Current Research

GFS Single Column Modeling

January 2012

Updates on tests with the DYCOMS simulation:
Sensitivity of adding subsidence advection, part II (part I below)
Sensitivity of adding a flag that effectively turns off shallow convection in this case.
Additional plots for above: looking more at the first hour of simulation. The scheme recieves this initial supersaturated moisture profile. But it takes quite a while for any cloud water to form, and cloud fraction rises much more quickly than cloud liquid water does. This is even more the case when the shallow convection scheme isn't firing due to the above flag. Here are also profiles illustrating this at t=0, t=30 min, and t=60 min. Note that "saturation deficit" is actually qv-qs here.

Updates on changes to shallow cu scheme
I discovered that the "bug" in Jongil's shallow cu changes were actually a bug in my implementation of said changes, namely, I wrote "ktcon" in a line that should have been "ktcon1". Below are comparisons of control, "EntrPE", and "EntrPE_UWparams" experiments now that this bug has been fixed. (This fix actually produces results identical to my own fix of changing "max" to "min", explained in "Exploring changes to shcu scheme" below, but I am posting for completeness). In "EntrPE", I've changed the entrainment and precipitation efficiency; in "EntrPE_UWparam" I've also changed the parameters of the vertical velocity equation to those.

Control vs EntrPE:
theta, mass flux, qt excess, precip, buoyancy, and wu2.

Control vs EntrPE_UWparams:
theta, mass flux, qt excess, precip, buoyancy, and wu2.

December 2011

Summary of week of Dec14-21, comparing control and "subadvect" experiments of the DYCOMS case with the GCSS SCMs.

I tried a run where I changed the line in the convection scheme that chooses cloud top to be the larger of kbm or that determined by wu2, to the smaller of those two. Then I tried a run with the new entrainment and grid scale detrainment parameters, but with precip efficiency = 0.
I also tried a run with just as above but with precip efficiency = 0.001 and with the wu2 equation having the same parameters as they have in the UW scheme. The vertical velocity equation in the model as Jongil sent to me is:
1/2 d(wu2)/dz = 3/8 buo - 1/6 epsilon*wu2
I'm not sure where these numbers came from. They are not only quite a lot smaller than in the UW scheme, but their relative magnitudes are almost opposite. UW scheme has the buoyancy coeffecient equal 1 and the entrainment coefficient equal 2. This formulation has the buoyancy coefficient more than twice the magnitude of the entrainment coefficient. So I tried a run with the coefficients equal to what they are in the UW scheme.

Plots comparing these two runs: mass flux, precip, and MSE excess.
(Compare to other runs by clicking on links below)

Compare buoyancy and wu2 in four runs, all with the vertical velocity equation: control, with my entrainment, detrainment, and precip efficiency parameters, with those parameters plus UW parameters, and with those parameters except no convective precip.

Summary of week of Nov 28-Dec 2, incorporating Jongil's changes to physics into the SCM at UW.

Exploring changes to shcu scheme

(these simulations also have vertical advection, which had little impact)
Plots from runs with no change to the PBL scheme, only the vertical velocity calculation in the shallow cu scheme. The main difference here is that cloud top is now set at the level where wu2, the square of updraft vertical velocity, goes to zero. The vertical velocity is diagnosed from buoyancy (which includes virtual effects). As these plots show, this change makes convection much too deep.
mass flux, PBL, cld base, cld top heights, precip, and MSE excess.

This plot of cu updraft MSE excess (which is used for the initial guess of cloud depth), cu buoyancy, and wu2 shows that there is a bug in this routine. At t=3 hours, hup < hsat at 1 km, while wu2 first crosses zero at 1.5 km. This should be the cloud top. Instead, cloud top is the value kbm, the max level it's allowed to be. This turns out to always be the case. (At t=3hrs, this happens to be a level where wu2 crosses zero a second time, but this is a coincidence, as I have found by printing out values of kbm, ktcon (cloud top), and wu2 at all times within the simulation.) The cloud depth would actually be quite reasonable if it chose the first place where wu2 goes negative.

In my first try at fixing this, I commented out the line after the above part of the code which sets cloud top to the max of either kbm or the level where wu2 goes to zero. This definitely reduces the clout top height, but precip is still too high (not as high, though) and convection gets somewhat jumpy. Here are mass flux, PBL, cld base, cld top heights, precip, (with a new axis), MSE excess, and buoyancy from this run. (Buoyancy isn't plotted in for other runs because only the new shallow convection scheme uses this buoyancy, which includes virtual effects.)

