Weather drones are filling the data blind spot that armies and traders both need

The artificial-intelligence weather boom has a dirty secret. For all the talk of models that see storms days ahead, an AI forecaster is only as good as the observations that feed it, and lately those have run thin.

According to Bloomberg, drones are now being sent aloft to plug the gap, quietly becoming sensors that armies and energy traders alike depend on.

The blind spot sits in the lowest few kilometres of the atmosphere, the layer where most of our weather happens. For decades that slice was measured by radiosondes, the instrument-laden balloons launched twice a day from stations worldwide.

As those launches thin out, so does the data, and the timing could hardly be worse for an industry racing to build the smartest forecasts money can buy.

The pitch for AI meteorology is seductive. Google DeepMind has claimed its system is the world’s most accurate 10-day weather forecaster.

But a model trained on yesterday’s readings still needs fresh measurements to tell it what the sky is doing right now. Starve it of ground truth, and even the cleverest neural network is essentially guessing.

The decline is partly self-inflicted. In the United States, staff shortages after federal layoffs pushed the National Weather Service to suspend or reduce launches at a string of upper-air stations through 2025, punching holes in a dataset forecasters had long taken for granted.

Enter the drone. Companies such as Switzerland’s Meteomatics fly what it calls Meteodrones, small uncrewed aircraft pitched explicitly as replacements for radiosondes.

They climb several kilometres, sampling temperature, humidity, pressure and wind on the way up, then land to be sent up again.

The economics help too. A weather balloon is a single-use affair, drifting off on the wind and rarely recovered, whereas a drone gathers its data and comes home to fly again, turning patchy coverage into a dependable feed.

And the idea is moving from demonstration to duty. Meteomatics says it delivered operational weather-drone data to the US National Weather Service for the first time earlier this year, part of a wider NOAA push to fold the aircraft into everyday forecasting rather than treat them as a science experiment.

That is where the money comes in. In power and gas markets, prices swing on wind, solar output and temperature, so a forecast even slightly sharper over the coming hours is a direct trading edge.

Bloomberg reports that energy traders are among the keenest customers for better low-altitude data, for the plain reason that it pays.

The military interest runs on the same logic. The very readings that help a trader position a gas book also help an army decide when to fly, when to fire and when to move, which is why Bloomberg frames the drones as serving soldiers and speculators alike.

It all lands on an awkward truth for the AI-weather crowd. Google DeepMind, whose latest forecaster has blown away the competition, and challengers such as Switzerland’s Jua, which claims to beat Microsoft and Google, are racing to build better models.

Yet the race for better raw data is arguably more important, and a good deal less glamorous.

For Europe, there is a strategic upside buried in the story. Meteomatics is a European firm selling the picks and shovels of a data gold rush into American forecasting, a rare case of the continent exporting the kit rather than importing it, even as the marquee AI models still emerge from Silicon Valley.

There is a neat irony in it. The future of forecasting was meant to belong to the software, to sprawling models trained on decades of history.

Instead it may hinge on who can put the most sensors in the sky, and who owns the readings they send back. Whoever does can make money or win battles, a fair prize for a fleet of drones doing the unglamorous work the algorithms cannot manage without.