Fróðskaparrit - 01.01.2007, Side 143

SAMANBERING AV ALDUFORSØGNUM FRÁ ECMWF VIÐ ALDUMÁTINGAR
GJØRDAR Á FØROYALEIÐINI
141
more energetic peak frequencies, and a
slightly narrower peak, one such example
is given in the upper plot in Figure 7.
Discussion
Before going into the detailed discussions
of the fit of the modelled wave parameters
at the different locations, a closer look into
the effect of the cut-off frcqucncy flu rh on
the measurements is needed. Using all the
measured wave spectra at WVD-4 (17196
measurements spanning from 10/2-1999 to
13/2-2004) as a reference database, it was
investigated what the effect of different
values for f were on the wave parame-
ters H g and Tm02. The inspection showed
that fhj h=0.50 Hz, corresponding to fhigh at
WV-2, introduced a positive bias (artificial
overshoot) of 0.43 s in Tm02 and a negative
bias in Hm0 of -0.01 m. \ífhjg=Q25 s, corre-
sponding to fhh at K7, the artificial posi-
tive bias in T would be 1.43 s and the
bias in H 0 would be -0.09 m. The parame-
ters from the WVD-4 site are, as men-
tioned earlier, compensated for the effect
of (. , whereas the measurements at WV-2
•'lugh
and K7 are not. When compensating for
Ihc effect frorn f. , on T ,,,-measurements
at WV-2 and K7 it becomes apparent that
the true 7^-bias is in the region of 1 s at
all sites for most of the events. It is there-
fore clear that the EW4 rnodel seems to
consistently over-predict the Tm02 wave pe-
riod.
Due to inherent random properties in
wind and wave fields, there is a lower limit
to the scatter index of wave parameters.
For the wave height this lower limit of the
scatter index, lies on average in the region
of 10-20% (Kornen et al., 1994). As re-
ported by Cardone et al. (2000), a gener-
ally accepted high skill level for continu-
ous hindcasts of Hm0 is represented by a
scatter index of less than 0.25 associated
with bias of less than 0.25 m and a correla-
lion of 0.90 or greater.
Looking at Table 4, and comparing it to
otlier operational models (Bidlot et al.,
2002; Nielsen, 2002, 2003a, 2003b, Jans-
sen et al., 2005), it is clear that the EW4
model prediction is good for all locations
and events.
When this is said, it is quite clear that
the modelled wave height performance is
seasonally dependent, having larger nega-
tive bias and smaller skill index in the win-
ter events. This seasonal performance vari-
ation in H „ of the EW4 mođel is also doc-
umented in Bidlot et al. (2002). Due to the
skill of the present wave models, poor
inodel performance says rnore about the
driving wind fíelds than the model itself
(Komen et al., 1994). This claim seems in
the present case to be supported by the
general under prediction of the wind fields
observcd at K7 in the winter events (Figure
2).
Another trend which is visible in all
events is the under prediction of the wave
height, in the peak of the storms. This
problem is well known from different ap-
plications, and the rnajor source for error
can usually be traced back to deficiencies
in the forcing wind field (Cardone et al.,
1996).
The rnodel prediction of Hm0 and T are
not affected by the same order of consis-
tent bias as it is seen in T and the bias of
m02