Fróðskaparrit - 01.01.2007, Blaðsíða 135

SAMANBERING AV ALDUFORSØGNUM FRÁ ECMWF VIÐ ALDUMÁTINGAR
GJØRDAR Á FØROYALEIÐINI
133
The last parameter is the skill index (also
known as reduction of variance);
Ík-^)2
S*/ = 1_J=L_----- (5)
SIO-O;)2
/ = 1
The bias value is good to reveal general
trends in the model performance, that is if
the model consistently over- or under-pre-
dicts the measurements. It can nevertheless
be misleading in cases where both over-
and under-estimations occur. The scatter
index does not reveal general trends, but is
good to evaluate the average model per-
formance. The correlation coeffícient is +/-
1 if there is a consistent linear relation be-
tween m and o, and 0 if m and o are inde-
pendent. If the correlation coefficient is
close to one this suggests that there is good
correlation between model and data (low
random error), but it does not mean that
there is no systematic error. The skill index
has similar properties as the scatter index,
but is included since it in many cases is
more sensitive than the scatter index. If
there is a good fit between model results
and data, then bias and scatter index will
tend to zero, whereas the skill index will
tend to one. If the skill index is less than
zero, this indicates that the measured mean
value of a parameter represents the rneas-
ured time series better than the modelled
parameter tirne series.
Results
The wind series from the K7 (Figure 2)
suggest that the ECMWF wind speed is
underestimated in the two winter events (2
and 4), whereas the ECMWF wind speed
correlates better with measurements in
Event 3 (if the modelled values at 10 m
height are expected to be 10-15% higher
than the measurcd values at 3 m height).
As expected, both wind speed parameters
give identical results at this location (K7)
since the irnpact of different inteipolations
is very small for point away from land.
Note also that wind observations from K7
are provided to the atmospheric model as-
similation scherne.
The measurements from the F8 station
show good correlation to the ECMWF
model winds (Figure 3). The clear differ-
ence in data-fit between winter and sum-
mer events seen in the K7 series is not ob-
vious at F8. The two model parameters
give slightly different wind speeds at this
location due to the differences introduced
by interpolations, but visual inspection of
the series does not find one of thern supe-
rior to the other.
The time series from locations WV-1
and WV-3 (omitted here) show surpris-
ingly good correlation between measured
and EW4 wave parameters, considering
the coarse resolution of the EW4 rnodel
and the often sheltered position of the
buoys. The measured time series do never-
theless reveal considerable fluctuations in
the wave height, which most probably are
due to the mooring issue discussed in
Niclasen and Simonsen (2005b). These se-
ries are thereforc not included in the EW4
validation.
The time series from the WV-2 and K7
buoys show good correlation with the