AI Contrails: Operation Blue Skies and Rerouted Planes

The surprising fact behind AI contrails Operation Blue Skies is that a plane’s exhaust gas is not its biggest climate problem. Contrails, the white lines behind aircraft, account for roughly 35% of aviation’s warming impact according to recent IPCC analysis, and only slightly over half of aviation’s warming comes from carbon dioxide at all. In August 2026, Google announced that its contrail-avoidance work with the UK Government and aviation industry partners, branded Operation Blue Skies, is expanding across the North Atlantic. The core result behind it: in a six-month trial with American Airlines covering 70 flights, AI predictions of contrail-forming regions let pilots adjust altitude and achieved a 54% reduction in contrails, at a cost of about 2% more fuel on those flights, which scales to roughly 0.3% fleet-wide.

Key Takeaways

• A contrail forms when hot, humid engine exhaust meets air cold enough for water vapor to freeze onto soot particles, typically above 8,000 m (26,000 ft) where air temperature is below about −36.5 °C.
• Short-lived contrails evaporate. The problem ones persist and spread into thin cirrus-like sheets, which trap outgoing heat from Earth more than they reflect incoming sunlight.
• Measured net effect of that added cloudiness is a radiative forcing of roughly 0.03 to 0.06 watts per square meter.
• Google’s system uses computer vision trained on hundreds of hours of labeled GOES-16 satellite imagery, plus weather data and flight paths, and can detect contrails within 30 minutes of formation.
• Trial result: 54% fewer contrails, 2% more fuel on trial flights (about 0.3% fleet-wide), 70 flights over six months, with American Airlines and collaborators including the Rocky Mountain Institute.

Why a cloud you can see for ten minutes matters for climate

Start with what a contrail is. Jet fuel combustion produces water vapor. At cruise altitude the air is extremely cold, often below −40 °C, and the exhaust plume mixes with it. If the mixed air becomes saturated with respect to ice, water vapor deposits onto soot particles from the engine and forms ice crystals. That’s the line you see.

What happens next depends entirely on the surrounding air. If the air is dry, the crystals sublimate and the contrail vanishes in seconds. If the air is already holding more water vapor than ice can normally sustain, the crystals keep growing by pulling in ambient moisture, the line widens, and it can persist for hours and spread into a sheet.

Those persistent, spreading contrails are the climate issue. A thin high cloud is very good at absorbing infrared radiation leaving Earth’s surface and re-emitting some of it downward, and comparatively poor at reflecting incoming sunlight. Net effect: warming. The estimated radiative forcing from the added cloudiness is around 0.03 to 0.06 W/m².

Here is the operationally crucial part: the regions where persistent contrails form are thin layers, often a few thousand feet deep. Fly 2,000 feet higher or lower and you miss them entirely.

How the AI contrails Operation Blue Skies system computes a route change

Step 1: Learn to see contrails. A computer vision model was trained on hundreds of hours of hand-labeled satellite imagery from GOES-16, a geostationary weather satellite. This is a straightforward supervised learning task with an unusual label set: humans marked which streaks in infrared imagery were aircraft contrails rather than natural cirrus. The trained detector finds contrails within 30 minutes of formation.

Step 2: Connect each detected contrail to the flight that made it. With flight track data, you can attribute a contrail to a specific aircraft at a specific altitude and time. That gives you labeled examples: this plane, in this atmospheric condition, produced a persistent contrail; that one did not.

Step 3: Forecast where the contrail-forming layers will be. This is the hard part, and it’s a humidity problem. Predicting temperature at altitude is relatively reliable. Predicting whether a thin layer will be ice-supersaturated is much harder, because upper-atmosphere humidity is poorly measured and highly variable. The satellite-plus-flight-data pairs give the model ground truth to learn from and to be scored against.

Step 4: Give the pilot something actionable. A forecast is only useful if it reaches the cockpit as a specific instruction. Google’s implementation put predictions into pilots’ flight tablets as altitude-adjustment suggestions on specific route segments.

