The prediction of the atmospheric and turbulence conditions are of great interest for the astronomical community and free space optical telecommunications. With the advent of the next generation of ELTs, the knowledge of atmospheric conditions several hours prior to observations has become essential. Besides the importance of reducing the lost in cost of observations due to bad atmospheric conditions, it is important to specify a site that provides an optimal optical quality. Therefore, it is crucial to overcome the effects of atmospheric turbulence. In general, the challenge is to find ways to reduce the effects of turbulence on the optical beam propagating through the atmosphere. Adaptive Optics (AO) methods aim to reduce these effects. However, the use of AO alone cannot overcome all the effects of turbulence. Moreover, these AO corrections are more efficient if the predicted conditions are more advantageous, for both prediction and performance optimization. Hence, emphasizing the importance of developing a robust and efficient tool to predict atmospheric turbulence conditions a few hours in advance in order to optimize the planning of astronomical observations called "flexible scheduling".

As well, in the field of free space optical telecommunications, the propagation of optical signals in the atmosphere is significantly influenced by weather conditions (clouds, fog, rain, etc.) and optical turbulence, which can cause signal losses. Optical ground stations must be installed at the most favorable sites, and it is necessary to have a tool to predict the best locations as well as the most favorable periods for laser links (transmission/reception).

Based on the work that have been developed previously in our team, a numerical approach based on the Weather and Research Forecasting (WRF) model coupled with different optical turbulence models has been used. The results were compared to in-situ measurements and optical measurements from the Calern Atmospheric Turbulence Station (CATS). In order to improve the statistical model, an optimization of the prediction has been performed using local measurements to improve the turbulence model and better take into account the local specificities of a given site. This method known as "site learning", has been tested at the Calern Observatory site, France and has brought improvement to the predictions results. On the other hand, a new estimation method of the outer scale of turbulence from the profile of meteorological data has been developed to improve the theoretical model. This method has been applied to the Cerro Pachon Observatory site, in Chile. The results show good agreement with the measurements.

Until now we have used weather forecast tools coupled with turbulence models to predict turbulence conditions. An approach using machine learning algorithms such as the Regression Ridge (ARR) and the Random Forest Algorithm (AFA), was also tested and showed good results on predictions. The goal is to build new models using advanced Machine Learning algorithms such as LSTM(Long short-term memory), to enhance the short term prediction.