Most "AI for climate" headlines are either a data centre's electricity bill or a vague promise. Here is one that is neither. On 18 August, Google and the UK government launched Operation Blue Skies — a trial that uses AI to predict where aircraft will leave climate-warming contrails, and then quietly steers planes around those patches of sky.

It is a small idea with a surprisingly large lever behind it, and it is worth understanding how it actually works.

Why the white lines are a climate problem

Contrails — the streaks jets draw across the sky — are not just water vapour that dissipates. Under the right cold, humid conditions they spread into thin cirrus clouds that can linger for hours, and at night especially they trap heat that would otherwise escape to space. Add it up and contrail warming is a meaningful slice of aviation's total climate impact — and, unusually, a cheap one to attack, because only a minority of flights, in a minority of atmospheric conditions, create the persistent contrails that do most of the damage. Miss those, and you have avoided a lot of warming for very little effort.

The catch has always been prediction. You cannot avoid a contrail you did not know was coming.

What the AI actually does

This is the part worth slowing down on. Google's system uses satellite imagery and flight data to learn where contrails have formed in the past, then layers that against weather patterns to forecast where the atmosphere is primed to produce persistent contrails next. Those forecasts go to NATS, the UK's air-traffic-control provider, whose controllers can then steer aircraft clear of the flagged zones.

The elegant bit is how small the intervention is. Controllers are not rerouting planes across the ocean; they are mostly requesting vertical adjustments of up to around 2,000 feet — nudging a jet into a slightly drier layer of air where a contrail simply will not form. A tiny change in altitude, informed by a good-enough prediction, can switch a lingering heat-trapping cloud on or off. That is the whole bet: cheap prediction plus a cheap manoeuvre equals an outsized climate return.

The honest scope

It would be easy to oversell this, so here is the shape of it. Blue Skies is a £5 million, 30-month trial — part-funded by £2.65 million from the Department for Transport — run by a consortium of Google UK, NATS, the Met Office, the nonprofit Contrails.org, Imperial College London and the University of Cambridge. The flying happens in two winter windows, across 2026-2027 and 2027-2028, with roughly 20 to 40 test days each winter, in NATS-controlled Shanwick oceanic airspace — the eastern half of the North Atlantic corridor.

That airspace accounts for something like 5% of global contrail warming, which is the point and the caveat. It is a large enough stage to prove the concept at real airspace scale — the organisers call it the first state-backed effort to manage contrail avoidance across an entire oceanic airspace — but it is a proof-of-concept, not a solved problem.

And there is a real open question the trial exists to answer: does the detour cost more than it saves? Nudging a plane off its optimal altitude can burn a little extra fuel, which means a little extra CO₂. The whole approach only works if the warming avoided by skipping the contrail reliably outweighs the warming added by the fuel — and getting that trade right, at scale, with imperfect forecasts, is exactly what 20-to-40 days a winter is meant to measure.

The read

Strip away the branding and Blue Skies is a good example of what useful applied AI looks like: a narrow, measurable prediction problem, tied to a small operational change with a large physical payoff, tested properly before anyone declares victory. No chatbot, no superintelligence — just a forecast accurate enough to be worth acting on, and a system honest enough to check whether acting on it actually helps. If it works over the next two winters, it is one of the more cost-effective climate levers aviation has. If it does not, we will at least know precisely why. Either way, that is how this is supposed to go.