Recent Climate Change at the Upper Danube—A temporal and spatial analysis of temperature and precipitation time series

Abstract

The following study investigates temperature and precipitation trends in instrumental time series between 1960 and 2006 at 88 meteorological stations located in the Upper Danube Basin. Time series were tested for inhomogeneities with several common homogeneity tests, trend magnitudes of annual and seasonal time series were calculated by least square fitting and the significance of trend values was checked and quantified by the Mann-Kendall test. The results confirm a particularly strong recent Climate Change in the investigation area. Increasing temperature trends show remarkably high trend values up to 0.8°C/decade in the summer season. The trends are highly significant for all investigated summer, spring and annual time series. Winter and spring temperature trends show consistently positive trend values as well even though some time series show no significance at all and the calculated trend values are smaller. Autumn temperature trends are mostly non-significant with low values (up to 0.3°C/decade) and several negative trends. Most of the highest trend values can be found in lower altitudes whereas stations situated in alpine regions tend to show low trend magnitudes and often exhibit non-significant results. Precipitation time series show positive as well as negative trends in the annual and seasonal analysis. At most stations a precipitation decrease in summer and autumn and an increase in winter was observed during the last 47 years whereas the spring and mean annual precipitation exhibits no change at all. But most time series are not conclusive since they show predominantly no significance and they often exhibit only low trend values.

1 Introduction

Global Change with its noticeable consequences like the increase of air temperatures or changes in the hydrological cycle (e.g. the increase of floods, droughts or extreme weather events) has been controversially discussed in the past among experts as well as in the public. Meanwhile the existence of a climatic change with a more or less high human influence factor is commonly accepted and proved by various studies, e.g. the 4th Assessment Report of the Intergovernmental Panel on Climate Change (IPCC). The IPCC concludes in its current study that global mean temperatures increased by 0.74°C ± 0.18°C in the last 100 years (1906–2005) whereas universally valid conclusions about global precipitation trends are difficult as precipitation shows a high spatial and temporal variability (Solomon et al. 2007). Rough estimates show a significant increase of precipitation amounts in the east of North and South America, northern Europe and northern and central Asia and a significant decrease of precipitation amounts in the Sahel, the Mediterranean, southern Africa and parts of southern Asia (Solomon et al. 2007).

Thus it appears that Global Climate Change is not uniform around the globe. In fact, some regions are more affected by Climate Change than others. For example, the highest temperature rise is observed over land regions in northern latitudes and it seems that especially alpine regions reflect an amplification of the global temperature trends (Solomon et al. 2007; Beniston et al. 1994). For that reason, it is important to know not only the spatial pattern of global climate trends but also to examine meteorological parameters on a regional scale. With precise investigations of climatic parameters measured at discrete weather stations it is possible to gain an impression of how strong the effects of Climate Change already are at a regional level. Such investigations in the form of trend analyses of meteorological station data have already been carried out all over the world in recent years with remarkable results. These investigations generally concentrate on a single country or just a part of a country. Studies to detect temperature or precipitation trends have been carried out for example in Saxony/Germany (Franke et al. 2004), Baden-Württemberg/Germany (Rapp and Schoenwiese 1995b), Austria (Matulla et al. 2003), Switzerland (Begert et al. 2005; Beniston et al. 1994), Italy (Buffoni et al. 1999; Buffoni et al. 1999), the European Alps (Böhm et al. 2001), Hungary (Domonkos and Tar 2003), the Czech Republic (Brázdil et al. 2009), Poland (Degirmendžić et al. 2004), Athens (Founda et al. 2004), Greece (Feidas et al. 2007), the Aegean (Good et al. 2008), the United States (Lund et al. 2001), Zimbabwe (Unganai 1995), China (Zhang et al. 2008), Japan and Tokio (Schaefer and Domroes 2009; Yue and Hashino 2003) or Ethiopia (Cheung et al. 2008).

Due to the importance of the parameters air temperature and precipitation—and their change—as components of the hydrological cycle, the following study concentrates not on a single country but on a hydrological catchment area, the Upper Danube basin. The catchment area shows an extreme sensitivity for Climate Change and associated possible future water based problems on the basis of high physiographic gradients and strong meteorological gradients as well as the complex use of water resources in the region. Therefore the watershed serves as test and validation area for the interdisciplinary research project GLOWA-Danube (Global Change of the Water Cycle), which focuses on the hydrological cycle at the Upper Danube under Climate Change conditions. Within the scope of the project (2001–2010) the decision support system DANUBIA was set up to investigate water relevant questions which are of interest for stakeholder from politics, economy and administration, i.e. the impact of less run-off. With the present work we intend to identify noticeable changes of the most important climatic factors temperature and precipitation in the investigation area by analysing seasonal and annual time series in order to reveal the importance of the regional decision support system for the development of adaptation strategies in water management.

