Large-Eddy Simulation Sensitivities to Variations of Configuration and Forcing Parameters in Canonical Boundary-Layer Flows for Wind Energy Applications

Abstract

Abstract. The sensitivities of idealized Large-Eddy Simulations (LES) to variations of model configuration and forcing parameters on quantities of interest to wind power applications are examined. Simulated wind speed, turbulent fluxes, spectra and cospectra are assessed in relation to variations of two physical factors, geostrophic wind speed and surface roughness length, and several model configuration choices, including mesh size and grid aspect ratio, turbulence model, and numerical discretization schemes, in three different code bases. Two case studies representing nearly steady neutral and convective atmospheric boundary layer (ABL) flow conditions over nearly flat and homogeneous terrain were used to force and assess idealized LES, using periodic lateral boundary conditions. Comparison with fast-response velocity measurements at five heights within the lowest 50 m indicates that most model configurations performed similarly overall, with differences between observed and predicted wind speed generally smaller than measurement variability. Simulations of convective conditions produced turbulence quantities and spectra that matched the observations well, while those of neutral simulations produced good predictions of stress, but smaller than observed magnitudes of turbulence kinetic energy, likely due to tower wakes influencing the measurements. While sensitivities to model configuration choices and variability in forcing can be considerable, idealized LES are shown to reliably reproduce quantities of interest to wind energy applications within the lower ABL during quasi-ideal, nearly steady neutral and convective conditions over nearly flat and homogeneous terrain.

The FastEddy® Resident‐GPU Accelerated Large‐Eddy Simulation Framework: Model Formulation, Dynamical‐Core Validation and Performance Benchmarks

“…In addition, sensitivity to some numerical settings such as grid resolution and grid cell aspect ratio were investigated with the general conclusion that while sensitivity to choice of model numerics and forcing conditions exist, all idealized LES cases resulted in reasonable agreement with observations. We use here, for the sake of brevity, a single set of forcing conditions from Mirocha et al (2018) which yielded good agreement with observations.…”

“…Initial random perturbations of ±0.25 K were included throughout the first 400 m above ground to instigate timely turbulence development within the ABL. Simulations spanned 3 hr of simulated time with analysis performed as recommended in Moeng and Sullivan (1994), after at least 5 to 6 eddy turnover times, during the latter 2 hr of simulation on results output at a rate of 1.0 Hz or every simulated second, consistent with the setup and analysis procedure of Mirocha et al (2018). Time‐averaged vertical profiles of wind speed from FastEddy during –3 hr are shown versus the 2‐hr mean profile from SWiFT instrumented tower observations in Figure 7.…”

“…The mean wind speed profile is under predicted in all three grid resolutions by up to 0.6 m s −1 , but is generally ≈0.1. Recall the forcing conditions applied here for FastEddy mimic a specific set of those prescribed for the model intercomparison in Mirocha et al (2018) for this convective case. The modest underprediction here is similar to that reported for multiple models in Mirocha et al (2018).…”

“…Recall the forcing conditions applied here for FastEddy mimic a specific set of those prescribed for the model intercomparison in Mirocha et al (2018) for this convective case. The modest underprediction here is similar to that reported for multiple models in Mirocha et al (2018). This bias is attributable to some degree, to the particular choice of idealized forcing conditions (geostrophic wind speed, surface roughness, and surface heat flux) in addition to the duration of model spinup prior to the assessment window.…”

“…Characterization of the mean and perturbation quantities were performed over 10‐min window intervals with 50% overlap as suggested in Welch (1967, 2016b). This procedure follows that used to calculate spectra in Mirocha et al (2018). Each column of the nine‐panel figure corresponds to spectra from a FastEddy simulation of (20, 10, 5) m resolution versus the fixed 10‐m resolution model results and tower observations reported in Mirocha et al (2018), respectively.…”