Step 5: Measure whether it worked. This is what makes the trial credible. Because step 1 detects contrails from satellite, you can check whether the flights that took the advice produced fewer contrails than baseline. That’s how the 54% figure was obtained: measurement, not modeling.

What it got right

A 54% reduction, measured by an independent detection channel rather than assumed from the model’s own predictions. And an honest accounting of the tradeoff: 2% more fuel on trial flights, about 0.3% fleet-wide, because only a minority of flights need to change anything.

What remains uncertain

Seventy flights over six months is a small sample. The humidity forecasting that drives the whole system is the weakest link, and mis-predicting a supersaturated layer means either a pointless fuel penalty or a missed contrail. Burning more fuel adds CO2 immediately and permanently, while a contrail warms for hours, so the arithmetic of the tradeoff depends on estimates of contrail warming that carry real uncertainty. And avoiding contrails cannot conflict with air traffic control separation rules, which constrains how much rerouting is available in busy airspace like the North Atlantic.

The analogy: choosing which cloud lane to drive in

Picture the sky as a stack of lanes. Most lanes are dry; if you drive through them, your exhaust disappears. A few lanes are so humid that anything you emit turns into a long cloud that hangs around all afternoon.

You can’t see which lane is which from the cockpit. But if someone gives you a map of the humid lanes, you can change lanes for twenty minutes, use slightly more gas, and avoid making the cloud at all.

That’s the entire strategy, and it’s unusually cheap as climate interventions go, because the lane change costs fuel only on the small fraction of flights that would have hit a humid lane.

When a contrail forms, persists, or vanishes

Condition at cruise altitude	Contrail forms?	Does it persist?	Climate consequence
Warmer than about −36.5 °C	No	n/a	None
Cold, and air is dry	Briefly	No, evaporates in seconds	Negligible
Cold, and air is near ice saturation	Yes	Short-lived, minutes	Small
Cold, and air is ice-supersaturated	Yes	Yes, hours, spreads into a sheet	The main warming case
Ice-supersaturated layer, but flight diverted 2,000 ft	Avoided	n/a	Small extra CO2 from fuel, no contrail
Night flights over ice-supersaturated air	Yes	Yes	Worse: no sunlight to reflect, only heat trapped

The last row is a genuinely non-obvious detail worth telling a kid. A daytime contrail at least reflects some sunlight back to space. A nighttime contrail only traps heat.

How to Teach Your Kid About Contrails and AI Rerouting

Ages 5–8: Breath on a Cold Day

On a cold morning, have your kid breathe out and watch the cloud. Ask why they can see their breath outside but not indoors. Same warm humid air, different surrounding temperature. Then point at a contrail in the sky: that’s a plane’s breath, and the sky up there is much colder than the coldest day here. Two minutes, no equipment, and the physics is exactly right.

Ages 9–12: The Persistence Log

Over two weeks, have your kid watch contrails and record whether each one disappears within a minute or hangs around and spreads. Note the date and whether the day was humid or dry. They’ll discover that contrail behavior clusters: some days everything vanishes, some days the sky fills with lines. That clustering is atmospheric humidity, and noticing it is the same insight the forecasting model needs.

Ages 13+: Track Flights and Compare Sky to Data

Use a free flight tracker like FlightAware or an open ADS-B site to identify which aircraft made a contrail your teen photographed, including its altitude. Then pull that day’s upper-air sounding from the NOAA/University of Wyoming sounding archive and look at temperature and humidity at that altitude. Ask: does the sounding explain whether the contrail persisted? This is real attribution work, done with free public data, and it’s exactly the step-2 problem in the pipeline above.

The question to ask: “If avoiding a contrail costs extra fuel, how would you decide whether it was worth it?”

What to actually do at home

Use it to break the “climate equals CO2” simplification

Aviation is the cleanest example available: only slightly over half of its warming comes from CO2, and contrails are about 35% of the impact. Kids who learn that climate has multiple physical mechanisms reason better about every subsequent policy argument.

