Google has conducted research, in partnership with Breakthrough Energy and American Airlines, on one of the lowest-cost climate solutions to aviation emissions, powered by AI. This research is a significant proof point that commercial airlines can verifiably avoid contrails and thereby reduce their climate impact.

Aviation is one of the hardest-to-decarbonize sectors in our transport infrastructure. It accounts for around 2.5% of global CO2 emissions, but the industry contributes closer to 3.5% of global warming. According to IPCC’s Sixth Assessment Report (AR6), contrails are responsible for a disproportionately large share – approximately 35% – of aviation’s global warming impact.

The good news is that planes can fly without creating contrails, but the challenge is knowing the routes that will create them. A team at Google Research learned of the impact of contrails and realized that they might be able to identify contrail-forming regions by training a machine learning algorithm to detect contrails in satellite imagery.

Google is committed to sustainability. One of the ways we invest in sustainability is by collaborating with our partners to explorate AI-enabled climate solutions. Our contrails research is an example of collaborating with our partners - American Airlines and Breakthrough Energy - to see how AI could enable climate-beneficial action.

The core technical capabilities include computer vision algorithms and large scale data processing. The solution requires processing large amounts of satellite, weather, and flight data. We applied the same technology that powers Waymo’s self-driving car to identify clouds from satellite imagery.

Contrails constitute ⅓ of aviation’s warming effect. Mitigating contrail formation through altitude changes has been discussed in academic literature as a potential cost-effective climate solution. Our goal was to test in a world-first real-world demonstration that verifiable contrail reduction is possible. We created a substantial dataset of labeled contrail images to use in developing precise models to guide real flights and detect whether or not they created contrails. We were able to demonstrate a 54% reduction of contrail creation using our approach.

We started working with American Airlines and Breakthrough Energy in late 2021, and we took time to develop the operational protocols and ensure the pilot and passenger experience would in no way be negatively affected. We flew our first flights in 2022 and completed our proof of concept earlier in 2023.

A group of pilots at American Airlines then flew 70 test flights over six months using the AI-based predictions. After these test flights, the team analyzed satellite imagery and found that the predictions reduced contrails by 54% compared to when pilots didn’t use the 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.

Using AI-based image recognition technology and satellite imagery, Google identified contrail forming regions and made slight adjustments to traditional flight paths that would avoid those areas.

Similar to the machine learning algorithms that are trained to pick out a cat in your photos, the team at Google Research spent hours training algorithms to identify contrails in satellite imagery. From there, they were able to predict ‘contrail likely zones’ for future flights and provide recommendations on what upper atmosphere zones planes should avoid when flying routes. The core technical capabilities include computer vision algorithms and large scale data processing. The solution requires processing large amounts of satellite, weather, and flight data.

The key cross-organization was the ability to safely integrate these new AI-based insights into the pilot’s workflow.

Using AI-based image recognition technology and satellite imagery, we identified contrail forming regions and made slight adjustments to traditional flight paths that would avoid those areas. After running live trials with pilots as they flew domestic American Airlines routes, we achieved our first proof point demonstrating verifiable, statistically significant, scalable and cost-effective contrail avoidance. A key accomplishment of this collaboration was the ability to safely integrate these new AI-based insights into the pilot’s workflow.

The work catalyzed investment in developing contrail avoidance as a scaled, cost-effective climate solution. The work catalyzed interest in contrails across the aviation industry, and the video has been shown from boardrooms to conferences. The video was shared with executives across the aviation industry and accelerated industry investment in contrails research. We plan to continue working with more aviation partners to scale the technology industry-wide.