Within the EC-project ALPCLIM (Environmental and Climate Records from High Elevation Alpine Sites) about 120 Central-European temperature station data of up to 240 years length were homogenised on a monthly base. For the procedure of relative homogeneity testing we used a multiple Craddock-curve intercomparison method, the MASH-test and a Bayesian rule based test of Meteo-France. Our experience shows that high-quality homogenisation does not depend on the statistical method alone but need a good combination of the statistical method, a powerful software system and detailed meta-data information. Comparison of homogenised data with raw data shows that the raw data are biased by an increasing trend of a too warm temperature level back in time. This can be explained by problems with screening and non-standardisation at the begin of temperature measurements, an overall trend due to relocation of climate station from city centres to more rural sites (airports) in the 1940ies and 1950ies as well as for the Austrian climate network by a change in observation time from 1970 to 1971. For easier data processing homogenised station data were interpolated to a 1deg grid for the region 43S/4E to 49S/18E. Comparison of ALPCLIM temperature data with the gridded temperature data of CRU (Climate Research Unit, Univ. East Anglia) confirms results of other studies that the CRU data set is not useful for describing regional climate variability.

High quality temperature data sets (station data or gridded data) of different spatial scale (regional to global) are sparse though they are very important in climate research. Moreover, only the minor number of investigators deal with the problem of data homogeneity. However, a lot of statements about climate change and also the outputs of GCM experiments are based on these data sets. On global scale and also on regional scale the data most widely used are the gridded temperature data of CRU (Climate Research Unit). Though the authors of this data set give some attention to the question of data homogeneity (Jones, 1994) the results of comparisons with high-quality homogenised regional data sets (eg. Moberg and Alexandersson, 1997, Auer a.o. 2000) show some remarkable differences in temporal trends. As a result of these comparison studies we conclude that the CRU temperature data set is able rather to describe hemispheric to global temperature trends than regional temperature variability.

A second important and widely used temperature data set is the station data set of GHCN (Global Historical Climatology Network, Peterson and Vose, 1997). This global data set is not homogenised but passed some quality control procedures (Peterson a.o., 1997). Consequently, for climate variability studies data from this data base have to be homogenised first. Other temperature data sets are on regional scale only. The minor number of them are carefully checked for homogeneity.

Central Europe is in the favourable position of a dense and long term climate network. Especially for air temperature instrumental data go back to the 18th century. In addition, for several countries this long-term data set is supported by detailed meta data information. Within EC-project ALPCLIM the high potential of central European instrumental temperature data should be used for calibration of proxy temperature information of high Alpine ice core data (dO18, dH2). Moreover the larger scale of a gridded temperature data set should be used for investigation of area representativity of the high alpine temperature signal. Following from the above mentioned aims of ALPCLIM and statements concerning homogeneity of available temperature data sets it was necessary within project ALPCLIM to homogenise a central European data set and interpolate it to grid points for the region 43S/4E to 49S/18E.

The data set used within this study is described by Figure 1 and Figure 2. Data sources are the respective data holders as well as for single series the GHCN data set. The longest series available started in 1750 (Basel, Genf) about 10 series go back to 1800. Most of the temperature series started between 1850 and 1900 (see Figure 2). Finally the ALPCLIM raw data set included 120 series of monthly air temperature data most of them longer than 100 years.

The method of homogenisation used at ZAMG (Central Institute of Meteorology, Vienna) was described in detail in several papers (eg. Auer, Böhm and Schöner 1999). The main features of this method are:
- the input of meta data information
- not to assume any series to be homogenous
- to fill gaps after homogenisation procedure
- to use a multiple-series comparison method for break point detection (Multiple Craddock curves intercomparison method and MASH-test; see Auer, Böhm and Schöner, 1999 and Szentimrey, 1999 for details)
- that homogenisation is done independently for regional groups on a monthly base

The western part of French temperature series were homogenised by a method described by Mestre (1999). From initial 120 digitised series 97 were found to be homogenisable. In Figure 3 the difference between the raw data and the homogenised data are shown for the examples of the mean of the Italian and the Austrian stations (mean of all stations of each subgroup respectively). These two subsets were selected for further comparison due to their high quality meta data information of the respective country.

Several clear features can be seen in Figure 3:
- the period up to about 1860 with highest values for breaks; these high values result on the one hand from the low number of stations for computation of mean value on the other hand from the low level of standardisation of meteorological networks (note that the first congress of WMO with important support for climate network standardisation was held in Vienna in 1873); moreover this first part of temperature measurements is most probably characterised by problems of screening of thermometers
- a conspicuous change in the 1940ies to 1950ies where many stations in Italy and Austria moved from city centre sites to airport or rural sites (for Italy these relocations continued also up to the 1970ies)
- the break in 1971 for the Austrian stations due to the change in observation time (from 7-14-21 to 7-14-19).

This systematic computation of inhomogeneity adjustments clearly indicates that the contribution of inhomogeneity function can not be filtered out by computation of multiple series means. Moreover we can conclude from Figure 3 that temperature increase of non-homogenised climate series of Italy and Austria since 1850 is to weak by about 1 K.

