Atmospheric turbulence has a significant impact on airplane motions. It can induce excessive stress and fatigue damage, shortening the aircraft lifespan. The identification of turbulence in aircraft service phase is of particular interest to estimate actual structural fatigue or to improve maintenance plans. Different methods exist to identify turbulence from in-board aircraft data. A common detection method is the use of the vertical component of air speed with regard to the ground to identify potential turbulence. This method is highly dependant on the incidence angle sensors accuracy, and is considering an identical airspeed over the entire plane. Cornman et al. calculate the aircraft’s response to vertical gusts from its structural characteristics [1]. They use the response function and acceleration measurements to estimate the intensity of turbulence, approximated by a Von Kármán spectrum . Li et al. proposed a method to detect turbulence using an isolation forest algorithm on aircraft sensor data [2]. Turbulence can indeed be investigated using anomaly detection techniques, since its occurrence is low in relation to the complete flight data and since it significantly modifies aircraft behavior. In their survey [3], L. Basora, et al. classifie them into the following families of methods. Anomalous data can be detected using clustering, by identifying data that does not belong to any cluster. Distance metric between data samples can help identifying outliers. Statistical methods provide the probability of outcome of a particular data sample. Reconstruction methods and prediction methods rely on a model which reconstruct the data or predict the next sample. These models struggle on samples which differs from the majority of the data, which allow to find anomalies. Recently, deep learning models have shown that they can perform well in detecting anomalies in temporal series, with reconstruction models [4], predictive models [5] and models computing the "reconstruction probability