2 Investigation area

The catchment area of the Upper Danube, one of the most important alpine watersheds, is part of the Danube watershed which is the second largest river basin in Europe with an area of more than 800.000 km². The Upper Danube watershed covers an area of 76.653 km². It is terminated by the discharge gauge “Achleiten” near Passau (Bavaria/Germany). The major part lies in Germany (73% of the area in the federal states Bavaria and Baden-Württemberg) followed by Austria (24%), whereas the remaining 3% of the area are split up between Switzerland, Italy and the Czech Republic (Ludwig et al. 2003) (Fig. 1).

Fig. 1

Investigation area upper Danube

The Upper Danube is characterised by an alpine topography which stretches from 287 m a.s.l. at gauge Achleiten up to 4047 m a.s.l. at Piz Bernina (Switzerland). The annual precipitation rates range from 550 mm in low altitudes to more than 2000 mm in mountain regions and the long-time annual air temperature lies between −5.3°C at alpine stations above 2000 m a.s.l. and 9.5°C in sheltered locations (e.g. valleys). The watershed includes agricultural areas as well as glaciers in alpine regions and is characterised by an intensive use of the available water resources e.g. for hydropower, farming or the production of artificial snow in the winter tourism industry (Mauser and Bach 2009).

3 Material and methods

Both mean temperatures as well as precipitation totals for the following investigation were taken from the meteorological database of the GLOWA-Danube-Project. This database contains various meteorological parameters in varying temporal resolutions measured at German and Austrian meteorological stations and provided by the German Weather Service (Deutscher Wetterdienst, DWD) as well as by the Austrian Central Institute for Meteorology and Geodynamics (Zentralanstalt für Meteorologie und Geodynamik, ZAMG). The database contains meteorological time series from 377 stations which are located in Austria and in the two German federal states Bavaria and Baden-Württemberg in altitudes between 300 and 3105 m a.s.l.. For the analysis only stations located in the Upper Danube watershed and in the border region of the catchment area were used.

Daily means of air temperatures were calculated using three daily measurements. The measurements are taken at 7:30, 14:30 and 21:30 Central European Time (CET) in Germany and at 7:00, 14:00 and 21:00 local mean time in Austria, respectively. As from 1 April 2001 the German Weather Service began to change measurement times at more and more weather stations to 7:00, 14:00 and 19:00 CET. The discrepancy between measurement times in the two countries and the change of the observation time in Germany may have a noticeable effect when comparing daily means of the air temperature for two adjacent stations (Gattermayr 2001). But since in this study annual and seasonal averages were used, the varying measurement hours were not taken into account for the analysis. The daily (24 h) precipitation totals were calculated from 7:30 to 7:30 CET of the following day in Germany and—similar to the change of temperature measurement times - from April 1, 2001 at more and more stations from 7:00 to 7:00 CET. The Austrian daily precipitation totals are continuously estimated from 7:00 to 7:00 local mean time. The minimal postponement of the measurement times of precipitation carries no weight in 3-month- and 12-month-sums and therefore has been neglected as well.

The German temperature and precipitation records available within the GLOWA-Danube project start in 1949, whereas the Austrian data does not start before 1960. Both data sets end in 2006. As a result of the shorter Austrian series of measurements only the period from 1960 to 2006, which is also used as the reference period in GLOWA-Danube, was used for the following trend analysis. For every available time series seasonal and annual temperature means and precipitation sums were calculated from the daily values. The winter season includes the months November, December and January and the summer season the months June, July and August. Consequently, the autumn season is represented by September, October and November, whereas the spring months are February, March and April.

The data has been quality checked by DWD and ZAMG and additionally it has been tested for outliers and errors before setting up the project database. Additionally the data set was tested for inhomogeneities before the start of the trend analysis. It is suspected that inhomogeneities occur for example by relocation of observation stations, changes in the observation instrumentation and observing practices. In most cases a non-homogeneous time series is not detected by visual methods. In fact only homogeneity tests are able to retrieve artificial abrupt or gradual changes in the data. The homogeneity tests were applied to the annual data set since effects of inhomogeneity emerge more clearly in annual data than in monthly or daily data (Rapp 2000; Rapp and Schönwiese 1995a).

The methods for testing the homogeneity of meteorological time series are classified into relative and absolute homogeneity tests. Relative homogeneity tests always compare the investigated data to a reference time series which consists of a more or less homogeneous mean of neighbouring time series or of a single time series which has been identified as homogeneous before. Absolute homogeneity tests scan single time series for abrupt or gradual changes on the basis of certain criteria, e.g. the inhomogeneity of the expectancy value by means of the variance (KLIWA 2005b). Temperature time series exhibit very similar fluctuation structures which can be confirmed by high values of correlation coefficients between individual meteorological stations even at high distances whereas the fluctuation structures of precipitation time series, even when situated close together, are very different in most cases as a result of differing atmospheric conditions (e.g. shower weather with localized rainfall) or regional and local factors (e.g. windward or leeward exposure) (Brázdil et al. 2009). On this account relative homogeneity tests were applied to temperature series whereas absolute homogeneity tests were applied to precipitation series. The significance level was set to 5% in all following homogeneity testing methods.

