The need for caution when interpreting velocity field structures predicted by LES

Large eddy simulation (LES) is an attractive alternative to expensive field experiments. It provides all the state variables at a density that field experiments cannot duplicate, and it is unbothered by instrument failure or vagaries of the weather gods. LES models run inexpensively, because smaller scales are treated as resolved-scale bulk effects. Given such advantages, it is unsurprising that LES results are often substituted for observations. A 2001 Stevens and Lenschow article in the Bulletin of the American Meteorological Society proposed criteria (or “conjectures”) to be satisfied to justify use of LES. The second of their two criteria is, “The statistics of the low-frequency modes that are explicitly calculated by LES are not sensitive to errors in the parameterization of SGS effects.” This restates the underlying LES assumption that the largest scales important to flow evolution are resolved, and that unresolved effects should be much smaller than those described directly. The validity of the Stevens and Lenschow conjecture is tested here by examining the detailed structure of the LES flow fields using different turbulence closure models.

Neutral, geostrophically forced (10 m s–1, ~45° N) flow over a flat, rough surface in a 1.5 km deep, 1.28 km square domain with periodic lateral boundary conditions was simulated with the Advanced Regional Prediction System. First, we used the basic Smagorinsky model and the Wong-Lilly dynamic model combined with five different methods and levels of velocity reconstruction on a 40×40×40 grid. Then, we tested the Smagorinsky model and the simplest dynamic reconstruction model on a fine scale 120×120×120 grid.

Differences in layer means among subfilter models were small (but usually statistically significant), but patterns formed in horizontal planes by vertical fluxes of u component momentum (w'u', as ejections, sweeps and upward momentum flux) were sometimes quite different. The coarser grid LES patterns are distinctly different near the surface for different subfilter models. Patterns of spatial correlation in the vertical motion fields also differed among models. Finer resolution generally reduced, but did not eliminate differences. There were significant differences in the length of the zero-flux isopleths, reflecting differences in interface complexity between regions of upward and downward flux. Finer resolution produces longer (more convoluted) interfaces. Differences found here between subfilter models suggest a need for considerable caution when interpreting LES experiments. The dynamic reconstruction model combinations tested produced more realistic mean profiles of wind speed and near-surface flux patterns. Dynamic models with reconstruction allow backscatter of energy that more faithfully mimics interactions between resolved and subfilter scales, thereby yielding a more active spectrum in the smaller scales of the resolved flow.