integration, modelling instrument and retrieval noise in a real spectral inversion.
Detection threshold — the AI classifier zeroes out any pixel below this concentration, exactly as a real detection algorithm suppresses sub-threshold noise; set it too high and the tail of the plume — and the estimate — gets clipped.
Detection confidence — a logistic score of the noisy peak concentration against the threshold, standing in for a CNN plume-classifier's output probability.
Resample scene — draws a fresh noise realisation, the way each satellite overpass gives an independent noisy retrieval of the same true source.

Plume voxels (concentration) Facility stack Wind vector

🛰 simulated satellite column view

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.