[A-AS06] Atmospheric (Stratosphere-troposphere) Processes And their Role in Climate

[AAS06-08] A Convolutional Neural Network for the Detection of Gravity Waves in Satellite Observations and Numerical Simulations

*Haruka Okui¹, Corwin Wright¹, Peter Berthelemy¹, Neil Hindley¹, Andrew Barnes¹ (1.University of Bath)

Keywords:Atmospheric gravity waves, Machine learning, Satellite observations, General circulation model

Observation-model comparisons of atmospheric gravity waves are crucial for evaluating the accuracy of general circulation model (GCM) simulations particularly in the middle atmosphere and for comprehensively understanding gravity wave characteristics. However, observational noise often obscures these waves, complicating such comparisons. To address this issue, we developed a gravity wave detection method using a convolutional neural network (CNN) for semantic segmentation. The CNN is trained on temperature measurements from the Atmospheric Infrared Sounder (AIRS) with labels indicating the presence or absence of waves based on the detection method proposed by Berthelemy et al. (2025, in review). Their original approach relies on detecting discontinuities in horizontal wavelengths caused by observational noise. In contrast, the CNN provides consistent results even when applied to smoothly-varying model data. Using this method, we conduct a multi-year comparison of stratospheric gravity waves in boreal winters between AIRS observations and a high-top gravity-wave-permitting GCM, JAGUAR.