Meteorol. Z, 22, 2013 
N.H. Schade et al.: Regional Evaluation of ERA-40 Reanalysis Data 
681 
Downloaded from www.schweizerbart.de 
Unauthorized distribution of this copyrighted material is strictly forbidden! 
Northern Europe, especially during NAO + phases in 
winter. It also should be noted that ERA-40 and the 
observations differ most at SLP values between 
1010 hPa and 1017 hPa in December. 
At last the ATs are evaluated for two different refer 
ence periods 1961 to 1980 and 1981 to 2000 (see 
Fig. 5) to investigate possible temporal changes. Box 2, 
which is closer to the coast, shows an increasing spread 
in the whole frequency distribution as well for the obser 
vations as for the ERA-40 data. However, there is hardly 
any difference of the median between the two periods for 
the ERA-40 data. Over the open sea (Box 1), only minor 
changes in the percentile values of both data sets can be 
detected. A closer look into the summer (JJA) and the 
winter (DJF) seasons reveals temperature increases in 
both boxes for both data sets and both seasons for the 
1981-2000 period compared to the 1961-1980 period: 
Slightly higher temperatures in all median values and 
an increased spread, but smaller minimum temperatures 
for DJF and higher maximum temperatures for JJA can 
be found in the southern Box 2, and a small overall 
increase in the higher percentiles and almost no changes 
in the median over the northern Box 1. The ERA-40 data 
show higher JJA maximum temperatures and lower DJF 
minimum temperatures than the observations, and also a 
smaller increase in the median. The differences between 
ERA-40 and GZS JJA maximum temperatures should be 
even higher when taking potential radiative heating errors 
and the uncertain fraction variability in GZS into account. 
The SLP data have no such differences between the 
two reference periods in neither of the data sets. The fre 
quency distributions of the percentile values suggest on 
one hand that the large scale circulation patterns have 
not or at least not noticeably changed. On the other hand 
it suggests that the SLP is reproduced well by the 
ERA-40 reanalysis system in the North Sea area in the 
investigated time periods. 
4 Summary and conclusions 
For the validation of regional climate models, high qual 
ity observations and/or reanalyses (i.e. ERA-40) data are 
needed as reference of the actual climate state. Compar 
isons of marine atmospheric in-situ observations with 
ERA-40 show a systematic cold bias for the air temper 
ature near the Danish coast in the reanalyses data for 
the winter months (DJF). This apparent land-induced bias 
does not appear in the analysis for sea level pressure, 
probably due to its larger scale and land-sea insensitivity. 
The bias may be caused by the interpolation process of 
ERA-40 from the spherical calculations to the final grid. 
Overall, ERA-40 agrees well with the observations of sea 
level pressure and, off the coast, also with the 2 m air 
temperature results. Near the coasts, one should be care 
ful using ERA-40 air temperature data. The cold bias 
compared to GZS temperatures could be partly related 
to measurement errors, prominently by radiative heating, 
and the uncertain fraction variability. It still should be 
accounted for, since ERA-40 itself is biased to the same 
degree near the coasts compared to sea-facing boxes. It 
should be further noted, that these results refer only to 
the two parameters air temperature and sea level pressure 
and base on two boxes in the North Sea area only. If a 
regularly sampled reanalysis product is compared with 
in-situ observations, the sampling bias has to be taken 
into consideration also. 
A higher resolution of the reanalyses data set might 
reduce the land-influenced bias and improve the quality 
of air temperature results near the coast. This would 
provide a better reference for subsequent high resolution 
modelling, coastal engineering, etc., which will become 
more and more important in the future. Further investiga 
tions are currently under way supporting our thesis: An 
evaluation paper, comparing amongst others ERA-40 
2 m air-temperatures with the ENSEMBLES RCM 
hindcast runs driven by ERA-40, is in preparation for 
KLIWAS. It indicates that the higher resolution of the 
regional models improves the 2 m air temperature results 
near the coasts. I.e., ERA-40 shows colder values in win 
tertime compared to the RCMs, too. Further steps have 
been undertaken by Saha et al. (2010) with the new 
NCEP Climate Forecast System Reanalysis (CFSR), 
using a global coupled atmosphere-ocean-land surface- 
sea ice system with an output of an hourly time resolution 
and a horizontal resolution of 0.5°. Validations in the 
same way as done here may hint on a better way of 
the development of a regional product. 
Also, the development of a regional coupled Atmo 
sphere-Ocean reanalysis has been discussed and might 
be implemented in future studies. At the moment, the 
Hans-Ertel Centre for Weather Research (HErZ) is con 
ducting a retrospective analysis of regional climate at the 
Meteorological Institute, University of Bonn (MIUB) and 
the Institute for Geophysics and Meteorology>, University 
of Cologne (IGMK). This regional reanalysis is based on 
the COSMO-EU(DE) model with a horizontal grid spac 
ing of 7(2.8) km resolution, as it is in operational use at 
DWD in the forecasting model suite. A two step nesting, 
using ERA-Interim (SIMMONS et al., 2006) as boundary 
conditions, is performed. Two main periods of the reanal 
ysis will focus on (1) a comparably short time frame of 
5 years (2007-2011) with the maximum amount of 
observational data, and (2) the past decades (1982— 
2011) with a reduced data basis, in order to aim at more 
homogeneous time series than typically available in long 
term reanalyses (KELLER et al, 2012). First results will 
be available soon. 
Our analysis clearly indicates the importance of in-situ 
observations over the sea to serve as validation set not 
only for reanalyses, but also for global and regional cli 
mate models. Even a sparse observational data base can 
help to the discovery and definition of problems in the 
models, if one accounts for the sampling bias. Neverthe 
less, pseudobiases and measurement errors are still