Verification of the dynamical downscaling of ERA40 by ALADIN

1
Verification of the dynamical downscaling of
ERA40 by ALADIN
Nedjeljka Žagar1, M. Žagar2, J. Cedilnik2, G.
Gregoric2 and J. Rakovec1
1Chair of Meteorology, University of
Ljubljana 2Meteorological office, Environmental
Agency of Slovenia
2
Motivation
Modelling the wind climatology for Slovenia at
scales around 10 km This is an intermediate step
towards a higher-resolution climatology for the
wind energy estimates
This talk ? optimal modelling setup for the
dynamical downscaling of ERA40 ? comparison with
observations during MAP-SOP (autumn 1999) and
with the ECMWF reanalyses of the MAP-SOP period
(KeilCardinali, QJRMS, 2004)
27th EWGLAM meeting
4 October 2005
3
Methodology
? Dynamical downscaling of ERA40 (10-year period)
? Mesoscale model a 10-km ALADIN model
10-km ALADIN
?120-km ERA40
27th EWGLAM meeting
4 October 2005
4
Direct vs. two-step nesting
ERA40 ? 30-km ALADIN ? 10-km ALADIN
ERA40 ? 10-km ALADIN
27th EWGLAM meeting
4 October 2005
5
Increasing spatial variability
An example building-up of the kinetic energy
spectrum during the first hours of the forecast
Model level 20, time 12
Model level 20, time 0
Model level 10, time 12
100 km
Model level 10, time 0
27th EWGLAM meeting
4 October 2005
6
Average spectra shape
Average kinetic energy
Average vorticity
Average divergence
Lowest model level
100 km
Average over 70 days of MAP-SOP and for model
levels 8-18 (4-9 km)
27th EWGLAM meeting
4 October 2005
7
Below 10 km
? Dynamical downscaling of ERA40 ? Mesoscale
model a 10-km ALADIN model ? Pure dynamical
downscaling (ŽagarRakovec, 1999) to 2.5 km
2.5-km ALADIN
10-km ALADIN
27th EWGLAM meeting
4 October 2005
8
Methodology
Define the optimal model setup by ? comparison
against observations, and ? comparison against
MAP-SOP reanalyses
40-km MAP-SOP reanalyses
10-km ALADIN
27th EWGLAM meeting
4 October 2005
9
exposed mountaintops
3-4 m/s
valleys, basins and lowlands
Adriatic coast
1-2 m/s
2-3 m/s
27th EWGLAM meeting
4 October 2005
10
exposed mountaintops
3-4 m/s
valleys, basins and lowlands
Adriatic coast
1-2 m/s
2-3 m/s
27th EWGLAM meeting
4 October 2005
11
Observed temporal variability
Frequency (cycles/day)
48
96
6
8
12
24
Period (hours)
Based on the time serie of the zonal wind during
MAP-SOP, data every three hours, totally 560 data
points
27th EWGLAM meeting
4 October 2005
12
Observed temporal variability
Frequency (cycles/day)
LTD D SD
86 12 2
exposed mountaintop station RO
27th EWGLAM meeting
4 October 2005
13
Observed temporal variability
Frequency (cycles/day)
LTD D SD
86 12 2
exposed mountaintop station RO
38 26 36
coastal station PO
27th EWGLAM meeting
4 October 2005
14
Observed temporal variability
Diurnal periods (26-22 hrs)
Subdiurnal periods (6-22 hrs)
Longer than diurnal periods (lt26 hrs)
Frequency (cycles/day)
LTD D SD
86 2 12
exposed mountaintop station RO
38 26 36
coastal station PO
valley station SG
52 7 41
27th EWGLAM meeting
4 October 2005
15
Specific questions
How well are these spectra reproduced by an
operational LAM?
Can we relate modelled spectra with traditional
verification scores (mae, rmse)?
How dependent are model-produced spectra on the
nesting method and the horizontal domain size?
How much better is the wind field from the 10-km
ALADIN than from the ECMWF model (40-km MAP-SOP
reanalyses) ?
27th EWGLAM meeting
4 October 2005
16
Modelled spectra
Exposed mountaintop station RO
27th EWGLAM meeting
4 October 2005
17
Modelled spectra
Coastal station PO
27th EWGLAM meeting
4 October 2005
18
Modelled spectra
Valley station SG
27th EWGLAM meeting
4 October 2005
19
Subdiurnal power range
mountain stations
coastal station
valley stations
27th EWGLAM meeting
4 October 2005
20
Longer than diurnal power range
mountain stations
valley stations
coastal station
27th EWGLAM meeting
4 October 2005
21
Traditional score comparison
3 ALADIN domains
Mean absolute error for various ALADIN-10
horizontal domains
24-h persistence
subdiurnal range
27th EWGLAM meeting
4 October 2005
22
Subdiurnal varibility a 12-h filter
Mean absolute error
Filtering improved scores for MAP-SOP
27th EWGLAM meeting
4 October 2005
23
Wind roses impact of a 12-h filter
OBS
ALADIN
MAP-SOP
After filtering
27th EWGLAM meeting
4 October 2005
24
Conclusions

• ? Downscaling of ERA40 by the ALADIN model - ok
• Conventional verification scores a 10-km ALADIN
  superior to the MAP-SOP reanalysis data
  expected
• However,
• More energy in the sub-diurnal (SD) range in the
  MAP-SOP data than in the ALADIN less expected,
  peculiar
• Spatial spectra slope does not follow 5/3 law
• Direct nesting (resolution jump 120?10 km)
  produced best conventional scores !?
• Smoothing MAP-SOP dataset improves its
  conventional scores
• Consequences
• The mesoscale model wind field displays
  significant senstivity with respect to the
  horizontal domain definition basin stations
  contain 30-50 of energy in the SD range

27th EWGLAM meeting
4 October 2005
25
Questions

• In dynamical downscaling, spatial variability
  increase is much more important than the increase
  of the temporal variabiliy?
• What shift from low frequency to high-frequency
  variability can/should be expected?
• -5/3 slope how strong constraint should it be?

27th EWGLAM meeting
4 October 2005