Last updated: 31 October 2007
After the retirement of Jürgen Steppeler, I took over the job of the Project Leader for the LM_Z. Jürgen did not put up a precise work plan after the last General Meeting, so I could be rather creative here. But the principle work was clear:
The work of the COSMO year 2006/2007 was then organized as follows:
Since the last General Meeting in 2006 there have been the following activities:
One test case crashed after 10 days during the Assimilation, while interpolating the results to pressure levels. The technical reason was, that the pressure was no more monotonic decreasing from bottom to top. This might result from a diverging of the p-and the z-grid. Therefore, Heinz-Werner Bitzer introduced a "nudging" of the p-grid to the z-grid. For that purpose, the z-grid values are interpolated to the p-grid (in addition to the swapping of the tendencies) and the p-grid values are recalculated with: p_grid_value = 0.75 p_grid_value + 0.25 z_grid_value.
Additional work was going on to reduce the necessary computing time for the LM_Z. Heinz-Werner Bitzer worked on a simplified horizontal diffusion scheme, which is much cheaper. To increase the stability of the dynamical core, he also implemented an implicit vertical advection in the slow tendencies part of the z-dynamics. Adding these two schemes, he experienced the same problem as explained above for the January-case, but in much more cases. The "nudging" of the p-grid to the z-grid also helps here.
The verification of one test case showed bigger problems for the LM_Z at 00 UTC (!), which indicated a problem of the assimilation. These problems got better during the forecast. Taking a closer look, a severe bug was found: the tendencies of the nudging, which are computed in the p-grid, were not transformed to the z-grid. So the z-grid did not know about the nudging at all.
After removing that bug, Version 1.9 was build and distributed to the partners. This version also contains some changes implemented in the COSMO-Model in the meantime and another bugfix in the turbulence scheme: For imode_turb=2/3, a term for the moisture convergence must be saved in turbdiff and added to dqvdt in the z-dynamics to trigger convection properly. The 3 test periods were started again with the new version.
Note, that the changes of Heinz-Werner (new horizontal diffusion, vertical implicit integration, nudging of p-grid to z-grid) are not in the official Version 1.9.
Lucio Torrisi worked at the implementation and testing of the implicit treatment of the vertical turbulent diffusion (imode_turb=1). In comparison to imode_turb=3-option (not documented) this version is more similar to the implementation used operationally in the TF model (no vertical varying implicitness and no horizontal average of diffusion coefficients in imode_turb=3). These changes are also not in the official Version 1.9.
Work was also devoted to improve the physical interface:
This work also is in progress and is not in the official Version 1.9.
Test cases were run at MeteoSwiss, HNMS and DWD:
Until May 2007: Version LM_Z 1.7:
Three flash-flood events in the Piedmont region, Italy (4-6 June 2002 and 1-2 September 2002) and the Cevennes region, France (5-9 September 2005) have been simulated using the COSMO-Model at horizontal resolutions of 7km and 2.2km. The events were predominantly convective. Two simulations per day are performed for the forementioned periods starting at 00 and 12 UTC. The simulation period is 24 hours. IFS data are used at the lateral boundaries and initial data are taken from a 12 hour assimilation run.
Three different model configurations are used for the simulations. A basic simulation is performed using COSMO-Model version 3.20 with terrain following coordinates (SET1). A sensitivity study is performed using Z-coordinates instead of terrain following coordinates (LM_Z version 1.7; SET2). The LM_Z version is based on an older COSMO-Model version, thus, the SET2 runs not just differ regarding the vertical coordinates, but also other settings: Leapfrog instead of Runge-Kutta scheme, no prognostic calculation of rain, shorter time steps (7km: 15s instead of 72s; 2.2km: 5s instead of 20s) and a slightly different orography. A third simulation is performed in order to separate the effect of Z-instead of terrain following coordinates: COSMO-Model version 3.20 is used with terrain following coordinates and the settings of the Z-coordinate runs (SET3).
The comparison of SET1 with terrain following coordinates and SET2 with Z-coordinates shows significant differences: The results with Z-coordinates show a more pixel-like structure compared to the more band-like structure with terrain following coordinates. Additionally, the amount of area averaged precipitation is higher in the simulations with z coordinates. The evaluation with rain gauge measurements shows that results of SET1 agree significantly better with measurements than those of SET2 with z coordinates. This result is confirmed by discharge simulations with hydrological models using the simulated precipitation as atmospheric forcing.
Major differences also exist between results with Z-and terrain following coordinates with the same settings (SET2 and SET3). The results with Z-coordinates have a more pixel like structure and the results with terrain following coordinates have a more band-like structure-like observed for SET1 and SET2. Additionally, the shape of the region with precipitation is very different. The comparison of simulated precipitation and rain gauge measurements shows a slight improvement if Z-coordinates are used. But a much stronger improvement is received by using terrain following coordinates with the settings of SET1 instead of SET3. Some simulations have been repeated changing just one of the settings per simulation. The results suggest that the strongest improvement is caused by using Runge-Kutta instead of Leapfrog scheme and prognostic precipitation.
Concludingly, the use of Z-coordinates instead of terrain following coordinates has a slightly positive effect on the simulation of precipitation, but a stronger positive effect results from using the Runge-Kutta scheme and the prognostic calculation of precipitation.
May-August 2007: Version LM_Z 1.9:
In the last weeks, the new Version 1.9 was tested with some single cases of the dates above. But for all cases no longer time step could be used, all runs with dt = 30s and also with dt = 20s were instable.
But the reason for this seems to be the 45-layer version used at MeteoSwiss. Because the LM_Z only has an explicit vertical advection, the time step has to be reduced, if the layers are thinner. Some short tests at DWD for one of these cases showed that the 40 layer version runs stable. Moreover, also the testversion by Heinz-Werner Bitzer with the implicit vertical advection runs stable using the 45 layers and a time step of 30s.
Version 1.8 based on COSMO-Model 1.8 has been released. At DWD 3 test cases have been started for the periods:
All tests were run with the full assimilation cycle. 2 Forecasts per day up to 30 hours have been performed (for 00 and 12 UTC). After experiencing the problem in the nudging, all periods were restarted and are still running. The January test with the pure binary from Version 1.9 crashed again with the same problem reported above. There seems to be a diverging of the grid.
No verification results are available right now.
Because of the problems experienced and described above, we did not start to write that report. It is very likely that this report will also NOT be available for the General Meeting in Athens. We hope to have at least some short verification results from the test periods of DWD until then in order to present them to the Evaluation Group.
There was no more work invested in the Semi-Lagrange Version this year.
See Section 3
| Sep-Dec | Jan-Apr | May-Aug | Sum | |
| Torrisi | 0.10 | 0.10 | 0.15 | 0.35 |
| Bitzer, Schättler, DWD | - | - | 0.10 | 0.10 |
| de Morsier, Dierer, Rotach | 0.15 | 0.25 | 0.10 | 0.50 |
| Avgoustoglou | 0.10 | 0.10 | 0.20 | 0.40 |