Windborne Systems Launches WeatherMesh 6 AI Weather Model

Windborne Systems released WeatherMesh 6 today, a new version of its AI weather forecasting model that the company says surpasses both traditional and AI forecasts from the European Centre for Medium-Range Weather Forecasting on several variables. The model produces a forecast every hour at 3 km resolution across Europe and the continental United States.

Kai Marshland, Windborne’s chief product officer, said WeatherMesh 6 “is as accurate five days out as a traditional forecast is the day before,” particularly on surface temperature measurements.

Traditional models generate forecasts every six hours using physics-based simulations that run on expensive supercomputers. The company operates about 400 balloons launched from 15 sites worldwide that collect sensor readings fed directly into the model. Joan Creus-Costa, Windborne’s head of AI, said the direct ingestion of balloon data is the main reason for the accuracy gains in WeatherMesh 6.

John Dean, Windborne’s CEO, noted that the company spent a year tuning and re-architecting its transformer-based model to maintain stability while incorporating the new data streams. “When we started doing data assimilation we were still very heavily reliant on ECMWF,” Dean said. ” Dean also questioned the viability of AI weather companies that lack proprietary data advantages.

“I don’t understand, personally, the business model of being an AI based weather company without a data set advantage,” he told TechCrunch. Windborne was founded in 2019 by Stanford students who initially focused on selling weather data collected by balloons.

After deep-learning weather models emerged in 2022, the company began building its own forecasting system alongside its data collection operations.