Forecasting the atmospheric refractive index structure constant profile with an altitude-time correlations-inspired deep learning model

Abstract

An accurate forecast of the atmospheric refractive index structure constant (Cn2) is vital to analyzing the influence of atmospheric turbulence on laser transmission in advance. In this paper, we propose a novel method to forecast the atmospheric refractive index structure constantCn2profile, which is inspired by the turbulence characteristics (i.e., the altitude-time correlations). A deep convolutional neural network (DCNN) is adopted in the hope that with the stacked convolutional layers to abstract the altitude-time correlations ofCn2, it can accurately forecast theCn2profile in the near future based on the accumulated historical measurement data. While the sliding window algorithm is introduced to segment the measured time series data of theCn2profiles to generate the input-output pair data for training and testing. Experimental results demonstrate its high forecast accuracy, as the obtained root mean square error and the correlation coefficient are 0.515 and 0.956 in the one-step-aheadCn2profile forecast case, 0.753 and 0.9046 in the 36-step-ahead forecast case, respectively. Moreover, the forecast accuracy versus altitude and its relationship with the distribution ofCn2against altitude are analyzed. Most importantly, with a series of experiments of various input feature sizes, the appropriate sliding window width forCn2forecast is explored, and the short-term correlation ofCn2is also verified.