DYCOMS II SCM email exchange

Bjorn Stevens (25 Sep 2003)

Hi all,

you are getting this email because you have all expressed an interest in doing the RF01 case using an SCM. Toward this end, it would be nice if we coordinate our ouput/forcings/configuration so we can compare with one another and the models. Martin Koehler had expressed an interest at one time in coordinating these activities. So what I propose is that we exchange emails, moderated by Martin, and define how the case will look like, then we can discuss how to synthesize results etc., in anticipation of the workshop.

Martin, would you be willing to do this. The current group includes:

Chris Bretherton (UW), Surabi Menon (Lawrence Berkeley), Pier Siebesma (KNMI), and me as an observer. If Chris forwards this to the wider GCSS list with an instruction to contact you if one is interested in participating at this level, this might be a way forward. I see no reason to maintain the September 30 deadline for this group.

Thoughts?

Bjorn

Martin Köhler (25 Sep 2003)

Hi DYCOMS SCM modelers,

thanks to Bjorn for suggesting a community based SCM case setup. I am quite happy to moderate this activity. I just had a little talk with Pier. Here are some ideas aimed at a quick start up:

configuration: I'd suggest full freedom in your model configuration such as vertical resolution. I would use the operational model. Pier suggested to use the old RACMO with a TKE PBL parameterization. The new RACMO uses ECMWF physics and it might be boring to submit two ECMWF SCMs.

initial state: Bjorn and Moengs's DYCOMS RF01 LES setup ("BM") should be sufficient for SCM initialization (http://www.atmos.ucla.edu/~bstevens/dycoms/dycoms.html). A potential problem might be to extend the initial state far above 1770m.

forcing: Vertical velocity, geostrophic wind and SST are also specified in BM. The horizontal advection terms, I guess, are assumed zero? The difficult question is about radiation (see cases).

cases: To start simple I would suggest to begin with a cases similar to (4). One could use the full radiation scheme and interactive surface fluxes according to each SCM. It sure would be ideal to then also try the cass (1), (2) and (3). But one might argue that the purpose of the LE simulations is to understand the processes involved. The SCM runs would then primarily be used to evaluate each set of parameterizations. Shortcomings can then be addressed ineach model.

output: Postscript. If one person volunteers to intercompare the results, we can agree on a format. A simple choice would be to use the EUROCS FIRE output format (ftp://ftp.phys.uu.nl/pub/jonker/eurocs_init.ps.Z). This might motivate some EUROCS SCM modelers to jump in this SCM excersise. Or one could use the netcdf format suggested by BM for LES submissions. This format should be much easier to be processed for an intercomparison.

I guess that Surabi is the first to try this SCM case. I'd be interested in your experience. We can help each other a lot on the way be communicating!

Good luck and see you in Colorado,

- Martin (and Pier)

Bjorn Stevens (25 Sep 2003)

Hi All,

for the vertical structure aloft one can use what I concocted in my RF01 analysis which is in our inpress QJ article (on my website). The basic gist is that we do the (z-zi)1/3 profile until the lapse rate gets to 3.5 K/km, and then we fix the lapse rate at that value. If one is using an interactive radiative scheme it is important to insert a moist layer aloft, per our description in the RF01 paper near page 11, around equations 8 and 9. if this proves difficult to digest I can provide the temperature and moisture soundings which I used to do the radiation calculations discussed in that paper.

as for horizontal advection. i see two possibilities: (1) lagrangian calculations to follow LES; (2) eulerian calculations where we assume a boundary layer gradient in properties implied by the time evolution of the lagrangian layer sampled in RF01. Overall this can be inverted to give an estimated upstream and downstream boundary layer structure. One problem I have with this is that it may artificially constrain the results, an alternative would be to relax the solutions to the initial conditions on fixed timescales (1 day for moisture, slightly longer for humidity, but maybe one day works for both) and look at how the solutions depart.

I do not have lots of experience in the SCM world so take these comments with that fact in mind.

Cheers, Bjorn

Martin Köhler (25 Sep 2003)

Thanks Bjorn for filling in the details!

for the vertical structure aloft one can use what I concocted in my RF01 analysis which is in our inpress QJ article (on my website). The basic gist is that we do the (z-zi)1/3 profile until the lapse rate gets to 3.5 K/km, and then we fix the lapse rate at that value. If one is using an interactive radiative scheme it is important to insert a moist layer aloft, per our description in the RF01 paper near page 11, around equations 8 and 9. if this proves difficult to digest I can provide the temperature and moisture soundings which I used to do the radiation calculations discussed in that paper.