When I combine the above method with the changed entrainment, detrainment, and precipitation parameters, it doesn't seem viable. However, changing those parameters seems to vastly improve the changed shallow cu scheme on its own.
Comparisons of mass flux, PBL, cld base, cld top heights, precip, (with a new axis), MSE excess, and buoyancy between runs with the updated Shcu scheme and runs with the updated scheme + changed entrainment, detrainment, and precip efficiency per the 2011 CPT annual meeting.
This plot shows that there is a problem with wu2 in this run.

Similar comparisons of the above run to control run and the run with the older version of shallow convection but with changed entrainment, detrainment, and prec efficiency.
Control: mass flux, PBL, cld base, cld top heights, precip, and MSE excess.
Entr_PE: mass flux, PBL, cld base, cld top heights, precip, and MSE excess.

Further exploration of changes to background diffusion and nonlocal mixing.

(these simulations also have vertical advection, which had little impact)
Comparisons between control and a simulation with nonlocal moisture mixing (no vertical velocity prediction in shallow cu, no changes to background diffusion).
mass flux, PBL, cld base, cld top heights, precip, and MSE excess

Comparisons between control and a simulation with decreased background diffusion of T & q (no vertical velocity prediction in shallow cu, no nonlocal moisture mixing).
mass flux, PBL, cld base, cld top heights, precip, and MSE excess

November 2011
Tests adding Jongil's code for vertical advection, changes to background diffusion and nonlocal mixing in moninq, and vertical velocity in shalcnv

Comparing Control to that with vertical advection
theta (note the different color scale from below)
u
v
mass flux
cloud water
precip
cu updraft MSE excess (from mean saturation MSE)

Adding background diffusion, then vertical velocity stuff
theta
u
v
mass flux
cloud water
precip
precip with a different color scale.
cu updraft MSE excess (from mean saturation MSE)

October 2011

Figs for CPT meeting

Mass flux and ql_up profiles at 3 hours
model-diagnosed cloud water
surface precip time series
time-height precip flux of SCM
time-height precip flux of LES
Env thetav and ql_up profiles at 3 hours
Updraft buoyancy profiles at 3 hours
SCM Updraft buoyancy time-height series
precip profiles at 3 hours

Combined sensitivity tests (multiple parameters changed)

SAM LES (precip allowed, Nc=100)

Cu udraft mass flux
Cu updraft virtual potential temperature excess
Cu updraft theta_v excess with color scale to match experiment below
Cu updraft total water excess
Cu updraft liquid water

Entrainment, detrainment, and precipitation efficiency all changed

Mass flux from shalcu & PBL scheme
Cu updraft virtual potential temperature excess
Cu updraft liquid water

Profiles at t=3 hours
Cu updraft theta_v and qt excesses
Cu updraft mass flux and ql

Entrainment rate c=1 and liquid water conversion to convective (grid scale) rain c0=0.001 (c1=2.5E-4)

Mass flux from shalcu & PBL scheme
Cu updraft virtual potential temperature excess
Cu updraft total water excess
Cu updraft liquid water
Cu updraft vapor
Model-diagnosed cloud water content
Model-diagnosed cloud fraction
rain mixing ratio
Surface precip rate
precipitation flux
cloud base, cloud top, and PBL heights
6 hour mean RH vs z and RH vs T
Shallow cu heating
Microphysics heating
Shallow Cu moistening
Microphysics moistening

Profiles at t=3 hours
Cu updraft theta_v and qt excesses
Cu updraft mass flux and ql
Moist static energy profiles (environmental profiles are for control run).
heating and moistening from shal cu & microphysics
vapor and entrainment rate
precipitation

Everything below this has incorrect thetav excess. Will be updated soon.

Entrainment rate c=1 and liquid water conversion to convective rain c0=0.001

Mass flux from shalcu & PBL scheme
Cu updraft virtual potential temperature excess
Cu updraft total water excess
Model-diagnosed cloud water content
Model-diagnosed cloud fraction
rain mixing ratio
Surface precip rate
precipitation flux
cloud base, cloud top, and PBL heights
6 hour mean RH vs z and RH vs T

Initial sensitivity experiments on BOMEX (only one parameter changed)

Initial notes: Compared to SAM LES (figures not shown), Cu updrafts in GFS have larger maximum theta_v (~315 K as opposed to ~310 K) and much smaller max ql (~1 g/kg as opposed to ~2.5 g/kg). All of the experiments below don't bring updraft properties to those of LES. Mass flux is closer to LES results.

