Optical Turbulence Profile Modeling in the Atmospheric Boundary Layer: A Random Forest Regression Approach

Abstract Imaging and targeting missions parallel or at an angle to the surface are impacted by spatial variations in the refractive index structure function coefficient in the atmospheric boundary layer. However, measuring and modeling in both the vertical and transverse directions in the atmospheric boundary layer remains difficult. Using a suite of atmospheric and surface measurements available through the DOE Atmospheric Radiation Measurement Facility Southern Great Plains near Lamont, Oklahoma, including radiosonde soundings, high‐acquisition‐rate meteorological towers, and an eddy covariance tower, we investigate the influence and importance of surface and atmospheric measurements on within the atmospheric boundary layer. Over a three‐year period from Apr 2020—Apr 2023, corresponding to 3,000 soundings, vertical profiles of in the boundary layer and surface‐based obtained from the meteorological towers are correlated and classified by atmospheric stability and surface fluxes. Vertical profiles are decomposed into an average profile, based on the Hufnagel‐Valley model, and a fluctuation profile, based on sounding statistics. Random forest regression is used to aggregate the Hufnagel‐Valley model parameters with surface measurements trained with surface measurements corresponding to soundings. A fourth year, Apr 2023—Apr 2024, of data is used for testing and validation. The most important surface parameters include boundary layer height estimates and surface measurements. Unstable conditions are more influenced by sensible energy surface fluxes than stable conditions. The random forest enables the modeling of in the atmospheric boundary layer based on surface measurements providing profiles at faster intervals between soundings.