Application of Neural Networks to Estimation and Prediction of Seeing at the Large Solar Telescope Site

Abstract

Abstract Optical turbulence limits the angular resolution of ground-based astronomical telescopes. The key parameter of optical turbulence is seeing. In this study, seasonal variations of seeing estimated from differential image motion monitor measurements at the Large Solar Telescope site are discussed. The Large Solar Telescope will be located at an elevation of 2000 m above sea level ( 51 ° 37 ′ 18 ″ N, 100 ° 55 ′ 07 ″ E ). The highest seeing values are observed in winter. The median of seeing is 2.″1. In summer, the median decreases to 1.″1. The best atmospheric conditions are observed in April–May, when the medians of seeing are low and the standard deviations are high. During this period, atmospheric situations with low values of seeing (∼0.″5–0.″6) are often observed. We simulated multilayer neural networks for the measured seeing by applying a group method of data handling. Modeled seeing is well described in terms of mean meteorological parameters, which include wind speed components and large-scale vorticity of air flows at different altitudes in the atmosphere. The 12-layer optimal neural network obtained has a high correlation coefficient between modeled and measured seeing values. The linear correlation coefficient is 0.77.