Sensitivity Analysis of Mesoscale-Coupled Large Eddy Simulations for Wind Energy Applications Using Machine Learning Approaches

Abstract:

Wind farm performance is influenced by wide-ranging scales of atmospheric motion, spanning from mesoscale variability to fine scales of turbulence. This presents a challenge for flow simulation, and high-resolution large eddy simulations (LES) over limited area domains have often represented the larger-scale flow in an idealized way. However, there is growing recognition of the importance of imposing more realistic large-scale forcing and the need to couple mesoscale and microscale simulations for wind energy applications. While the sensitivity of large eddy simulations of the atmospheric boundary layer to turbulence closures has been an active area of research for decades, prior studies have mostly considered relatively idealized scenarios. Here we investigate the uncertainty associated with turbulence closure parameters for more realistic LES performed using the Weather Research and Forecasting (WRF) model nested from horizontal resolutions of a few kilometers down to tens of meters. We vary closure parameters to generate simulation ensembles for two case studies of highly sheared, convective boundary layers observed in the Columbia Basin of Oregon and Washington during the Wind Forecast Improvement Project II (WFIP2). Machine learning techniques are used to explore the parameter sensitivity of key output variables while accounting for nonlinear model responses and interactions between parameters.