Meteorol. Z, 22, 2013 
N.H. Schade et al.: Regional Evaluation of ERA-40 Reanalysis Data 
677 
Downloaded from www.schweizerbart.de 
Unauthorized distribution of this copyrighted material is strictly forbidden! 
literature, e.g. by Gandlin (1987), or ISHn et al. (2003), 
so they are described only briefly here: 
HQC is based on the program by HOEFLICH et al. 
(1975). In a first step, the data are checked on doubles 
(i.e. already existing data or simply data received twice 
or more often), the correct contents of each column, i.e. 
allowed values in the column “air temperatures” etc., 
and a date/time and position tracking (“cruise control”) 
by means of reported position, course and speed. For this, 
all observations have to be related to previous and subse 
quent observation times, i.e. date and time, the geograph 
ical positions, and observed values. But also comparisons 
to other observations in the vicinity of the checked data, 
spatial as well as temporal, are carried out (“vicinity 
test”). 
Climate and consistency checks follow. All data have 
to be within reasonable boundaries (minimum and max 
imum values) depending on the geographical position. In 
this “climate test”, sea level pressure as well as air and 
sea surface temperatures are compared to the marine data 
base of DWD (MARDAB), containing the number of 
observations, mean value and variation for each 2.5° or 
5° grid world wide. Correspondingly, the MARDAB 
boxes in the North Sea area include a high number of 
observations and the climate test is supposed to be reli 
able. Here, the boundaries for air temperatures are 
—25 °C and +40 °C, for sea level pressure 930 hPa 
and 1050 hPa. 
Special routines check for consistency and abidance 
by the laws of physics (e.g. wet-bulb temperature < air 
temperature). As a result of the quality control, all obser 
vations are flagged (values of 0, “no control” up to 9, 
“missing”) and, in case of an erroneous or doubtful 
value, checked manually, corrected if possible and finally 
written in file. However, manual correction can only be 
applied for a limited number of non real-time weather 
reports, which is actual 3% of the incoming data. 
2.2 Preprocessing 
All those different observational data contain specific 
errors that can not be detected by the HQC and have to 
be addressed in preprocessing: Biases known to be 
related to the composition of nations, to which the ships 
belong, instrumental random errors and sampling errors. 
ISHE et al. (2005) noted inhomogeneities in the qual 
ity of the historical observations due to missing call signs, 
ship identifications or changes in the observation meth 
ods (q.v. Cardone et al., 1990). Low pressure biases 
might be caused by “suction”, i.e. air flowing from 
inside a ship leeward to the outside (Hayashi, 1974), 
and air temperatures may be to a certain extent affected 
by the solar heating of the instruments and their sur 
roundings (Berry et al., 2004). Kent and Berry 
(2005) found random measurement errors averaged for 
each month in the period 1970 to 2002 of 1.2 ± 0.3 K 
(air temperature, with height correction) and of 
2.4 ± 0.9 hPa (sea level pressure) in the North Sea area. 
Figure 1: Distribution of the total number of air temperature 
observations in GZS in the North Sea area. Resolution: 0.5°. Base 
period: 1961-2000. 
Both estimates have been found to decrease towards the 
end of the period, assumed to be related to an increased 
understanding of the importance of positioning and qual 
ity of the instruments. Gulev et al. (2003) investigated 
VOS wave observations and found fair weather biases 
due to the fact, that ships avoid stomiy conditions. 
Observations in the North Sea, however, are “contam 
inated” by discrete shipping routes, leading along the 
coasts and around Scotland into the North Atlantic, 
displayed in Fig. 1. Shown is the distribution of air 
temperature observations in GZS on a 0.5 degree resolu 
tion from 1961-2000 that passed the HQC. Obviously, 
for most grid boxes in the North Sea area the occupancy 
is insufficient for a robust investigation. Two criteria were 
therefore chosen to select the boxes of investigation: (1) 
Enough high quality observations of good timely distri 
bution and (2) no remarkable land influence. The two 
boxes that passed these criteria are shown in Fig. 2 (left). 
Since the ERA-40 reanalysis data are stored on an irreg 
ular grid of 1.125 degree in longitude and between 1.121 
degree and 1.125 degree in latitude, all grid boxes for the 
investigations are chosen thus, that four ERA-40 grid 
points are centred in the middle. All observations were 
averaged around the 6-hourly ERA-40 output and only 
those time points for which data of both data bases exist, 
further referred to as “GZS-like sampling”, were 
evaluated. 
Comparing the number of observations in box 2, ICO- 
ADS includes about twice as much air temperature data 
in the early 60s and 70s compared to GZS, whereas 
the rest of the period is in good agreement, especially 
since the late 70s. Box 1 shows the exact opposite (not 
shown). This may be due to slightly different thresholds 
in the different quality control procedures and to a certain 
extent different processed data. Since time series