Data & Methods
The methods used in this study follow those used by Moberg
et al. (2006) and Chen et al. (2006). A brief description is
given below, while detailed information can be found in
these references, except for the regional division used here.
For the regional analysis, we followed the regional divisions
used by the recent IPCC special report (IPCC 2012), rather
than the approach used by Moberg et al. (2006) and Chen
et al. (2006).
EMULATE collected a dataset containing daily tempera-
ture and precipitation observations over European locations
having data starting before 1901. This dataset is the basis for
the analyses in this atlas. Up to four daily climate variables
are available: minimum and maximum temperature (Tmin/
Tmax), mean temperature (Tmean), and precipitation (Prec).
We focus on three periods: 1801-2000, 1851-2000, and
1901-2000. Since a trend analysis is extremely sensitive to
the starting and ending of the time series, particular attention
is paid to the missing data in these beginning and ending pe-
riods. A data completeness criterion has been applied to filter
out stations with insufficient data for the proposed analyses.
We divided each analysis period in question into three sub
periods: one 20-year period at the beginning, one 20-year pe-
riod at the end of the series and the entire period in between
(160, 110, or 60 years according to the period analyzed). A
station passed the filter if it did not have more than 4 % of
missing values in the two 20 year blocks at the beginning and
end and at most 6 % missing values in the longer block in be-
tween. All the stations that passed the criterion and are used
in this work are listed in Table 2.1 and shown in Figs. 2.1 ,
2.2 and 2.3 .
For the 200-year period (1801-2000) we can use only
three stations with Tmin/Tmax, and seven stations with
Tmean measurements. No precipitation observations are
available for this period. Looking at the 150-year period
(1851-2000) the number of observations is increased to nine
for Tmin/Tmax, thirteen for Tmean, and nine for precipita-
tion. For these two periods the analyses were carried out for
each station with sufficient data. The number of stations for
these two periods is too few to undertake any averaging or
regionalization approaches. For the 100-year period (1901-
2000) significantly more stations are available: we have 57
observation sites for Tmin/Tmax, 54 for Tmean, and 100 for
Precipitation. The stations are fairly unevenly distributed
over the study area, although we find the highest station den-
sity in Central Europe. The relatively high number of obser-
vation sites for this period provides the possibility to group
the stations into regions, which enables a regional analysis
and inter comparison between regions.
Using regional divisions for the regional analysis fa-
cilitates direct comparison of the regional trends with other
estimates and model simulations. As indicated in Figs. 2.1 ,
2.2 and 2.3 , the following three regions were created:
NEU (Northern Europe), CEU (Central Europe), and SEU
( Southern Europe).
The climate indices described in the next section were
computed for all stations belonging to a particular region.
Afterwards the time series have been averaged arithmeti-
cally. The resulting averaged time series was taken as the
regional mean and was subject to a linear trend analysis over
the whole 100-year period.
The quality of the station series selected for this study
varies. All series have been corrected for obvious errors
whereas efforts towards homogenizing the database could
not be undertaken. Relatively few series have undergone in-
tensive testing and correction for inhomogeneities, among
them the very long records for some Swedish stations (Mo-
berg and Bergstrm 1997). Some of the series have been
used more frequently than others in the literature and are in
this context also more quality controlled. We have to keep
in mind that inhomogeneous data can cause errors in esti-
mated trend values as demonstrated by Venema et al. (2012).
In particular with respect to the method of linear regression,
which is more sensitive to values at the beginning and end
of the respective time series analyzed for linear trend. Dur-
ing the recent past substantial efforts have been undertaken
to improve the availability and quality of long-term climate
observations. For example, 557 monthly long-term observa-
tion series for the Greater Alpine Region where collected,