The U.S. Air Force will purchase data collected by WindBorne Systems’ fleet of autonomous weather balloons under a new contract, which could be worth nearly $50 million to the startup.
WindBorne currently has approximately 700 weather balloons flying globally, collecting atmospheric sensor readings. The company sells this data to customers while also feeding it into its own AI-powered weather prediction model. Founded in 2015 and headquartered in Redwood City, California, WindBorne is among the first organizations to directly input real-world measurements into deep learning weather models.
The company has long collaborated closely with the U.S. government, particularly the National Weather Service, the U.S. Air Force, and the Navy, but this contract represents the single largest contract the company has ever undertaken. The contract value could reach up to $49 million, but the Air Force will pay only as needed, placing orders for specific regions at specific times. To date, the Air Force has placed orders totaling $1.7 million in data, and management expects additional orders in the future.
WindBorne co-founder and CTO Andrey Sushko told TechCrunch: “This is largely a vehicle to support the continued growth of our constellation, while also funding some R&D to further improve the hardware and performance of the balloons.” He added that the contract, awarded without a competitive bidding process, represents “the government’s assessment that we are the only company capable of providing this capability—collecting global data—in a way that meets its operational needs.”
In August, WindBorne CEO John Dean told TechCrunch that the company had completed a $37 million Series B funding round to continue expanding its operations. Earlier this summer, the company released the latest version of its forecasting model, WeatherMesh 6. The company claims that this model is more accurate than traditional models run by the European Centre for Medium-Range Weather Forecasts (ECMWF), which simulate the atmosphere using physical laws.
Since the emergence of the Transformer model architecture and the release of a large-scale weather dataset by ECMWF several years ago, AI-based weather forecasting has become one of the most productive applications of deep learning techniques. Previously, meteorologists relied on expensive supercomputers to run detailed equations to approximate atmospheric behavior. Today, end-to-end learning can absorb decades of weather data and deliver forecasts more quickly.
Models developed by leading laboratories and government agencies are demonstrating greater accuracy than traditional forecasts, although the field remains challenging, and both types of predictions are still essential for forming a useful understanding of weather. Google DeepMind’s latest weather model provides important new capabilities, and its cyclone prediction tool is increasingly becoming a core component of the U.S. government’s response to hurricane season.
In August of this year, WindBorne became the first organization to send a powered-free flight system—meaning a system without an engine—into the eye of a typhoon to collect data. The company controls the balloon by adjusting its altitude, releasing gas, or dropping sand carried as ballast, allowing the aircraft to enter the appropriate wind field and be carried by the wind to its target location.
Sushko said, “This is a fairly novel step, and achieving it involves two key elements: first, having accurate wind and weather models and knowing how to use them for navigation; second, having a large enough constellation to ensure balloons are in the right locations.” He added, “This marks the first time deep learning models have been used for actual operational decision-making.”