Addressing Numerical Challenges Associated with WRF LES Modeling: Stratocumulus Cloud in the DYCOMS-II RF02 Case
Poster PDF
Authors
Andrew M. Vogelmann — Brookhaven National Laboratory
Yangang Liu — Brookhaven National Laboratory
William I. Gustafson — Pacific Northwest National Laboratory
Hailong Wang — Pacific Northwest National Laboratory
Jerome D Fast — Pacific Northwest National Laboratory
Wuyin Lin — Brookhaven National Laboratory
Satoshi Endo — Brookhaven National Laboratory
Heng Xiao — Pacific Northwest National Laboratory
Branko Branko Kosovic Kosovic — National Center for Atmospheric Research (NCAR)
Bill Skamarock — National Center for Atmospheric Research (NCAR)
Category
General Topics – Cloud
Description
We investigate an issue raised by Yamaguchi and Feingold (2012) who report that the Weather Research and Forecasting (WRF) model can have an unexpected convergence issue when run in a large-eddy simulation (LES) configuration. The issue was discovered for the GCSS/GASS DYCOMS-II RF02 stratocumulus case, where simulations would not converge in terms of liquid-water path unless an acoustic time step was used that was smaller than expected from the Courant–Friedrichs–Lewy (CFL) condition. To understand this issue further, we use the DYCOMS-II RF02 case to determine conditions that are predisposed to the non-convergence issue using a series of sensitivity tests to initial thermodynamic and wind profiles, surface and large-scale forcings, grid spacing, dynamics, physics, and numerical options. We found that the dependency to acoustic time step is sensitive to the initial horizontal wind speed near the inversion height and water vapor difference across the inversion. The non-convergence is associated with small-scale fluctuations in horizontal wind components that are not present in the converged runs and are likely responsible for the lack of convergence. We find that these fluctuations can be detected early in the simulation, before the onset of boundary layer turbulence, through a metric derived from filtering the wind variance. Using this metric, we show that larger fluctuations tend to occur with the combinations of larger wind speed and water vapor difference at inversion height. A Galilean transformation, which simulates dynamics on a moving coordinate instead of static coordinate, can be used to reduce the wind speed in the advection calculation and significantly reduce the acoustic time step dependency. We also found that the Weighted Essentially Non-Oscillatory (WENO) advection scheme, which is designed to inhibit overshooting and undershooting of the solution, can inhibit the fluctuations and attain convergence for the DYCOMS-II RF02 case even with a large acoustic time step. Thus, the results show that the acoustic time step dependency in WRF model reported in Yamaguchi and Feingold (2012) for large-eddy simulations of stratocumulus can be effectively eliminated even under the condition of high wind speeds at the inversion height. See the Fast et al. poster for detailed examination of the same issue using shallow cloud cases observed at ARM SGP site.
