lanl / palm_lanl

LANL Contributions to PArallelized Large-eddy simulation Model (PALM)
3 stars 5 forks source link

High momentum diffusivity for ice melting, AMD cases #69

Open cbegeman opened 4 years ago

cbegeman commented 4 years ago

@xylar @vanroekel @katsmith133 I'm running into high km values in the second grid level below melting ice interfaces as a result of high buoyancy gradients. This problem was noted here, at which time the problem seemed limited to just the first grid level below the interface.

I'd like to get your thoughts on possible remedies, whether that's in code changes to AMD, a more restrictive km_max or different simulation configuration. Thanks! sa_6hr_z_profile k_6hr_21zmax_z_profile Note, km_max = 0.1. Namelist file here: /lustre/scratch3/turquoise/cbegeman/palm/jobs/test_ocean_amd_mcphee_lowres_disturb5h_cd1e-2_rotation

cbegeman commented 4 years ago

A few more figures for the same case momflux_z_6hr_21zmax_z_profile e_res_6hr_21zmax_z_profile

vanroekel commented 4 years ago

@cbegeman one other possibility besides altering K_max is to allow the Poincare constant to have a length scale that is a function of stratification (similar to what Schuman did for Smagorinsky), which would limit SGS fluxes in high stratification scenarios.

cbegeman commented 4 years ago

@vanroekel Per our conversation yesterday, I looked into the l_stable variable currently used in PALM (Deardorff, 1980) and it is a function of the SGS TKE, which we do not solve for with AMD. Are you aware of other mixing length scales that are a function of stratification without a dependence on e?

vanroekel commented 4 years ago

shoot of course, I forgot that. Let me think about this some more.

cbegeman commented 4 years ago

One idea is to limit the mixing length to l_wall as in PALM's closure, without the stratification dependence. This could help somewhat.

vanroekel commented 4 years ago

It might be worth a shot, I worry though that we may find ourselves continually lowering the influence for different tests.

cbegeman commented 4 years ago

Summary of our conversation yesterday: The stratification term is only dominant when TKE is low, and in that case the viscosity is unrealistically high. Cat Vreugdenhil has found that in a stratified Couette flow, the stratification term is not dominant. It could be the case that the stratification term is not important when the flow is well-resolved by LES but is important for these simulations with dz >= 0.25m. In the absence of more information we proceed with simulations in which AMD is used but the stratification term is removed (see PR https://github.com/xylar/palm_les_lanl/pull/72). The scale-dependent Lagrangian dynamic model of Bou-Zeid et al. (2005) is a good candidate for an alternate sub-grid scheme.