Machine learning for optical turbulence prediction in geographically similar regions

Abstract

The refractive index structure parameter (Cn2) is of interest because it characterizes turbulence, which affects optical propagation through the atmosphere, including free space optical communications, laser propagation, and imaging. This work seeks to develop a geography-agnostic model that can predict Cn2 and received signal strength index (RSSI), with as few input parameters as possible. This work trains models including the Gaussian process regression, neural network, and bagged decision tree types, and use r-squared and root-mean squared error to compare model performance. Most of the data used to train and test the algorithms is collected in San Diego, a Csa-type climate (hot-summer Mediterranean climate) according to Köppen climate classification. We then demonstrate application of the trained models to a different site with similar climate, using available common input parameters, and quantitatively assess each model's respective efficacy.