7. Operational version in Czech Republic
(more details vana++at++chmi.cz)
Since 1st January 2003 every LACE Member exploits its own
operational application of ALADIN. The recent ALADIN/LACE application operated
in Prague LACE centre became the ALADIN/CE (CE standing for Central Europe)
configuration of CHMI. In the beginning of 2003 it naturally kept its
continuity with the previous LACE suite.
1. Evolution of the ALADIN/CE application
The ALADIN/CE suite switched to the version COCONUT+++ of the
model physics :
27/05/2003 at 12 UTC network time for the production run
and at 06 UTC network time for the assimilation cycle : COCONUT+++.
This physics package became operational in ARPEGE mid-April
2003. According to the original plan in Prague the switch to this model
version was thought only for the end of June, together with the move to the
new computer NEC-SX6 and to a higher cycle of ALADIN (AL25T1). However, with
the start of warm season, the operational version CYCORA_ter+++ was not
providing reasonable values of CAPE; these were extremely high. This fact
together with the requirement of correct validation between the old cycle AL12
and the new one AL25T1 lead to the back-phasing of the COCONUT physics to the
still operational cycle AL12 on the old computer NEC-SX4 (insufficient memory
on the SX4 machine was the blocking factor). The COCONUT package was validated
and gave the same results with both cycles. Meantime, already in March, the
parallel testing started in comparison with the operational CYCORA_ter+++
version. These tests have shown that the COCONUT version as implemented in
ARPEGE does not give completely satisfactory scores (at least neutral). For
example, the bias of 2 m temperature got considerably worse, it became too
cold (suite ABZ). Many other tests were made, see the description below.
Finally, the satisfactory solution was found when coming back on the radiation
scheme (+), on the nebulosity scheme (+) and on the horizontal diffusion
coefficients (+) to the CYCORA_ter tunings.
Impact on the forecast : neutral scores, the CAPE
values get reasonable compared to CYCORA_ter+++.
Technical impact : none.
The ALADIN/CE suite switched to cycle AL25T1 on new computer
NEC-SX6 :
30/06/2003 at 12 UTC network time for the production run
and at 06 UTC network time for the assimilation cycle : AL25T1 on NEC-SX6
In the beginning of 2003, CHMI purchased a new computer, the
starting configuration is NEC-SX6/B with 4 processors and 32 Gbytes of shared
memory (one node). The machine was installed in March and the acceptance tests
were successfully completed on 10th June 2003. The switch of the complete
ALADIN/CE suite took place on 30th June 2003, in its version COCONUT+++ on
cycle AL25T1. As first future step, the resolution of ALADIN/CE will be increased.
Beside the model configurations needed for the operational
suite, a considerable effort was put to the installation and validation of the
ODB software. Not all ODB tools were however validated before the end of June.
Impact on the forecast : neutral.
Technical impact : important increase in the memory
consumption due to the data flow, different from the old shared memory code.
2. Parallel suites
The following parallel tests were launched to assess the
impact of different modifications:
- Suite ABY : It was the test of new tuning of the gravity-wave
drag and orographic lift when the envelope orography is abandoned (following
the study of Dunja Drvar). The scores were not satisfactory and the
mean-sea-level pressure maps have shown that the drag and lift do not
compensate enough the blocking effects of the envelope. Further study will
take place.
- Suite ABZ : It tested the COCONUT version of physics; at the same
time the horizontal diffusion intensity was increased twice. The scores were
not the best. There was a too cold bias of 2m temperature, computed against
SYNOP observations and compared to the operational version.
- Suite ACA : This test is based on the ABZ suite, but it comes
back on the cloudiness scheme. The scores compared to ABZ are worse and they
remain not satisfactory, compared to the operational version.
- Suite ACB : This suite should have been comparable to the model
version known under the name DICORA (used in ARPEGE before COCONUT). The
scores were not satisfactory either. Later on it was found that there was a
mistake in leaving the switch LSRCONT
=.true. (it should be switched off with DICORA).
- Suite ACC : This test was based on the COCONUT version (suite
ABZ) however a different tuning was used for the mixing length. The goal was
to find the reason of the worse scores of COCONUT. The results were not really
changed by this retuning.
- Suites ACD, ACE, ACF : All these tests were based on the DICORA
version, adding the shear-linked convection or using various tunings. The
results were not better and they did not help much to discover the reason of
the bad COCONUT scores. On top, all these tests were affected by the mistake
(setup of LSRCONT ) made in
the preparation of DICORA (suite ACB).
- Suite ACG : Since the DICORA version did not lead to any
improvement or to any clue on the COCONUT scores, we came back to COCONUT in
this test. On top of coming back to the operational version (CYCORA_ter) for
the cloudiness scheme, the same was made for the radiation scheme. This test
gave reasonably neutral scores for a trial period in winter. As it was a
promising configuration, it was also re-tested for a summer period, this time
even with a little improvement of the geopotential score. Within the summer
period test it was noticed, however, that the field structures of humidity
convergence or CAPE were not as sharp as in the operational version. Therefore
several single case tests were made in order to see what could cause the damping.
- Single test ACI, ACJ, ACK: In these tests the convection tuning
was modified (for parameters GCVPSI
, GCOMOD,
LSRCONT respectively) with
respect to the ACG suite. The results (field structures) were almost the same
as in case of ACG.
- Single test ACL : This test is based again on the ACG settings,
but the horizontal diffusion coefficients were set back to the values of the
operational version. Indeed, too strong diffusion in ACG was the reason of too
smooth fields. Therefore it was decided to keep the horizontal diffusion
tuning without change. The ACL settings then became operational on 27th May
and it got the name COCONUT+++ (first (+) for cloudiness scheme, second (+)
for the radiation scheme and third (+) for the horizontal diffusion).
- Suite ACH : This test was inspired by a study made with ARPEGE,
where a better precision of computation of the semi-Lagrangian trajectory was
tried (using three iterations instead of two in the trajectory computation).
The test was based on the ACG suite and it was compared both with ACG and with
the operational version. Surprisingly, the scores were worse in altitude,
namely in bias of geopotential, which made worse also the RMSE geopotential
score. This test was then left aside but perhaps some study can be later made
around this issue.
- Suite ACM : test of the AL25T1 cycle with the COCONUT+++ settings
on the SX6 machine. It was compared to the AL12 cycle with the COCONUT+++
settings on the SX4 machine (operational version since 27th May). The scores
were perfectly neutral. The comparison was confirmed by the same result
obtained for the 24 h cumulated precipitation forecast with both suites. The
ACM suite became operational on the SX6 machine on 30th June 2003.
The results of parallel tests may be consulted on / pages.