678
N.H. Schade et al.: Regional Evaluation of ERA-40 Reanalysis Data
Meteorol. Z, 22, 2013
Downloaded from www.schweizerbart.de
Unauthorized distribution of this copyrighted material is strictly forbidden!
60°N
57°N
54°N
51 °N
48°N
1
2°E
14000
12000
10000
8000
6000
4000
2000
0
Box 2, Air temperature observations
per year, ICOADS / GZS
III
1965 1970 1975 19801985 1990 1995 2000
Year
Figure 2: North Sea area showing the two investigated grid boxes (left). Both boxes consist of four sub grid boxes, each referring to one
ERA-40 grid point (grey plus). Also shown: Elistogram of air temperature observations by ICOADS (black) and GZS (grey) in Box 2
(right). Base period: 1961-2000.
of yearly mean air temperatures derived from monthly
means show a good agreement between GZS and
ICOADS in both boxes, both data sets should be compa
rable here. The same arguments apply for sea level
pressure observations which are in even better agreement.
To address biases in marine observational data, infor
mation about the metadata are needed. Kent et al. (2007)
used the WMO Publication No .47 (e.g. WMO, 1994),
which contain infomiation about the contributing VOS
ships and the instruments used. Until the late 1980s,
the thermometer type was mostly unknown, also was
the method of exposure. Therefore, no correction has
been applied. Biases related to changing measurement
heights (about 0.01 K/m) due to the fact that ships
became larger would lead to a cold bias in global air tem
peratures of about 0.07 K in 2000, relative to 1970 val
ues. In the North Sea area, the mean measurement
height changed from the 1970-1979 to the 1995-2004
period from 10 m to 20 m, which result in a 0.1 K cold
bias (or 0.18 K, if corrected at 2 m). A height correction
would therefore reduce the random measurement error by
Kent and Berry (2005) to 1.1 (1.0) ± 0.3 K.
Further, there is a variety of nations contributing to
GZS, associated with specific biases. Some Russian ships
are known to deviate about 2-3 K in air temperature
observations (personal communication). Most of these
substantial biases should have been sorted out by the
“vicinity test”, unless there are no neighbouring observa
tions. For the investigation only those observations were
used that appear to be inconspicuous according to the
HQC. Furthermore, the GZS data in both boxes were
manually evaluated concerning possible biases caused
by single vessels and fixed positions, but showed no
noticeable inhomogeneities as to space and time, in spite
of the fact that some vessels and platforms (prominently
the Gemían fishing trawler “Walther Herwig III” and the
Dutch oil rig “Maersk Endeavour”) in Box 2 have sub
mitted a relatively high number of observations. Since
Table 1: Total Sampling Error (TSU) for air temperatures (AT) and
sea level pressure (SLP) for both North Sea boxes for the whole
period 1961-2000 and the respective summer (August) and whiter
(December) month.
Box 1
Box 2
TSU AT
±0.2 K
±0.36 K
TSU AT (summer)
±0.32 K
±0.25 K
TSU AT (winter)
±0.15 K
±0.56 K
TSU SLP
±0.3 liPa
±0.47 hPa
TSU SLP (summer)
±0.18 hPa
±0.34 hPa
TSU SLP (whiter)
±0.51 hPa
±0.77 hPa
only time averaged values have been further used, and
the exclusion of those vessels shows no substantial differ
ences in the results, this fact can be considered negligible
for our investigations (which of course does not mean it
should be assumed to apply for other boxes and areas as
well without prior investigation).
Concerning sampling errors we followed the method
of GULEV et al. (2007) and compared monthly means
of our GZS-like sampled ERA-40 data with regularly
sampled ERA-40 data to obtain the total sampling uncer
tainties (TSU). These would be inherent in both datasets
compared in this study. Results are shown in Table 1:
TSU of air temperatures for the whole period 1961—
2000 are within = 0.2 K for Box 1 and = 0.36 K for
Box 2 with higher values in the summer (winter) in
Box 1(2). Furthermore, the TSU winter values for
Box 2 are slightly increasing towards the end of the per
iod, whereas they remain constant in summer and
throughout the whole period in Box 1. Sea level pressure
data show TSUs of ± 0.3 hPa for Box 1 and = 0.47 hPa
for Box 2. Here, TSU is close to zero since the late 80s in
Box 1 but shows an increase up to = 1 hPa in winter in
Box 2. Since we investigate GZS-like sampled
ERA-40 data with GZS data, these sampling errors are
not accountable for any differences between the two data