Operational ALADIN forecast in

1
26th EWGLAM 11th SRNWP meetings 4th - 7th
October 2004,Oslo, Norway Zoran Vakula, Lovro
Kalin, Martina Tudor and Stjepan
Ivatek-Šahdan Meteorological and Hydrological
Service, Gric 3, HR-10000 Zagreb,
Croatia vakula_at_cirus.dhz.hr, kalin_at_cirus.dhz.hr,
tudor_at_cirus.dhz.hr ivateks_at_cirus.dhz.hr
Operational ALADIN forecast in Croatian
Meteorological and Hydrological Service
Verification
In the Croatian meteorological service ALADIN is
operationally run twice a day, for 00 and 12 UTC.
Coupling files are retrieved from ARPEGE
(Meteo-France global model) via Internet and
RETIM2000. Model resolutions are 12.2 km for LACE
domain, 8 km for Croatian and 2 km for the
high-resolution dynamical adaptation domains. The
execution of the suite is controlled by the
OpenPBS (Portable Batch System) as the queuing
system. Horizontal resolution of the LACE domain
is 12.2 km, 37 vertical ?-levels, time-step 514
sec, 229x205 grid points (240x216 with extension
zone). Corners SW (34.00,2.18), NE
(55.62,39.08). Horizontal resolution of Croatian
domains is 8 km, 37 vertical ?-levels, time-step
327 sec, 169x149 grid points (180x160). Corners
SW (39.00,25), NE (49.57,22.30). Initialisation
of ALADIN on LACE domain is provided by Digital
Filter Initialisation (DFI). Coupling frequency
and frequency of output files for the LACE and
Croatian domains are 3 hours. When the 48 hours
forecast on LACE domain is finished 48 hours
forecast for Croatian domain starts without
initialisation with coupling files from LACE. 6
domains for the dynamical adaptation of the wind
field in the lower troposphere to 2-km resolution
orography for mountainous parts of Croatia.
Dynamical adaptation is run sequentially for each
output file, with 3 hour interval. In the
dynamical adaptation meteorological fields are
first interpolated from input 8-km resolution to
the dynamical adaptation 2-km resolution. The
same file is used as a initial file and as a
coupling file that contains boundary conditions
for the model. Visualisation of numerous
meteorological fields are done on LINUX PC.
Comparison of forecasts with SINOP data are done
hourly for today's and yesterdays forecast. The
products are available on the Intranet
Internet. Internet address with some of the
ALADIN products, total precipitation and 10 m
wind http//prognoza.hr/aladin_prognoza_e.html
http//www.dhmz.htnet.hr/prognoza/aladin_prognoza_
e.html.
Skill scores for probability of precipitation
made of ranked probability scores of quantitative
ECMWF and ALADIN CROATIA precipitation forecasts
for "1st" and "2nd" day for Zagreb Maksimir
(14240), from summer 1997 to summer 2004
(left). Bias of ALADIN CROATIA precipitation
forecast (rain versus no rain) for Zagreb
Maksimir (14240), Gospic (14330) and Split Marjan
(14445) for year 2003 (below left). Mean errors
and standard deviation of ALADIN CROATIA wind
speed for Split Marjan for the year 2003 (below).
PBL height determination
Operational version of ALADIN used in Zagreb is
AL25T1_op2. PBL height is diagnosed using
modified formula of Ayotte (1996) where PBL
height is determined using virtual potential
temperature vertical integral. If the value at
the level z is significantly larger than ts
vertical integral value between this level and
the ground, PBL height is detected. However, as
can be seen in the top figure on the right, using
this formula, PBL height is detected very low in
statically stable situation. The significant
drawback of this method is that the effect of
wind shear on dynamical stability is
neglected. The virtual potential temperature
profile is modified to include wind shear impact
on stability. The value at surface remains the
same, above it, the values are modified by wind
shear.
The new values of virtual potential temperature
are used in the integral. First the deviation
between the level value and the integral value of
the virtual potential temperature is calculated.
The contribution of this layer to the PBL height
is calculated.
These values are summed up to the level where PBL
top is detected. If the wind on the lowest model
level is strong, the diagnosed PBL height are
high. In statically stable situations with weak
wind, the diagnosed values are low. The vertical
integral makes the method robust to oscillations
the model variables might have in the vertical.
The figures show forecasted PBL height before
(left) and after (right) the wind shear impact
introduction.
Root-mean-square errors of maximum temperature
for day 1 (left) and day 2 (right) forecast of
ALADIN LACE and CROATIA for direct model output
and model output statistics, for Zagreb Maksimir
(14240), from summer 1997 to summer 2004 (top
row) and the last two years (bottom row). MOS are
are done using regression equations (yaxb)
which were calculated from historic data for warm
(April to September) and cold (October to March)
part of the year.
Adriatic storm case

• On 24th March 2004 03 UTC a cyclone stroke a
  southern part of Croatian coast in the Dubrovnik
  area. Unfortunately, the movement was forecasted
  too fast and the depth of this cyclone was
  severely underestimated. Comparison of the
  forecasted mean sea level pressure with 12-km
  (red) and 8-km (orange) with measurements from
  the SYNOP (violet) stations (below). Mean sea
  level pressure forecast (below left) and analysis
  (below right) for 00 UTC 24th March 2004.

Seasonal (top row) and monthly (bottom row) mean
errors, mean absolute errors, root-mean-square
errors and skill scores (according
root-mean-square error and regression
persistency) for day 1 (left) and day 2 (right)
maximum temperature forecast of ALADIN LACE and
CROATIA for direct model output, for Zagreb
Maksimir (14240), from January 2003 to august
2004. Left The same like above, but for the day
2 minimum temperature forecast.
Bi
Gr
Du