WindBorne's WeatherMesh-6 beats ECMWF with hourly AI forecasts
A Stanford startup's AI matches 5-day forecasts in just 1 day using 400 balloons.
Deep Dive
WindBorne Systems released WeatherMesh-6, an AI weather model that out-forecasts the European Centre for Medium-Range Weather Forecasts (ECMWF). The model is as accurate five days out as traditional forecasts are one day before, produces hourly (vs. 6-hourly) updates, and has 3 km resolution in Europe and the continental U.S. WindBorne uses about 400 balloons for data collection and direct data assimilation, reducing reliance on government datasets. The company has raised $25 million and sells its balloon data to NOAA and the U.S. Air Force and Navy.
Key Points
- WeatherMesh-6 matches ECMWF's 5-day accuracy in 1 day for surface temperature predictions.
- Model produces hourly forecasts at 3 km resolution, vs. traditional 6-hourly at coarser grids.
- WindBorne's 400-balloon network enables direct data ingestion, reducing reliance on ECMWF inputs.
Why It Matters
Hyper-accurate, hourly weather AI can save billions across agriculture, logistics, and energy sectors.