I agree, either of the two method concerning the moisture aloft should be used. Maybe the easiest and most realistic way would be to provide the T and q profiles and then we can interpolate them to our SCM grids (for above 1770m).

as for horizontal advection. i see two possibilities: (1) lagrangian calculations to follow LES; (2) eulerian calculations where we assume a boundary layer gradient in properties implied by the time evolution of the lagrangian layer sampled in RF01. Overall this can be inverted to give an estimated upstream and downstream boundary layer structure. One problem I have with this is that it may artificially constrain the results, an alternative would be to relax the solutions to the initial conditions on fixed timescales (1 day for moisture, slightly longer for humidity, but maybe one day works for both) and look at how the solutions depart.

I think (1) the Lagrangian calculation is preferable for its simplicity and consistency with the LES setup.

- Martin

Bjorn Stevens (26 Sep 2003)

Hi All,

Per Martin's request attached is a prototype sounding. It might differ in small details from what is specified for the LES. If anyone finds the discrepancies too egregious we can adjust things as needed.

Please comment --- Bjorn

PS the respective columns are: z, p, T, qt, ql, Dt/dt, LW_up, LW_dn, for winds one should use the LES specs...

Martin Köhler (26 Sep 2003)

Thanks, Bjorn, for the profiles. That should be sufficient.

And welcome to Luo Yali, a student of Steve Krueger and Adrian Lock, who considers to apply both UMKO SCM and the SCM he is working on in Canada (TKE scheme) to the DYCOMS II case. (Please add both emails to any discussion.)

- Martin

Martin Köhler (8 Oct 2003)

Hi all,

Anning Cheng has joined the DYCOMS SCM group - welcome!

And thanks to Chris and Ping for contributing their initial results (http://www.ecmwf.int/staff/martin_koehler/docs/dycoms/ZhuBretherton). It was interesting to see the small sensitivity to the choice of radiative scheme and vertical resolution. The drastic impact of shallow convection scheme on/off reflects the experience of some modelers in the EUROCS FIRE SCM intercomparison. We had opted there for simulation with all model default specification on (e.g. shallow convection on and model inherent radiation scheme). A second run was then suggested with a more idealized setup. I experienced from my current PBL scheme implementation that the interaction between the PBL and shallow convection parameterization is paramount for successful low cloud cover (and low level wind).

In terms of output, I think we should adopt the same NetCDF output as suggested by Bjorn for the LES group - excluding the impossive like dataset D (TKE budget). This simplifies the comparison of SCM and LES results and would also help a combined publication as was done in the EUROCS FIRE stratocumulus case (Dynkerke et al 2003 submitted to QJ).

Realizing the producing this output might take some time I suggest each SCM modeler to send me two simple ASCII files of

(1) time averaged profiles (0-6hours) of temperature, total water and liquid water, and

(2) time evolution LWP of 48 hours.

The simulation choosen should use all model default setups. I could then precede the ~5min presentations of each SCM with some common plots to put them im relation. It would still be nice if you could submit your postscript files for addition to the web page.

- Martin

ps. simple output sample

height   temperature   total water  liquid water
  [m]        [K]         [kg/kg]      [kg/kg]
 1000        302          0.010        0.001
 ....        ...          .....        .....

time     LWP
 [s]   [kg/m^2]

  0    0.100
3600   0.150

Ping Zhu (15 Oct 2003)

Hi, Martin:

I don't remember if anyone said how long we should run the case. As for
LES, Bjorn set it to 4 hours. This is certainly too short for us (SCM).
Right now I am running this case for 7 days.

Also Surabi just asked me how I specified the initial profile. Since
rf01_idealb.dat does not specify U and V, I just used geostrophic winds
suggested by Bjorn as the initial profile and let the mode adjust by
itself. So I did not use first-day simulation. I also tried two type of
surface forcings. One specifies surface heat fluxes, and the other specifies
SST.

For the LES-comparison, Bjorn requests the results from the last two-hour
simulations. What about us (SCM)?

Could you please clarify this so that we can provide you useful data for
comparison.

Thanks

ping

Martin Köhler (16 Oct 2003)

Hi all,

thanks to Adrian for submitting his results including the simple ascii output using the Canadian SCM!

I also produced results for the standard ECMWF SCM. Doing the runs, I have stumbled over some SCM specific setup issues (initial state, nocturnal, location & time, horizontal and vertical velocity). I recommend other newcomers to follow those.

You will find both added results and my setup comments on the DYCOMS SCM web page (http://www.ecmwf.int/staff/martin_koehler/docs/dycoms).

Ping had some further questions:

(1) SCM simulation length: I recommend at least 48 hours. While the focus should be on the first hours for comparison with LES, I consider it interesting towards which equilibrium a SCM tends.

(2) U and V are initialized with the geostrophic wind.

(3) Output: For the conference I am processing your simple ascii output, as defined in the email discussion on the web page. In Broomfield we can discuss further analysis of the SCM case.

- Martin