CleanSky

ML-Powered Predictions to Minimize Contrails

Problem & Motivation

As you look up in the sky, those mesmerizing white streaks formed by aircraft engine exhaust interacting with the atmosphere, have long fascinated us with their beauty. However, beyond their aesthetic allure, these trails represent a significant environmental concern. They act as artificial clouds, trapping heat within the Earth's atmosphere, intensifying the greenhouse effect, and aiding global warming and climate change. As the aviation industry continues to boom, the proliferation of contrails poses an ever-growing threat to the delicate balance of our planet's climate system, urging us to explore sustainable solutions for air travel and its environmental impact.

At CleanSky, our mission is to empower ARPA-e-funded companies with cutting-edge B2B software, enhancing their devices' capabilities for in-flight contrail prediction. We are committed to revolutionizing contrail prediction by providing indispensable insights into contrail severity, leveraging advanced algorithms and seamless integration of atmospheric data, thereby enabling our clients to navigate the skies with unprecedented accuracy and foresight.

Our CleanSky Contrail Severity Model (CSM) leverages machine learning algorithms and real-time atmospheric data to revolutionize contrail forecasting. We're hoping for positive change by enabling the prediction, interpretation, and navigation of contrail forecasting with unparalleled precision, ensuring safer and more informed skies for all.

Data Source & Data Science Approach

Through our communication with Dr. Roger Teoh (Faculty of Engineering at Imperial College London), our team was able to receive a parquet data file of recorded flight data that he used in his contrail research project. This dataset contained data on atmospheric weather conditions as well as predicted contrail formation forecasts. Using this dataset we were able to extract meaningful insights on the types of atmospheric conditions that resulted in the formation of a persistent contrail.

At a high level, our clients would feed live, onboard recorded atmospheric and contrail data to our CMS model which would output real-time predictions of contrail length and age. It would then be in the pilot's hands to decide whether or not to re-route the aircraft to airspace with a lower likelihood of creating a contrail.

Evaluation

Our model evaluation strategy progressed from simple to complex, starting with linear regression and advancing through neural networks and ensemble methods like Random Forest and XGBoost.

The accuracy of our models was quantitatively assessed using R-squared, RMSE, and MAE, with XGBoost emerging as the top performer for predicting contrail age and persistent contrail length. This approach allows for real-time, environmentally conscious adjustments to flight routes by forecasting contrail characteristics with relatively high precision.

Key Learnings & Impacts

Throughout this capstone adventure, our team delved into the intricate relationship between atmospheric conditions and contrail forecasting. Together, we gained valuable insights into the complexities of atmospheric data analysis, honed our skills in model development and optimization, and discovered the critical roles of feature selection and data wrangling for accurate predictions. 

Beyond bolstering our technical proficiency, Project CleanSky has illuminated the potential for proactive measures in aviation sustainability, paving the way for more informed decision-making regarding flight routes and emissions reduction. We're excited to now realize that there is so much more that can be done at the intersection of meteorology and data science to further shape a more efficient and eco-conscious future for aviation.

As our semester comes to a close, we envision plenty of model improvements via parameter discovery, data supplements, and more. Some other use cases for contrail forecasting that have come up this semester (besides overall global warming effects) include the following...

1. Military and Air Force: a reduction in contrail formation on fighter jets could provide an extra sense of stealth allowing for more discrete missions
2. Air Traffic Management: contrail forecasting can assist in optimizing flight paths to minimize congestion and delays
3. Public Health and Safety: forecasting can inform policymakers to develop regulations for sustainable aviation practices, ensuring public health and safety
4. Agricultural Practices: contrails could affect local weather conditions, so forecasting could help agricultural planning and disaster preparedness

Acknowledgments

The CleanSky team would like to thank our Capstone professors Puya Vahabi and Danielle Cummings for guiding us through our endeavors this semester. Our vision would not have been possible without your support, feedback, and industry knowledge.

We would also like to thank the following people for contributing to Project CleanSky.

• Dr. Roger Teoh || Faculty of Engineering at Imperial College London
• Dr. Scott Geraedts || Physics at Princeton and Climate & Energy Research at Google
• Dr. Lembit Salasoo || Senior Engineer, Analytics at GE Global Research
• Our teams' amazing UC Berkeley MIDS professors from past semesters