The Captain Has Turned the Seatbelt Sign On: 43 Years of Reanalysis and AI to Predict Clear Air Turbulence

Clear Air Turbulence (CAT) presents significant challenges for aviation safety, leading to injuries, increased fuel consumption, and operational delays. Recent accidents and an increase in CAT-related incidents have highlighted the growing concern within the aviation industry. This trend may be exacerbated by changing atmospheric conditions, possibly linked to climate change, which is believed to be altering jet stream patterns and increasing the frequency and severity of CAT events. Traditional methods of forecasting CAT rely on analyzing atmospheric parameters; however, the unpredictable nature of CAT calls for more advanced predictive techniques.

In this study, we leverage the power of machine learning, specifically using Long Short-Term Memory (LSTM) networks combined with Convolutional Neural Networks (CNN), to develop CAT indicators based on reanalysis data spanning 43 years. This approach allows us to effectively capture the temporal and spatial patterns critical for predicting CAT, though other models such as Transformers could also be considered for future work. Our approach utilizes the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) dataset to capture essential atmospheric variables at corresponding flight levels (FL) where CAT is more prominent (FL 200 – 300). Although MERRA-2 has a coarse spatial resolution, it provides a long, continuous dataset that captures the larger-scale atmospheric patterns driving CAT. The LSTM component captures temporal dependencies, while the CNN extracts spatial features, enabling a robust analysis of the complex interactions that lead to CAT events. The resulting CAT indicators show promise for improving turbulence prediction, with the potential to enhance flight safety and operational efficiency.