source: techcrunch ai: this ai weather startup is out-forecasting government agencies
level: technical
windborne systems released weathermesh-6, an ai weather model that outperforms the european centre for medium-range weather forecasts on key measures like surface temperature. the model produces forecasts every hour instead of every six hours and reaches 3 km resolution in europe and the continental us. the startup began by building better weather balloons and now operates about 400 balloons globally, feeding sensor data directly into its deep learning model.
the improvement comes from better data assimilation, the process of turning raw sensor readings into a unified picture of the atmosphere. windborne previously relied on ecmwf's initial conditions but now ingests its own balloon data and other sources directly. this shift required a year of tuning the transformer-based model to maintain stability. the company says its five-day forecast is as accurate as a traditional one-day forecast.
windborne sells balloon data to noaa and the us military, and forecasts to traders, but focuses on model and data infrastructure over commercial software. it has raised $25 million and uses ads-b to steer balloons away from aircraft after a minor collision last year. the startup sees direct data collection as essential for ai weather companies, arguing that without a dataset advantage, the business model is unclear.
why it matters: better ai weather forecasts can improve planning for agriculture, energy, and disaster response, and show how proprietary data can give startups an edge over government models.
source: techcrunch ai: this ai weather startup is out-forecasting government agencies