An organisation called "Czech National Climatic Programme" deals among others by research of climatic change in the Czech Republic region. This programme consists of the following parts:

•    Climate monitoring
      -    Measurements (observation)
      -    Homogenisation of climate series
      -    Homogenisation of climate series
•    Scenarios of climate changes (testing and modifying of different world models)
•    Impacts (analysis of assumed climate change influences on agriculture, forestry, water economy, human being etc.).
•    The presented paper gives some experiences and results getting at solving the task to homogenise 30 Czech precipitation
      time series in period 1961 - 1998.

Software MASH v. 1.0.1 was used ([1]). This method is a relative homogeneity test. For our purposes we used a variant, which does not presumption of homogeneity of reference series. The mentioned software can compare at most 10 station at the same time. One of them is tested. The software algorithm chooses combination at most 4 the other series as a reference series. More references series is offered, obviously.

It has to be noted: Test results of used homogenisation method are exact from a statistical point of view. Practical use of this method contains a certain subjectivity. We think it is useful to tell something about our practical experiences with the use of this method, as followed:
•    50 stations was used
•    Cumulative model with critical value 16.64 ([1]) was used.
•    All stations were formed into 8 groups from a view of their geographic position. (Fig. 1).
Some stations were included to more groups. Altitudes of stations are shown by Figure No. 1a.
•    For each group an artificial series (REF001...REF008) was made, as an average from data of given group.
•    A homogeneity test of annual data was made at first. If software judged the series as homogenous, the series was signed as homogenous without further correction. In the opposite case were made tests of quarterly totals. Corrections were made on the base of those tests. After some experiments we left an idea to correct individual months on the base of monthly precipitation totals tests. Precipitation totals have too large variability. Corrections founded for quarter was used for every single month of this quarter.
•    Metadata. We have tried to respect metadata in the highest measure. If we found statistical significant inhomogeneity we have made correction. But in more cases we did not find a cause of inhomogeneity in metadata. In those cases we corrected only very strong shifts, coincident detected in relation to more different reference series.
•    Some of the stations were tested in different groups. The station was accounted as homogenous, if was homogenise at least in one group.
•    Breaks: Breaks were verified on the base of original daily data and included daily data from surrounding stations.
•    Auxiliary graphs. Graphs of cumulative sums of relative differences of monthly precipitation totals from average 1961-1998 was made according to followed formulae:

( (a_ij-a_j)/a_j - (b_ij-b_j)/b_j)), where i = 1,...,38, j = 1, .., 12,

a_ij, b_ij is precipitation total in i-th year and j-th month, a_j, b_j is average monthly totals for the period 1961-1998 on the tested station and the regional station REF respectively. A letter a is related to the tested station, a letter b to the station REF of given station group.

The graphs are very good and precisely show differences in behaviour of tested series in relation to given regional series (Fig. 2). Most of the breaks detected by MASH are highly visible. The risen parts of the curve in the graph correspond to time intervals, when differences between monthly totals of tested station and of REF station were higher then average for all period, and contrary.
The result of homogenisation of the data from the station Usti nad Orlici (H2USTI01) is clearly shown on pictures X and Y. Picture X (data before homogenisation) is a graph made out of two visibly different parts. Before relocating the station, rainfall was higher, and the graph curve goes up. After relocating, the rainfall decreased, and the graph curve goes down. Picture Y shows the state after homogenisation.
A change of critical value follows sometimes to appearing a new breaks and shifts.
Graphs are very useful for decision in cases when the tested series shows different breaks to different reference series.

Executed homogenisation was very useful for better understanding of precipitation series in the Czech Republic. The process of homogenization led in more cases to repeat verifying of daily data. We have also discovered more insufficiencies in the metadata records.We have made many efforts to make metadata better. However, we are somewhat sceptical to possibilities of founding further metadata of studied stations. Homogenization of some stations is problematic and is not definite.
In opposite to temperature we suppose, that is not useful to make only one common reference series for a bigger region, for example the whole of the Czech Republic. It is possible, but only if we are studying the climate changes in a correspondingly large region. This distinguishing ability should be recorded as a part of metadata taking of the homogenisation process.

[1] Szentimrey, T.: "Statistical procedure for joint homogenization of climatic time series", Proceedings of the First seminar of homogenization of surface climatological data, Budapest, Hungary, 6-12 October 1996, p. 47-62.