Entrainment rate

The GFS shallow convection scheme parameterizes cumulus updraft entrainment rate = c/z. In LES experiments (e.g., Siebesma et al, 2003) the coeffeicient c ~ 1.0. The GFS uses a smaller value: c = 0.3. Below are comparisons of 6 hours of BOMEX simulation with c = 0.3 and c = 0.6; c=1.

These figures are for c = 1
This produces much cooler updrafts with less liquid water and a smaller mass flux
Mass flux from shalcu & PBL scheme
Cu updraft virtual potential temperature excess
Cu updraft total water excess
Model-diagnosed cloud water content
Model-diagnosed cloud fraction
rain mixing ratio
Eliminates all grid scale precip
Surface precip rate
precipitation flux
latent and sensible heat fluxes
cloud base, cloud top, and PBL heights
6 hour mean RH vs z and RH vs T

These figures are for c = 0.6
Increasing entrainment reduces both vapor and liquid water in updrafts, decreasing theta_v and convective mass flux (due to decreased buoyancy?).
Mass flux from shalcu & PBL scheme
Cu updraft virtual potential temperature
Cu updraft specific humidity
Cu updraft liquid water
rain mixing ratio
Increasing entrainment not only reduces precipitation, it also reduces the fraction of precip that is grid scale.
Surface precip rate
precipitation flux
This experiment has little effect on surface fluxes or the heights of cloud base, top, and PBL.
latent and sensible heat fluxes
cloud base, cloud top, and PBL heights

conversion to convective precip

The operational shallow cu scheme converts updraft liquid water to convective rain at a rate of c0 = 0.002 m-1. Below are comparisons of the control run with a run with c0 = 0.001 m-1.

Mass flux from shalcu & PBL scheme
Cu updraft virtual potential temperature excess
Cu updraft total water excess
Model-diagnosed cloud water content
Model-diagnosed cloud fraction
Interestingly, decreasing the conversion actually increases the rain water mixing ratio later on. This is because the model shifts toward a greater fraction of the precip being grid scale. Convective precip has large precip rates and small rain water mixing ratio; grid scale precipitation is the opposite.
rain mixing ratio
Surface precip rate
precipitation flux
cloud base, cloud top, and PBL heights
6 hour mean RH vs z and RH vs T

conversion to grid scale precip

The operational shallow cu scheme converts updraft liquid water to grid scale rain at a rate of c1 = 5*10^-4 m-1. Below are comparisons of the control run with a run with c1 = 2.5*10^-4 m-1.

Doing this creates overactive and jumpy convection that over-stabilizes in some time steps.
Mass flux from shalcu & PBL scheme
Cu updraft virtual potential temperature
Cu updraft specific humidity
Cu updraft liquid water
Convective precip is increased
rain mixing ratio
Surface precip rate
precipitation flux
latent and sensible heat fluxes
cloud base, cloud top, and PBL heights
This parameter should probably only be decreased in conjunction with a decrease in the conversion rate to convective precip, if at all.

August 2011

DYCOMS with output every time step, first 8 hours only

Control run: with shallow convection on regardless of cloud top relative to PBL top

potential temperature
specific humidity
pressure of pbl top
cloud fraction
liquid water content
precipitation rate
PBL heating
shallow cu heating
microphysics heating
PBL moistening
shallow cu moistening
microphysics moistening

Test run: with shallow convection off if cloud top one or fewer levels above PBL top

Transition (Reference) with Coriolis acceleration on the wind tendencies

Surface Line Plots

potential temperature
water vapor
precipitation
cloud water content
liquid water path
model diagnosed cloud fraction
vertical diffusive heating
vertical diffusive moistening
shallow cumulus heating
shallow cumulus moistening
Microphysics heating
Microphysics moistening

DYCOMS Flight 1 with full radiation & ShCu on, compared to LES with full rad

potential temperature
water vapor
precipitation
cloud water content
liquid water path
model diagnosed cloud fraction
vertical diffusive heating
vertical diffusive moistening
shallow cumulus heating
shallow cumulus moistening
Microphysics heating
Microphysics moistening

