image: Graphic of an aircraft encountering microbursts during takeoff and approach. Generally an aircraft would first be exposed to increasing head wind which brings higher airspeed and leads to increased lift; subsequently the aircraft would suffer from an abrupt strong tailwind, which causes an undesired airspeed drop and results in dangerous stall, loss of balance, and loss of control, until it escapes from areas affected by low-level wind shear.
Credit: Ji Song et al.
Wind shear, a sudden change in wind speed or direction, is a major cause of aviation incidents; it was responsible for 18% of aviation accidents in 2022. Predicting wind shear events is a priority for aviation safety, as it would allow pilots to avoid areas where wind shear is likely. Currently, aircraft-based wind shear detection relies on the F-factor, an index that captures current wind speed and direction, as well as current aircraft speed. But the F-factor cannot predict future wind shear events. Xiaowei Yue and colleagues propose a physical-mechanism-aided, transformer-based model capable of producing reliable predictions of wind shear. The model was trained on 19 key parameters from the NASA DASHlink Sample Flight Dataset, primarily from mechanical, power, and control systems of an aircraft, as well as the external flight environment. When tested on real-world in-flight datasets, the model enabled pilots to have at least 15 seconds of warning before potential wind shear risks. The model’s outputs had deviations from real outcomes within 5% across all forecast horizons. According to the authors, the study suggests that machine learning, combined with physical measurements, can improve aviation safety.
Journal
PNAS Nexus
Article Title
A mechanism-aided transformer may transform in-flight aviation safety
Article Publication Date
9-Jun-2026