For easier data handling and for climate modelling purposes ALPCLIM station data were interpolated to a 1 to 1 degree latitude longitude grid for the area of 43S/4E to 49S/18E. For interpolation a Gaussian filter function was selected with filter weight 1 for distance 0km decreasing to filter weight 0.1 for distance 200km. Some additional limiting features should avoid unrealistic interpolation results:
- for each grid point interpolation is truncated at the first year of the longest series within 200km distance
- interpolation is interrupted at the Alpine main divide, at 1500m level of elevation and at the border between coastal and continental regions (see Figure 4).

Figure 5 shows the derived ALPCLIM gridded temperature data sets (one for low-level sites (<1500m) and one for high level sites(>1500m)). Data are relative values based on the 20th century mean (period 1901-1998). The ALPCLIM data set is available from the ZAMG free of charge.

As a first application of the ALPCLIM temperature data set we compared its temporal trend with that of the CRU temperature data set. As the CRU data set is based on 5 to 5 degree latitude-longitude grid boxes the ALPCLIM grid points have to be computed to the same area extent first. However, this is possible only for two CRU grid boxes with a coverage of same area of 80% (5-10E/45-50N and 10-15E/45-50N). All other CRU grid boxes have a lower common area coverage with derived ALPCLIM grid boxes. We selected the grid box 10-15E/45-50N for the following intercomparison study.

In Figure 6 the temporal trend of the CRU grid box 10-15E/45-50N is compared to the ALPCLIM 10-15E/45-50N grid box. As an independent measure of air temperature for the same region also a modelled temperature curve is included in the figure (mean East Alpine column temperature derived from air pressure measurements using barometric height equation, see Böhm a.o., 1998 for details).

The figure shows very clear the too weak trend of CRU temperature series in the period since about 1950. This result of ALPCLIM temperature series is well confirmed by the modelled temperature series. The ALPCLIM data set confirms other investigations mentioned earlier that the CRU temperature data set is not able to describe regional trends.

The paper deals with the experiences of homogenisation of a central European ALPCLIM temperature data set. Results of homogenisation underlines the importance of this data pre-processing part for climate variability investigations. It came up quite clear from this investigation, that quality of homogenisation does not depend on a powerful statistical method alone but on a good combination of meta data information with efficient software including mathematical background. Temperature data are systematically biased and computation of several stations mean can not filter out the inhomogeneity part of series. Comparison of ALPCLIM temperature data set with CRU temperature data set clearly shows that the CRU data set is not able to describe climate variability for central Europe. Findings of our study emphasis the postulation of further homogenisation work of regional to global climate data sets - not only for other regions but also for other climate elements.

Temperature data were provided by DWD (Offenbach), Meteo Swiss (Zürich), MHSC (Zagreb), SHMI (Ljubljana), HMS (Budapest), SHMI (Bratislava), HA-APB (Bolzano/Bozen), Soc. Met. Subalpina (Torino), UEA-CRU (Norwich) and NOAA-GHCN (Asheville). The study was supported by EC-project ALPCLIM (contract ENV4-CT97-06389).

Auer, I, R Böhm and W. Schöner, 1999: ALOCLIM - Austrian - Central European long-term climate - creation of a multiple homogenized long-term climate data-set. In: Proceedings of the 2nd seminar for homogenisation of surface climatological data. Budapest, 9-13 Nov. 1998, WCDMP-No.41, WMO-TD No.962, 47-71

Auer, I, R Böhm and W. Schöner, 2000: Long climatic series from Austria, In: Jones PD a.o. (Ed): 2nd International Climate and History Conference, 7th -11th Sep. 1998, Norwich, in press

Böhm, R, I Auer, W Schöner and M Hagen, 1998: Long alpine barometric time series in different altitude as a measure for 19th/20th century warming. Reprints of the 8th Conference on Mountain Meteorology, Flagstaff, Arizona, AMS, Boston, 72-76.

Jones, PD. 1994. Hemispheric surface air temperature variations: a reanalysis and an update to 1993. J.Climate, 7, 1794-1802.

Mestre O. 1999. Step by step procedures for choosing a model with change-points In: Proceedings of the 2nd seminar for homogenisation of surface climatological data. Budapest, 9-13 Nov. 1998, WCDMP-No.41, WMO-TD No.962, 15-26

Moberg, A and Alexandersson H, 1997: Homogenization of Swedish temperature data. Part II: Homogenised gridded air temperature compared with a subset of global air temperature since 1861. International Journal of Climatology 17, 35-54.

Peterson, TC and RS. Vose, 1997: An overview of the Global Historical Climatology Network temperature data base. Bulletin of the American Meteorological Society, 78, 2837-2849.

Peterson, TC, RS Vose, R Schmoyer, and V Razuvaev, 1997: Quality control of monthly temperature data: The GHCN experience. International Journal of Climatology, submitted.