DYCOMS Flight 1 with full radiation & ShCu off, compared to LES with full rad

DYCOMS Flight 1 with new physics & full radiation

shortwave radiation on, ShCu on

potential temperature
water vapor
precipitation
model-diagnosed PBL height
cloud water content
liquid water path
model diagnosed cloud fraction
vertical diffusive heating
vertical diffusive moistening
shallow cumulus heating
shallow cumulus moistening
Microphysics heating
Microphysics moistening

shortwave radiation on, ShCu off

shortwave radiation off, ShCu on

shortwave radiation off, ShCu off

Transition (Reference Composite) with new physics

BOMEX SCM comparisons with old and new cumulus schemes

Analyzing the heat budget

Profiles at 3 & 6 hours
Profiles at 9 & 12 hours
Profiles averaged over first 12 & 24 hours
Column-integrated heat budget

Time-height comparisons of the GFS with new schemes to SAM cloud-resolving model

potential temperature
water vapor
relative humidity
cloud water
liquid water path
total heating?

Time-height comparisons of the two schemes

potential temperature
water vapor
relative humidity
moist static energy
precipitation
model-diagnosed PBL height
cloud water content
liquid water path
vertical diffusive heating
shallow cumulus heating
shallow cumulus moistening
deep cumulus heating
deep cumulus moistening
Microphysics heating
Microphysics moistening

Profile comparisons of both with SAM

potential temperature
relative humidity
water vapor
moist static energy
cumulus heating profiles
cumulus moistening profiles

December 7, 2010

Using the new operational PBL scheme, with the modification that z/L = -3 (specified), produces reasonable looking results without having to specify STRESS. The theta profiles have a nicer-looking structure than previous specified STRESS runs, but the 9 hr profile is about half a degree warmer than LES (more near the surface).

Theta and vertical diffusive heating
PBL height and d/dt of column-integrated theta
Theta profiles at 06 and 09 hours
Theta profiles at 42, 45, and 48 hours

December 3, 2010

Plots for Telecon

Control case

Column DSE
theta and vertical diffusive heating
Lowest level temperature and PBL h

150 W/m2 case

Column DSE
theta and vertical diffusive heating
Lowest level temperature and PBL h

SST 295 case

Column DSE
Lowest level temperature and PBL h

SST 305 case

Column DSE
Lowest level temperature and PBL h

DCBL temperature
DCBL theta
DCBL theta for first 9 hours
DCBL diffusive heating
diffusive heating profiles at forecast hours 3, 9, and 24.
model-estimated PBL height
theta profiles at 06 and 09 hours

ARM observations at Graciosa Island

Comparing observations (sounding, clouds, and precipitation) at an ARM mobile facility on Graciosa Island in the Azores to analyses and forecasts by NCEP GFS and ECMWF.
Go here for more

Hadley Cell project

Using the dry Held-Suarez model to explore the dynamics of the Hadley Cell, specifically the extent to which it is axisymmetric vs. "eddy-driven".
Go here for more

CIN closure project (COGS, Oct 2008-Sept 2009):

COGS talk (May 7, 2009)ppt
COGS talk (May 7, 2009)pdf

Looking at output from several well-verified SAM CRM runs forced by idealized omega profiles from the KWAJEX, TOGA COARE, and ARM Great Plains IOP's, in order to study the relationship between CIN and cloud base mass flux, as opposed to that between cb mass flux and some measure of CAPE, which is typically used in cumulus parameterizations.

Week of August 16, 2009: scatterplots of Cu frac against CIN/TKE and CBMF against closure, with new and improved fit lines.
Weeks of July 27 & August 3, 2009: CIN PDFs
Week of June 23, 2009: updated figs with TKE correction (previous plots were of 2TKE, not TKE)
Week of May 4, 2009 (more plots for COGS talk)
Week of April 27, 2009 (extra plots for COGS talk)
Week of March 9, 2009
Week of March 2, 2009
A finalized set of figs
An updated set of paper figs
Week of February 2, 2009
Week of January 26, 2009
Week of January 19, 2009
A more narrowed-down set of figures for my COGS paper
Week of January 12, 2009
Weeks of January 4 & 12, 2009 (paper figures)
Week of December 29, 2008

Prior to the week of December 29, 2008:

Some investigation into the reasons for the differences between cloudy updraft CIN & sounding CIN (December 2008)
Initial CIN calculations for KWAJEX, TOGA COARE, and ARM (Fall 2008)

SCAM project (~August 2008)

go here for more