Project Contrails

How do contrails contribute to global warming?

Contrails can persist as cirrus clouds for minutes or hours, depending on the conditions. Normally, the side of the Earth not facing the sun releases heat overnight. At night natural clouds and airline contrails trap heat.  During the day, they also reflect back incoming sunlight and warmth.  Nighttime contrails are more warming than daytime contrails because they do not reflect incoming sunlight and only trap heat, creating a net warming effect. A recent IPCC report noted that clouds created by contrails account for roughly 35% of aviation's global warming impact.

Using Google AI to help airlines prevent contrails

Planes don’t always make contrails. Contrails only occur when planes fly through humid regions.  If planes intentionally avoid flying through these regions, then they can avoid creating warming contrails, with minimal impact on fuel usage.

The challenge is knowing where these regions will occur. By combining massive quantities of weather data, satellite data, and flight data, AI can create state-of-the-art predictions of when and where contrails are likely to form. Pilots and dispatchers can then use this information to adjust the altitudes of their flights. 

Our team developed an approach to leverage satellite imagery and computer vision to detect contrails and attribute them to associated flights, based on hundreds of hours of labeling tens of thousands of satellite images. The image below shows our AI detecting contrails over the United States, based on satellite imagery.

Testing with American Airlines

A group of pilots at American Airlines flew 70 test flights over six months using our AI-based predictions to avoid routes that created contrails. After these test flights, we analyzed satellite imagery and found that the predictions reduced contrails by 54% compared to when pilots didn’t use our predictions. We also saw that flights that avoided contrails burned 2% more fuel, which would translate to 0.3% more fuel when scaled across an airline’s fleet. (This is because not all flights make contrails, and with AI predictions, only a fraction of flights need to be adjusted.)  Together, this suggests contrail avoidance costs could be in the range of $5-25/ton CO2e, which would make it one of the most cost-effective climate solutions.

Together with American Airlines, we achieved the industry’s first proof point that commercial airlines can verifiably avoid contrails and thereby reduce their climate impact. Read more about our work with American Airlines.

Watch the story of this research

How does it work?

1. Labeled data: Our team of engineers analyzed satellite imagery and spent hundreds of hours labeling tens of thousands of contrail images collected by the geostationary satellite, GOES-16.

2. Train a state-of-the-art computer vision model: We used labeled data to train a state-of-the-art computer vision model to detect contrails when they are formed.  The detection model can identify contrails that are visible within GOES imagery in a matter of 30 minutes. 

3. Combine large datasets: We combined our computer vision model with large-scale weather data, flight data, and satellite images to develop a more accurate prediction model.

4. Integrate AI insights into pilot workflow: We partnered with American Airlines to integrate contrail likely zone predictions into the tablets that their pilots used in flight, so they could make realtime adjustments in altitude, just as they do to avoid turbulence, to avoid creating contrails.

5. Analyze performance results: We evaluated the prediction model's performance using satellite imagery, comparing the number of contrails produced in flights where pilots used predictions to avoid contrails to contrails created in flights where pilots didn’t use contrail predictions. For our first proof point with American Airlines, we saw a reduction of 54%.

In the future, we will extend our models to geostationary satellites over Europe, Africa, the Indian Ocean (Meteosat Third Generation), East Asia and Western Australia (Himawari).

Read more

A human-labeled Landsat-8 contrails dataset, Kevin McCloskey, Scott Geraedts, Brendan Jackman, Vincent R. Meijer, Erica Brand, Dave Fork, John C. Platt, Carl Elkin, Christopher Van Arsdale

OpenContrails: Benchmarking Contrail Detection on GOES-16 ABI, Joe Yue-Hei Ng, Kevin McCloskey, Jian Cui, Vincent R. Meijer, Erica Brand, Aaron Sarna, Nita Goyal, Christopher Van Arsdale, Scott Geraedts

Estimates of broadband upwelling irradiance from GOES-16 ABI, Kevin McCloskey, Sixing Chen, Vincent R. Meijer, Joe Yue-Hei Ng, Geoff Davis, Carl Elkin, Christopher Van Arsdale, Scott Geraedts

A scalable system to measure contrail formation on a per-flight basis, Scott Geraedts, Erica Brand, Thomas R. Dean, Sebastian Eastham, Zebediah Engberg, Ulrike Hager, Ian Langmore, Kevin McCloskey, Joe Yue-Hei Ng, John C. Platt, Tharun Sankar, Aaron Sarna, Marc Shapiro, Nita Goyal

Also check out our Kaggle competition for detecting linear contrails from satellite imagery.

Contact us

If you are a research-focused organization interested in a novel research partnership with the contrails team or interested in exploring our contrail predictions through the Trusted Tester Program, please fill out this form.