COVID-Contrail: Automated Contrail Detection through Probabilistic Deep Learning Model

Contrails potentially influence the Earth’s energy balance by altering the radiative forcing (i.e., heating or cooling), particularly in middle latitudes where a cold and sometimes moist upper-troposphere (UT) intersects the mean jet-flight path altitude. Such contrail radiative forcing should be amplified when multiple occurrences of these anthropogenic clouds are grouped spatio-temporally as ‘contrail outbreaks’ and contrail cirrus on at least 1 × 103 km2, extending over approximately 104~105 km2, and persisting for approximately 1-6 h. To more definitively determine the net radiative forcing of contrails and their role in contemporary and future climate changes, and to establish appropriate climate change mitigation strategies (e.g., jet flight activity regulation), it is essential to reliably detect contrail incidence and assess the spatio-temporal distribution of contrail outbreaks as well as understand the underlying mechanisms of contrail formation.

Thanks to recent advanced technologies, one can obtain flight tracking data as well as automate an individual contrail occurrence detection on the satellite or aerial images and alleviate the burden of labor-intensive manual contrail detection tasks. In this context, our COVID-Contrail project aims to (1) assess the changes of contrail occurrence patterns and relevant atmospheric factors in the continental United States (CONUS) before and during COVID-19 and (2) develop an automated data processing and analysis system to construct and publish a massive contrail incidence database in high spatio-temporal resolution. This project is expected to make significant advances in diverse aspects of contrail and climate studies. In the poster presentation, we will introduce our COVID-Contrail project and present work in progress of our project.