Short-term atmospheric turbulence prediction in a satellite-to-ground laser link with limited observed data

Satellite-to-ground laser communications can download massive amounts of data from satellites, thus playing an important role in space science. However, the performance of data downloading will decrease severely if there is stronger atmospheric turbulence (AT) in the link. Therefore, predicting AT strength in advance is crucial, as it will provide efficient strategies for data transmission and for following adaptive optical systems. Current schemes based on machine learning and deep learning typically require large amounts of training data, which demands substantial manpower and computational resources. This paper proposes a hybrid model to predict short-term AT strength, represented by the Fried parameter r₀, with limited historical r₀ data. We capture r₀ practically by using a differential image motion monitor in Miyun, near Beijing, China. Then, we decompose the r₀ series into four feature series to reduce the prediction difficulty, followed by an auto-regressive integrated moving average model to predict each feature series. Finally, the predicted r0 is obtained by summing up the predicted modes. The favorable results indicate that our proposal can predict r₀ for the next two hours with an average mean absolute percentage accuracy of 82.10% using only six days of historical data. To the best of our knowledge, this work achieves the shortest historical data requirement for AT strength prediction and effectively promotes related research and applications.