Selection of features for an image-based machine learning model to predict atmospheric optical turbulence

Direct measurements of atmospheric effects on light propagation often require equipment and access that are unavailable in remote or complex environments. In these situations imaging data may be able to provide estimates of atmospheric turbulence levels suitable for performance predictions of laser-based systems. To select the most significant image features, a supervised machine learning model is developed using partial-reference image data and scintillometer-based measurement of atmospheric optical turbulence via the index of refraction structure parameter, C_n². Both the images and the scintillometer data come from a 1-km over-water path adjacent to the Chesapeake Bay. The specific image features identified with the partial-reference model will then be used to develop machine learning models for atmospheric optical turbulence using no-reference image data.