Low (web) Quality - BALTEX

83
Does dynamical downscaling with regional atmospheric models add value to
surface marine wind speed from re-analyses?
Jörg Winterfeldt and Ralf Weisse
Institute of Coastal Research, GKSS Research Centre, Geesthacht, Germany, ralf.weisse@gkss.de
1. Assessment of added value using buoy wind
The hindcast surface marine wind speed fields from the
regional atmospheric model REMO in two configurations
and the regional atmospheric model CLM are investigated
with regard to their added value in comparison to the driving
wind field from the NCEP/NCAR reanalysis (NRA_R1). To
do so in a first step wind speed measurements from buoys,
light ships and platforms in the eastern North Atlantic are
considered as “truth”. Added value from the regional models
is obtained when correspondence with both the measured
statistical distribution and instantaneous wind speeds is
higher than that of the reanalysis. Wind speed fields from
NRA_R1, REMO and CLM are bilinearily interpolated to
measurement locations and statistically compared.
Winterfeldt and Weisse (2009) demonstrated that for
instantaneous wind speeds the regional models do not have
an added value both in "open ocean" areas and the German
Bight. However, in the English Channel, where local
topography and associated local wind regimes become
important, the regional model shows an added value for
instantaneous wind speeds.
Concerning the wind speed distribution there's a clear
indication for an added value of the regional models in
coastal regions, especially for higher wind speed percentiles,
while in "open ocean" areas the NRA_R1 is better reflecting
observed distributions.
2. Assessment of added value using QuikSCAT
wind speed retrievals
The added value of the dynamically downscaled wind was
assessed with satellite data, namely QuikSCAT Level 2B
12.5 km (L2B12) wind speed retrievals following the
thorough validation of the quality of L2B12 data with buoy
winds in the eastern North Atlantic (RMSE: 1.5 m/s) by
Winterfeldt et al. (2009). For that purpose L2B12, REMO
and global NCEP/NCAR reanalysis (NRA_R1) data were
co-located for the years 1999-2007. Co-location criteria
between REMO and L2B12 data are within 0.1° and 0.06° in
longitudinal and latitudinal distance from REMO model grid
points and within 20 minutes, leading to a high quality
L2B12 gridded wind speed data set with REMO grid
dimensions. NRA_R1 data was interpolated in time and
space onto the REMO grid.
Figure 1. Modified Brier Skill Score calculated from colocations
between QuikSCAT L2B12, NRA_R1 and SN-
REMO (SN stands for use of spectral nudging) in the wind
speed range from 3 to 20 ms-1 and the years 2000 to 2007,
where QuikSCAT L2B12 serves as ``truth``, NRA_R1 as
reference ``forecast`` and SN-REMO as ``forecast``. Blue
areas indicate value lost, while red areas indicate value
added by dynamical downscaling.
3. Seasonal variability of added value
Strong interseasonal differences exist, in winter enhanced
cyclonic and meso-cyclonic activity increases the potential
of dynamical downscaling. In winter time the added value
is more pronounced around Iceland and Greenland, south
of Iceland and within the Gulf of Lyon/Mistral region (see
Fig. 2 and 3).
Summarizing QuikSCAT is a valuable tool to identify
marine regions where dynamical downscaling of wind
speed makes sense.
A slightly modified Brier Skill Score (BSS) was used to test
to what extent the regionally modelled wind gives a better
reproduction of QuikSCAT wind speed than the NRA_R1.
The modified version of the BSS varies between -1 and +1
and simplifies the comparability of positive (value added)
and negative (value lost) scores.
Fig. 1 confirms the point stated by Winterfeldt and Weisse
(2009) for a wide area including the eastern North Atlantic,
the Baltic, Mediterranean and Black Seas: dynamical
downscaling does not add value to NRA_R1 wind speed in
open ocean areas (blue), while it does for complex coastal
areas (red).
Figure 2: As Figure 1 but for Summer (JJA).