adc_parameters_optimization

[amrapallig]

As other parameters did not show much difference, I checked with CwwE and CwwD for wind-only and cooling-only cases and saw a factor of 10 difference for these values. Before moving forward, I want to clarify if this due to a significant change of length scale for both convection and shear cases or some other error.

c2, c4, c16 are cooling-only test cases with a heat flux of -100 W/m^2, -200 W/m^2, and -800 W/m^2 respectively. w1, w2, w5 are wind-only test cases with wind stress of 0.01 N/m^2, 0.02 N/m^2, and 0.05 N/m^2 respectively. I have shown only 1m and 2m resolutions plots here.

Parameter space best suited for the cooling cases are:
config_adc_alpha_tracer1 = 0.6
config_adc_alpha_tracer2 = 1.0
config_adc_c11 = 0.1
config_adc_Cmom = 0.5
config_adc_Ctherm = 0.5
config_adc_Cmom_w3 = 10.0
config_adc_c_slow = 2.0
config_adc_slow_w_factor = 0.875
config_adc_c_slow_tracer = 2.0
config_adc_dissipation_constant = 10.0
config_adc_CwwE = 1.0
config_adc_CwwD = 1.5

Cooling cases:





Now, wind case with similar parameters as in cooling case:

Here w^3 is of the order of -3.00002576e-14 below the mixed layer and hence the areaFraction is 0.01., due to the condition: if(w3check < 3e-14_RKIND) w3check = 1.0_RKIND.
If I change w3check to 0.0 instead of 1.0, area fraction becomes 0.5 below the mixed layer.

Now For wind case:
(keeping other parameters same as above and following changes):
config_adc_CwwE = 10.0
config_adc_CwwD = 15.0

Now for wind+cooling cases (t1w: -50W/m^2+0.01 N/m^2, t2w: -100W/m^2+0.01 N/m^2, t4w: -200W/m^2+0.01 N/m^2), as heat flux increases, CwwE and CwwD need to change from wind case values (10,15) to cooling case values (1,1.5).

[amrapallig]

@BrodiePearson , here are the plots for one cooling and one wind case.

[BrodiePearson]

@amrapallig Thanks for making these subplots. One thing concerns me:

• The wind case has the same amount of TKE as the LES, but in the wrong components! In the LES this energy is all in the along-wind component, but in the ADC this energy is spread across along-wind, cross-wind, and vertical components.

The increased $\overline{w'w'}$ is probably causing the rapid entrainment that we're seeing in the wind cases. When we increase the entrainment/detrainment parameters this artificially dissipates the excessive vertical velocity variance but is not the ideal solution - since the other TKE components are still poorly emulated. I think we should delve into why the TKE is being rapidly redistributed between velocity components. Normally I would associate this with the pressure-strain terms (or in this case the parameterization for the pressure terms). Could you run a couple of tests to diagnose the source of this transfer:

1. Turn off the rapid pressure-strain term, which directly puts some shear production in the cross-wind and vertical TKE budgets and reduces the along-wind TKE production rate. Do this by setting alpha_1 and alpha_2 to zero.
2. If that doesn't work, try switching off the return-to-isotropy term which automatically tries to make all TKE components equal in size. Do this by setting c_slow to zero (this might crash the model though...)

In both cases just run these changes for the wind-case using the setup of the red-dashed line simulation above (convective diffusion parameters). It would be useful to see plots of the three TKE components and the vertical momentum fluxes ($\overline{u'w'}$ and $\overline{v'w'}$) with lines for the original setup, the alpha terms set to zero, and the return-to-isotropy term set to zero.

[amrapallig]

With alpha_1 and alpha_2 set to zero, solutions are unstable even though it runs for 5 days, and with c_slow set to zero, code breaks in a few time steps.

