A new AI-powered weather forecasting tool launched today by startup WindBorne Systems delivers more frequent and accurate predictions on key variables than the world-leading system developed by European governments. This improvement stems from advances in how sensor readings are integrated into deep learning models.

Founded in 2019 by a group of Stanford students, WindBorne initially focused on building a better weather balloon with the intention of selling weather data. However, with the emergence of weather-forecasting deep learning models in 2022, the team realized they could capture greater value by developing their own model as well.

Today marks the release of the sixth version of that model, called WeatherMesh. The company claims this version is more accurate than both traditional and AI forecasts produced by the European Centre for Medium-Range Weather Forecasting (ECMWF), the intergovernmental organization widely regarded by meteorologists as the leading provider of accurate weather predictions. According to WindBorne's chief product officer, Kai Marshland, one simple way to understand the improvement is that WeatherMesh 6 "is as accurate five days out as a traditional forecast is the day before," particularly for surface temperature measurements.

WeatherMesh 6 generates a forecast every hour, compared to every six hours for traditional models. Its resolution has been refined to 3 kilometers in Europe and the continental United States, where data quality is highest. Traditional weather forecasts rely on complex physics models that require expensive supercomputers and take a long time to run. AI models—developed by startups and major labs like Google DeepMind—generally move faster than physics-based models, but currently lack the same resolution, number of variables, or long-term accuracy. Still, AI weather technology is improving rapidly and is already being used by major government agencies worldwide. Researchers are working to integrate it into systems that aggregate weather data and produce public forecasts.

WindBorne benefits from its unique combination of model-building and data collection. The company now has approximately 400 balloons in flight at any given time, gathering sensor readings from 15 launch sites around the world. The advances in its current model come from improvements in how the data collected by these balloons is fed into the models. The ECMWF's superiority has traditionally been attributed to its expertise in "data assimilation"—the process of converting disparate sensor readings into a comprehensive, machine-readable picture of the world. Currently, AI weather models depend on data sets produced by the ECMWF and the US National Oceanic and Atmospheric Administration. But WindBorne and other organizations are working to feed data directly into their models. Joan Creus-Costa, the company's head of AI, says that directly ingesting data from their balloons and other sources is the key reason for the improvement in the new version of WeatherMesh. It took a year of tuning and re-architecting the transformer-based model to deliver these forecasts without losing stability. "When we started doing data assimilation, we were still very heavily reliant on ECMWF," said Dean. "I predict today, if we removed ECMWF's initial conditions, we would actually still do pretty good."

The company faced a scare last year when a United Airlines jetliner struck one of its balloons. Although the plane sustained minor damage, no one was injured, partly because WindBorne followed US regulations limiting the size of its sensor package. Now, the company has added transponders to its balloons that report their location through the global aviation surveillance system, ADS-B, in an effort to reduce the risk of another collision.

WindBorne, which has raised $25 million in venture funding with a reported valuation of $85 million in 2024, sells its balloon data to NOAA, where it is used in American weather forecasting, as well as to the US Air Force and Navy. The company also sells its forecasts to investors and commodity traders. However, Dean says the company remains focused on building out its model and data infrastructure rather than commercial products, partly due to the evolving nature of the information environment. "I'm not trying to invest a massive team into building a SaaS product, if the way people want consumer information two years from now is through an agent, right," Dean said.