Abstract
Wavefront distortions caused by atmospheric turbulence significantly reduce the quality of astronomical observations. A key characteristic of atmospheric turbulence intensity is the structural constant of the refractive index Cₙ². Direct measurements of its altitude profile are complex and labor-intensive, which stimulates the development of forecasting methods. In this work, we present a method for estimating the altitude profile of Cₙ² at 12 levels (from 0.5 to 22.6 km) using machine learning. The gradient boosting model was trained based on ground measurements of turbulence characteristics, local meteorological data, and vertical profiles of meteorological variables extracted from the ERA-5 reanalysis. The performance of the resulting model was compared with that of a random forest model. The comparison showed that gradient boosting outperforms the random forest method at most heights, starting from 0.71 km. The greatest improvement in accuracy (up to 7.4 % according to Pearson's correlation coefficient) was achieved at heights above 5 km. Analysis of the importance of features for the best model revealed that the most significant parameters are temperature and wind speed at isobaric surfaces. The results demonstrate the effectiveness of gradient boosting for predicting the Cₙ² profile, which simplifies and speeds up the evaluation of astronomical sites where optical telescopes are located.