Can machine learning improve the model representation of turbulent kinetic energy dissipation rate in the boundary layer for complex terrain?

• Published in: Geoscientific Model Development 2020-09, Vol.13 (9), p.4271-4285
• Main Authors: Bodini, Nicola, Lundquist, Julie K, Optis, Mike
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Alternative energy sources; Analysis; Anemometers; Atmospheric boundary layer; Atmospheric models; Aviation; Bias; boundary layer: complex terrain; Boundary layers; Data mining; Energy (Physics); Energy dissipation; Energy exchange; Energy industry; GEOSCIENCES; Kinetic energy; Kinetic energy dissipation; Learning algorithms; Machine learning; model representation; Numerical weather forecasting; Numerical weather prediction; Parameterization; Precipitation; Prediction models; Representations; Sample variance; Sonic anemometers; Terrain; TKE dissipation rate; Turbines; Turbulence; Turbulence (Fluid dynamics); Turbulence measurement; Turbulent kinetic energy; Weather forecasting; wind energy
• ISSN: 1991-9603; 1991-959X; 1991-962X; 1991-9603; 1991-962X
• DOI: 10.5194/gmd-13-4271-2020

Current turbulence parameterizations in numerical weather prediction models at the mesoscale assume a local equilibrium between production and dissipation of turbulence. As this assumption does not hold at fine horizontal resolutions, improved ways to represent turbulent kinetic energy (TKE) dissipation rate (ϵ) are needed. Here, we use a 6-week data set of turbulence measurements from 184 sonic anemometers in complex terrain at the Perdigão field campaign to suggest improved representations of dissipation rate. First, we demonstrate that the widely used Mellor, Yamada, Nakanishi, and Niino (MYNN) parameterization of TKE dissipation rate leads to a large inaccuracy and bias in the representation of ϵ. Next, we assess the potential of machine-learning techniques to predict TKE dissipation rate from a set of atmospheric and terrain-related features. We train and test several machine-learning algorithms using the data at Perdigão, and we find that the models eliminate the bias MYNN currently shows in representing ϵ, while also reducing the average error by up to almost 40 %. Of all the variables included in the algorithms, TKE is the variable responsible for most of the variability of ϵ, and a strong positive correlation exists between the two. These results suggest further consideration of machine-learning techniques to enhance parameterizations of turbulence in numerical weather prediction models.