面向星地激光通信的大气湍流预报研究进展 (特邀)

Significance The prediction of atmospheric turbulence has great significance both in science and engineering, which provides key parameters and references for domains like astronomical observation, site selection, satellite-ground laser communication, and remote sensing. Especially in satellite-ground laser communication, predicting key parameters of atmospheric turbulence can schedule satellite-ground data transmission links in advance, and pre-deploy adaptive optical schemes to compensate turbulence effects, so as to establish effective communication links and suppress the performance degradation of data transmission. Therefore, atmospheric turbulence prediction is crucial and become an important issue, which needs to be addressed for most of laser scenarios in atmosphere. Progress This review consists of three sections. In the first section, firstly, the widely used meso-scale numerical prediction scheme to forecast atmospheric turbulence is introduced in detail. This scheme is accomplished by turbulence parameterization schemes, which establishes the relationship between the turbulence characteristics and the conventional meteorological parameters output from mesoscale meteorological model. Mesoscale meteorological model has been well developed, the most representative models include Meso-Nh(Nonhydrostatic mesoscale atmospheric model), MM5(Mesoscale Model 5), WRF(Weather Research & Forecasting Model) and Polar WRF. Many achievements have been made in turbulence parameterization schemes, including Hufmagel model, Tatarski model. Then, the relevant work of using mesoscale numerical prediction method to forecast atmospheric turbulence in typical regions is reviewed. The second section presents recent advances regarding deep learning in atmospheric turbulence prediction, and discusses its advantages and limitations. This section first introduces the research achievements of deep learning in meteorological forecasting, and then introduces the research advances of deep learning in atmospheric turbulence forecasting. Based on a large amount of data, deep learning scheme can establish a relationship between the input data and the target label without any prior formula. In atmospheric turbulence prediction, deep learning is used to establish the relationship between meteorological parameters and atmospheric turbulence parameters, but the prediction accuracy is also limited by the accuracy of meteorological parameters. In the third section, a short-time atmospheric coherence length prediction method called TsVMD-AR is introduced. TsVMD-AR model uses VMD (variational mode decomposition) algorithm and AR (autoregression) algorithm to forecast the short-term atmospheric coherence length. This scheme reduces the interference and coupling between the multi-scale feature information in the dataset, makes the complex internal features of the dataset easier to obtain. The results show that the established TsVMD-AR model is obviously superior to other models and is suitable for daily atmospheric turbulence prediction. Prospects We hope this review will provide more valuable information for people who is working in scenarios of laser applications in atmosphere turbulence, and inspire more wonderful ideas towards abilities of more accurate and faster turbulence grasp.