[amrapallig]

24 hr simulation. lest0w2_palm has f=0.28e-4, $\tau=0.02N/m^2$ $Q_h=-5W/m^2$, other two are with 0.175e-4 and $\tau=0.02N/m^2$

[amrapallig]

[amrapallig]

Test of $\alpha_1$ and $\alpha_2$: All cases are with CwwE = 1.0, CwwD = 1.5
case0: $\alpha_1=0.9, \alpha_2=0.5, \alpha_1/3+\alpha_2=0.8, -2\alpha_1/3=-0.6, \alpha_1/3-\alpha_2=-0.2$
case1: $\alpha_1=0.6, \alpha_2=0.4, \alpha_1/3+\alpha_2=0.6, -2\alpha_1/3=-0.4, \alpha_1/3-\alpha_2=-0.2 $
case2: $\alpha_1=0.6, \alpha_2=0.3, \alpha_1/3+\alpha_2=0.5, -2\alpha_1/3=-0.4, \alpha_1/3-\alpha_2=-0.1 $
case3: $\alpha_1=0.375, \alpha_2=0.375, \alpha_1/3+\alpha_2=0.5, -2\alpha_1/3=-0.25, \alpha_1/3-\alpha_2=-0.25 $

[amrapallig]

[amrapallig]

@BrodiePearson @katsmith133 @vanroekel

temperature profiles_ (parameters similar to the plot above(red,blue,cyan,magenta))

uw profiles

It seems uw profiles kickback to zero in some cyclic manner after encountering a positive value that contributes to a shallower mixed layer and again wind forcing acts on it with a initial mixed layer. I ran for 50 days and adc still shows a similar pattern over some time intervals.

[amrapallig]

• According to the above plots, there are two issues: 1)outburst and 2)over-mixing(even before outburst and
  for any parameter sets).
  -I turned the constants of namelist files as per Canuto 2007. : alpha_tracer1 = 0.2175, alpha_tracer2 = 0.3625 (alpha3=0.29), Cmom_w3 = 5.0, c_slow =2.5 and c_slow_tracer = 5.936 which helped with the outburst.

• Fixed two bugs in the code (correct me if I am getting it wrong or overlooked any previous discussions)

  -ln 439: tauw3(k,iCell) = C_mom_w3*KE / (1.0E-15_RKIND +sqrt(2.0_RKIND) lenAv), I think sqrt(2) is not necessary as KE is defined as  sqrt(0.5_RKIND*(u2av+v2av+w2av)) 
  
  -ln 726: (alpha_1 + alpha_2)*w2(i1,k,iCell))*Vz,  I think this should be (alpha_1 - alpha_2)*w2(i1,k,iCell))*Vz -this fixes outburts.
  

-Reduced slow_w_factor to 0.121, which helped dampen the over-mixing. (This needs proper justification as the convective case might get affected as well)

-Following two plots are comparison of les_ncar, les_palm and adc with above mentioned parameters for 24 hr (upper pannel) and 120 hr(lower pannel)

-Time series:

• It seems for earlier time up to 48 hrs, adc well represents les_palm. However, towards a later time 4-5 days, the same tuning of ADC better presents ncar_les. Here are plot showing the mixed layer and growth of other profiles showing inhibited mixing in les_palm 48hrs onwards and overmixing in vw profile of adc. The last plot is showing w2, w3 and area fraction for adc data.
• dotted lines are adc, solid lines are less (left column: les_palm, right column: les_ncar). Legend shows different time stamps.
  
  
  
  
  
  
  

-areaFraction ($\sigma$ ) is unstable near the bottom boundary. In the second plot, CwwE and CwwD are changed to 0.2 and 0.3 showing sigma approaches 0.5 below the mixed layer (as w3 is exactly zero, For the first case w3 is ~e-14). This change does not affect any other plots (qualitatively) as discussed above.

[BrodiePearson]

I think that the original vwtend2 term was correct (the sum of alpha_1 and alpha_2 rather than their difference). This links to the analogous term in uwtend2 a few lines before. The pressure term that uses the difference of these parameters are the ones containing v2 and u2 respectively.

Equations (12) and (19) of Overleaf document state the general formulation of these pressure terms, so if there is a mistake we should figure out where that is and fix it in those equations.

[amrapallig]

I was referring to equations 1a and 1b of Canuto 2007. I will check overleaf with the general formulation.