About GaIA Echo

GaIA Echo is a forensic climate console: it reads ECMWF ERA5 reanalysis of catastrophic events and narrates them in real time using Claude Opus 4.7 with reasoning visible. It also runs a managed agent that watches the equatorial Pacific for the next El Niño, and a Monte-Carlo simulation engine for atmospheric scenarios.

It is not an operational forecast. It does not replace AEMET, the NHC, ECMWF or any national weather service. The simulation engine is a parametric Monte-Carlo wrapper around physical first-order equations — useful for stress-testing scenarios and training, not for issuing public warnings.

Forensic analyses are grounded in 8×8 grids at 0.5° resolution from Open-Meteo's ERA5 archive, six-hourly cadence, with a hardcoded 1991–2020 climatology baseline per event. The Climate Dial counterfactual applies Clausius–Clapeyron (~7%/°C column water), super-CC (~14%/°C peak hourly), Emanuel potential intensity (~5%/°C TC wind, ~10%/°C TC rainfall), and Perkins-Kirkpatrick (~+2 days/°C heatwave duration). The Guardian reads NOAA CPC weekly Niño-3.4 and monthly standardized SOI; it is forbidden from inventing numbers. The Simulation engine ensembles 200 perturbed members against five historical analogues with live initial conditions from Open-Meteo.

ERA5 grids do not include sea-surface temperature, soil moisture, vapour pressure deficit or vertical wind shear; when Opus invokes them it is required to mark the statement as inference. The simulation engine is parametric, not a GCM; it is good at envelopes, not exact rainfall totals. Climate Dial outputs are first-order — they say what the physics demands, not what a coupled model would simulate.

Next.js 16 / React 19 / TypeScript strict, Anthropic SDK with Claude Opus 4.7 (adaptive thinking, summarized display, xhigh effort), Mapbox GL JS for 2D, Three.js for 3D, lucide-react for icons. SQLite for persistence. No authentication. Open source under MIT.