WindBorne’s groundbreaking deep learning model delivers real-time, medium-range global weather forecasts, surpassing previous state-of-the-art records. The model now powers the company’s constellation of smart, long-duration sensing balloons.WindBorne Systems, the environmental and weather intelligence company behind the only constellation of autonomous, long-duration sensing balloons, announced today its record-breaking AI forecasting model, named WeatherMesh. The new weather model now generates the most accurate global weather forecasts through innovative deep learning and data techniques.
WeatherMesh, a deep learning numerical weather prediction (DLNWP) model, has reached unprecedented levels of global prediction accuracy for real-time, 24-hour forecasts, surpassing records previously held by Google DeepMind’s AI weather model, GraphCast. WindBorne’s state-of-the art model surpassed GraphCast by a significant 11% margin in accurately predicting a primary forecasting metric, 500 millibar geopotential height. This measure of pressure describes the movement of weather systems, evaluated at 0.25-degree resolution.
The model also surpassed the gold-standard for traditional physics-based weather modeling, known as numerical weather prediction (NWP), held by the European Centre for Medium-Range Weather Forecast (ECMWF)’s medium-range model (ENS). WeatherMesh outperformed ENS by 6%-14% for five-day forecasts in ensemble mean.
WindBorne’s WeatherMesh, which predicts temperature, pressure, wind speed and direction, dew point, cloud cover, incoming radiation, and precipitation, represents the equivalent of years-worth of forecasting advancement. WindBorne achieved these accuracy and operational feats while being substantially more efficient than both other AI-based and traditional NWP models, which run on supercomputers using hundreds of times more compute. WeatherMesh can run on a gaming laptop.
WindBorne’s breakthrough can be attributed to its innovative machine learning architecture, modeling techniques, and vast, powerful data sources. WindBorne’s forecasts are uniquely suited for live, operational use cases. WeatherMesh’s generative AI underpinnings and transformer-based architecture enable it to easily process large amounts of data. This includes both live input operational data, as well as extensive historical data sets, such as the ERA-5, the ECMWF’s estimate of hourly weather variables for the past eight decades.
WeatherMesh generates forecasts with one-hourly resolution, making them ideal for real-time applications. Other top AI models, by contrast, produce forecasts with six-hourly resolution. WeatherMesh also benefits from WindBorne’s unique, proprietary atmospheric data, which includes real-time measurements of temperature, wind speed and direction, pressure, and humidity, captured through WindBorne’s balloon constellation.
WeatherMesh now powers WindBorne’s entire autonomous balloon network, making WindBorne the first company to control real-world assets with AI forecasts at scale and replacing the company’s use of NWP forecasts for balloon navigation.
WindBorne’s forecasting debut represents a major inflection point in the company’s product line, market opportunity, and climate impact. Before expanding into forecasts, WindBorne, founded in 2019, spent the last five years inventing and scaling an unparalleled, atmospheric data-as-a-service offering. Its exceptionally sustainable, cost-effective balloon network can now observe critical atmospheric data anywhere in the world, unlocking up to 150x more data per cost dollar.
WindBorne’s growing data set is the environmental data most urgently needed to significantly improve forecasts. Accurate forecasts rely on real-time observations of the atmosphere from the surface through the stratosphere, yet the WMO has deemed 85% of the atmosphere inadequately measured. Because weather blows in from data deserts like the Pacific Ocean, even the best-instrumented downwind land masses like the U.S. lack weather certainty.
WindBorne is making rapid progress as the only company poised to close the atmospheric sensing gap. Alternative atmospheric data collection methods such as traditional weather balloons can only capture a single data profile directly above their land-based launch point. Without WindBorne, in-situ atmospheric profiles over oceans and the most remote places can only be gathered through aircraft-dropped devices called dropsondes, a highly costly approach that poses risk to operators.
By contrast, WindBorne’s sustainable balloons fly autonomously for weeks at a time toward programmed targets, gathering dozens of vertical profiles in a single flight. This in-situ data isn’t possible to collect through other weather observation technology such as ground stations or satellites.
As extreme weather becomes more frequent and severe in the wake of climate change, demand for more accurate forecasts is growing. Forecasts, which have the potential to improve dramatically, are critical for equipping people, governments, and businesses with better safety and decision-making intelligence. More accurate forecasts also have the opportunity to reduce carbon emissions by half a gigaton per year, through more intelligent aviation, transportation, and logistics routing, better wildfire management, and better use of energy grids to accelerate the transition to renewables.
WindBorne is revolutionizing end-to-end weather intelligence by fusing the power of essential, proprietary data with unprecedentedly accurate, operation-ready forecasts.
“WindBorne is the only company that is delivering 100x-scale improvements on both better weather data and weather modeling,” said WindBorne co-founder and chief executive officer, John Dean. “This also means WindBorne has a critical opportunity to make forecasts more trustworthy for everyone, including individuals and governments navigating how to prepare for and manage disasters, as well as companies and decision-makers who can now make real-world choices with better information.”
Already, governments spend $10 billion annually on weather observation. In addition to public sector applications, WindBorne’s forecast offering unlocks vast commercial applications across industries such as transportation and logistics, agriculture, aviation, travel, events, utilities, insurance, energy grid operations, and more, as part of the growing $100 billion addressable forecasting market. WeatherMesh also deepens WindBorne’s ability to support humanitarian-driven weather initiatives.
WindBorne has partnered with major government agencies such as the National Oceanic and Atmospheric Association (NOAA), the U.S. Air Force, the Office of Naval Research, the National Center for Atmospheric Research (NCAR), and the Defense Innovation Unit (DIU), delivering on contracts every year since founding and generating millions in revenue to date. The Air Force has already partnered with WindBorne to support its deep learning innovation.
“Navigating today’s weather and climate events with current forecasts is like stumbling in the woods in the dark,” said WindBorne co-founder and Chief Product Officer Kai Marshland. “WindBorne has invented the equivalent of the flashlight. We’re in the unique position to use both data and real-time forecasts to slash weather uncertainty and tackle climate change from both sides. First, sharing intelligence that makes every individual safer in the event of anything from a heat wave to a hurricane. And second: leveraging this intelligence to reduce emissions themselves.”
WindBorne has attracted a world-class engineering and scientific team, recruiting from the Stanford Space Initiative, as well as some of the world’s most renowned meteorological organizations. WindBorne’s full-stack technical talent spans deep learning and generative AI, electrical and sensor engineering, and autonomous software. The team holds a range of patents for its atmospheric sensing technology. The company’s meteorology team has decades of industry experience. WindBorne is backed by Footwork, Khosla Ventures, Ubiquity Ventures, Susa Ventures, Pear VC, Jetstream Ventures, and Harvest Venture Partners, among others. The company has raised $10 million to date, most recently a $7 million seed round in May 2023. The company was founded by Andrey Sushko, Joan Creus-Costa, Paige Brocidiacono, Dean, and Marshland in 2019.
