Read the article
here.This summer, WindBorne put its next-generation weather technology to the test at RIMPAC 2026, the world's largest international maritime exercise, which brought together the U.S. and 30 partner nations in and around Hawaii. Throughout the exercise, WindBorne delivered new atmospheric observations and twice-daily AI weather forecasts to U.S. Navy meteorology and oceanography teams.
The effort combined two parts of WindBorne's weather system: Atlas, the world's largest constellation of long-duration weather balloons, and WeatherMesh, its AI weather forecasting model. Atlas provided observations from data-sparse regions of the atmosphere. WeatherMesh delivered 48-hour forecasts that worked within Navy software and produced results comparable to the Navy's established weather model.
The work was conducted under WindBorne's Cooperative Research and Development Agreement (CRADA) with Commander, Naval Meteorology and Oceanography Command (CNMOC).
Bringing a new weather model into existing tools
Ahead of RIMPAC, WindBorne worked with the U.S. Naval Research Laboratory to make WeatherMesh output compatible with the Interactive Scenario Builder (Builder), software the Navy uses to predict how weather conditions affect radar and communications performance.
By the start of the exercise, WeatherMesh forecasts could load directly into Builder, allowing Navy teams to evaluate a new AI weather model without adopting a new platform or changing the software they already rely on.
WindBorne delivered the forecasts in GRIB2, the standard format for gridded weather data, with each run structured in three-hour steps to match the Navy's own model output. Navy teams received both a deterministic forecast and an ensemble mean, the average of multiple runs used to account for atmospheric uncertainty.
"The Navy teams at RIMPAC set a high bar. They tested WeatherMesh against operational requirements and gave us fast, candid feedback that will help shape the next generation of the model," said John Dean, Co-Founder and CEO of WindBorne. "We couldn't ask for better partners in building weather technology that works in the real world."
AI forecasting side by side with an established model
During the exercise, Naval Oceanography Operations Command (NOOC) used WeatherMesh data to create electromagnetic maneuver warfare planning products for RIMPAC participants and compared its output with COAMPS, the Navy's standard physics-based weather prediction model.
"The WeatherMesh evaluation yielded valuable data on AI model performance during the exercise. Its forecasts produced broadly similar results across our comparisons, and the model's speed offers a promising look at how we might accelerate future weather planning workflows” said LT Tristan Walker of NOOC.
More observations from the world's oceans
Weather forecasting has a fundamental data problem: much of the atmosphere is difficult to observe. Traditional weather observations rely heavily on fixed infrastructure, which is far more limited over the world's oceans than on land. That leaves vast areas of the atmosphere with few direct measurements.
Atlas is built to reach these areas. Its balloons profile the atmosphere from near the surface to above 80,000 feet, measuring temperature, pressure, humidity, and wind along the way. During RIMPAC, 133 WindBorne balloons collected measurements across the waters around Hawaii and made them available to Navy teams in near real time. More observations in these gaps give forecast models a better starting point, which means better forecasts downstream.
Looking ahead
RIMPAC also gave WindBorne a clear focus for the next generation of WeatherMesh: higher vertical resolution in the planetary boundary layer (PBL), the lowest part of the atmosphere. Conditions in the PBL can change significantly over short vertical distances, so representing it in greater detail can help capture features that coarser models may miss. Changes in temperature and humidity in the PBL can also affect how radio signals travel near the surface, making this part of the atmosphere particularly important for communications planning.
WeatherMesh already includes more vertical pressure levels than most current AI weather models. WindBorne is now working on the PBL from both sides: adding more near-surface detail to WeatherMesh, and flying Atlas balloons lower to measure the boundary layer directly.
"Some of the hardest places to observe and forecast weather are over the open ocean," said Sidney Dills of CNMOC's N95 Office of Research and Technology Applications. "New atmospheric observations and rapid AI forecasts give us a real shot at closing gaps that have existed for decades. This CRADA provides a vital framework to evaluate these emerging capabilities in real-world scenarios, putting cutting-edge forecasting directly into the hands of our Sailors."
