Estimation and forecasting of optical turbulence by physical model-based and machine learning-based approaches

Accurate estimation and forecasting of optical turbulence (typically quantified by Cn2) in our atmosphere is of great significance for both civil and military applications. Cn2 values can be measured by various types of research-grade instruments (e.g., scintillometers, thermosondes). However, due to logistical and financial issues, these types of instruments, and associated data, are not widely available. In the absence of observational data, physical model-based approaches are often utilized for the estimation of Cn2. The physical models (e.g., mesoscale models, large-eddy simulations) numerically solve the conservation equations for mass and momentum (known as the Navier-Stokes equations). They also solve the thermodynamic energy equation and the conservation equations for various phases of water (e.g., water vapor, cloud water). Most of the relevant physical processes (e.g., turbulence, radiation, microphysics, land-atmosphere interactions) are parameterized in these models. In order to diagnose Cn2 from the explicitly modelled meteorological variables (such as wind speed, temperature, turbulent kinetic energy) various schemes have been developed over the years. Some of these schemes make use of simple regression equations, while others employ higher-order turbulence parameterizations. During this presentation, we will discuss the strengths and weaknesses of these schemes with varying complexity. In addition, we will elaborate on the sensitivities of the simulated results with respect to spatial resolution, initial and boundary conditions, and turbulence parameterization. We will briefly touch upon the future of optical turbulence forecasting research in which state-of-the-art machine learning approaches (e.g., gradient boosting machines) coupled with physical models will play vital roles.