Make the tradeoff explicit, not moral

More fuel, fewer contrails. CO2 lasts centuries; a contrail lasts hours. That’s a real quantitative tradeoff with genuine uncertainty in it, and treating it as an arithmetic question rather than a virtue question is how engineers actually think. Our piece on AI in airplane autopilot systems covers the adjacent question of what else is automated in the cockpit.

Point out that measurement made the result believable

The 54% figure is credible specifically because contrails were detected from satellite independently of the model that predicted them. Whenever your kid reads a claim about an intervention working, ask how the outcome was measured. That habit is worth more than any single fact.

Connect to the weather-model story

The whole system depends on forecasting humidity at altitude, which is the same class of problem that WeatherNext 2’s ensemble forecasting addresses. Better atmospheric forecasts make contrail avoidance more practical, which is a nice illustration of research compounding.

What not to do

Don’t let this become a conspiracy conversation. Contrails are water ice condensing on soot in cold air, a well-documented process since the 1940s, photographed from satellites, and now labeled by hand for machine learning training sets. The interesting story is the measurement and the tradeoff, and that story is more than interesting enough.

What to Watch For Over the Next 3 Months

• Week 4: Your kid can explain why some contrails vanish and others spread, using the word humidity.
• Month 2 red flags: They treat all contrails as equally bad. Only the persistent ones matter, and distinguishing them is the whole point of the forecasting work.
• Month 3 self-check: Ask what would make the 54% result stronger evidence. If they say more flights, or a longer trial, or a control group, they’ve understood what a small sample means.

Frequently Asked Questions

What is Operation Blue Skies?

A partnership announced by Google with the UK Government and aviation industry leaders that uses AI-powered forecasts to help flight crews and air traffic controllers adjust routes to avoid forming contrails. As of August 2026 it is being expanded across the North Atlantic, one of the world’s busiest and most contrail-prone air corridors.

Are contrails really 35% of aviation’s climate impact?

That figure comes from recent IPCC analysis as cited in Google’s contrails research, and the companion point is that only slightly over half of aviation’s warming comes from CO2. Contrail warming estimates carry more uncertainty than CO2 estimates, which is a real caveat, but the conclusion that non-CO2 effects are large is well established.

Doesn’t burning more fuel defeat the purpose?

It’s a genuine tradeoff and the numbers are what decide it. Trial flights used about 2% more fuel, which is roughly 0.3% across a fleet because only a small fraction of flights need to divert. Whether that is a net win depends on contrail warming estimates, which is precisely why measurement-based trials matter more than modeling alone.

How does AI detect a contrail in a satellite image?

A computer vision model was trained on hundreds of hours of satellite imagery in which humans labeled which streaks were aircraft contrails rather than natural cirrus. Once trained, the model finds contrails in new imagery, within 30 minutes of formation, which is fast enough to link them to specific flights.

Can my family see this happening?

Yes, with your eyes. On some days contrails vanish within seconds; on others they spread into a haze that covers the sky by afternoon. That difference is the difference between dry and ice-supersaturated air aloft, and once a kid can spot it, they are reading the upper atmosphere from the ground.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Google Research. “Project Contrails” (American Airlines trial results, detection methodology, IPCC context). https://sites.research.google/contrails/
2. Google. (2026, August). “Google AI updates, August 2026” (Operation Blue Skies with the UK Government, North Atlantic expansion). https://blog.google/innovation-and-ai/technology/google-ai-updates-august-2026/
3. Contrail formation physics, temperature thresholds, and radiative forcing estimates. https://en.wikipedia.org/wiki/Contrail
4. NOAA / University of Wyoming. Upper-air sounding archive (temperature and humidity at altitude). https://weather.uwyo.edu/upperair/sounding.html
5. NOAA National Weather Service. GOES satellite imagery and products. https://www.weather.gov/
6. Google DeepMind. “WeatherNext 2” (atmospheric forecasting that underpins humidity prediction). https://blog.google/technology/google-deepmind/weathernext-2/