For testing the homogeneity of the temperature series the relative Craddock homogeneity test and the relative standard normal homogeneity test of Alexandersson (SNHT) were applied to the annual data (Craddock 1979; Alexandersson 1986 and 1995; Štěpánek 2005). As references, mean series which were calculated from correlating neighbouring stations (r > 0.9) as well as homogeneous correlating single series were used. Further the annual and seasonal data was checked for inhomogeneities by comparing neighbouring stations with each other (Weber et al. 1997). From 84 temperature series in the investigation area which are suitable for the trend analysis, at least 14 series have shown inhomogeneous test results. Only for four stations the inhomogeneity could be traced back to a relocation of the stations. Due to the unavailable station history at the 10 remaining stations, the reason for their inhomogeneity remains ambiguous.

The homogeneity of the precipitation series was tested by calculating several common absolute homogeneity tests: the cumulative deviations, the Bayesian Procedures, the Mann-Whitney-Pettit test and the standard normal homogeneity test for single series (Buishand 1982; Alexandersson 1986 and 1995; KLIWA 2005b; Štěpánek 2005). The time series were classified into “homogeneous classes” and only the time series which were likely to be homogeneous (at least two homogeneous results with most weight on the SNHT-test) were used for the trend analysis (Rapp and Schönwiese 1995a). Nearly all precipitation series (76 out of 84) turned out to be homogeneous on a significance level of 5%.

In case of inhomogeneous test results there exists a prevalent approach of adjusting the time series so that a reapplied homogeneity test reveals a homogeneous result. Both temperature and precipitation series were not adjusted when inhomogeneities occurred but excluded from the following trend analysis. The reason for this approach lies in the inadequate and often missing station history which therefore can not be adducted for the determination of the inhomogeneity’s magnitude and points in time when one or more inhomogeneities appeared (Böhm et al. 2001; Rapp and Schönwiese 1995a). Moreover Rapp and Schönwiese (1995a) allude to the fact that an adjustment of the data with most methods for homogenization simply means a data transmission from homogeneous time series to inhomogeneous time series and therefore provides no new information.

Consequently there exist 70 stations with reliable temperature records and 76 stations with reliable precipitation records for the selected period from 1960–2006. For the summer temperature trends only 69 stations could be used to calculate the trend values due to missing data. Attention should be paid to the fact, that despite the related number of temperature and precipitation time series, the location of the investigated parameters is not automatically the same (Fig. 2 and Table 16).

Fig. 2

Spatial distribution of meteorological stations with homogeneous time series of precipitation and temperature used for the trend analysis

The magnitude of the seasonal and annual trends was calculated by linear regression (least square method) with the regression slope in the form

$$ {y_n} = a{x_n} + b $$

(1)

y _n :

ordinate values of the regression slope at the times of x _n

a, b :

regression coefficients

The temperature trends were declared as absolute trends Tr _abs with

$$ T{r_{{abs}}} = \left( {{y_n} - {y_1}} \right) $$

(2)

y _n :

last value at time t _n

y ₁ :

first value at time t₁

The precipitation trends were declared as relative linear trends Tr _rel with

$$ T{r_{{rel}}} = \frac{{T{r_{{abs}}}}}{{\bar{y}}} \cdot 100\% $$

(3)

$\bar{y} = {\text{ mean}}$ of the investigated time series

With the relative method of trend calculation the change is demonstrated in % of the mean value compared to the beginning of the investigation period (Rapp and Schönwiese 1995a). The advantage of using relative precipitation trends in maps and analyses is the independence of the trend values from the extent of the precipitation sums, which are strongly influenced by altitude. Accordingly the topography is not embedded in the results and trend values from stations in varying altitudes become comparable.

The significance of the linear trend values was assessed by the rank-based non-parametric Mann-Kendall-Test (Mann 1945; Kendall 1975; Salmi et al. 2002). This test is robust to influences of extremes and is used for detecting trends in hydro-meteorological time series, in which data is occasionally missing or data is not normally distributed (Yue and Hashino 2003; Partal and Kahya 2006). According to the Mann-Kendall test, the null hypothesis H ₀ states, that the time series show no trend at all and the observed values x _i are randomly ordered in time. However, the H ₁ hypothesis states the existence of an increasing or decreasing monotonic trend in the time series.

The Mann-Kendall test statistic S and the variance of S Var(S) were calculated using

$$ S = \sum\limits_{{k = 1}}^{{n - 1}} {\sum\limits_{{j = k + 1}}^n {{\rm sgn} \left( {{x_j} - {x_k}} \right)} } $$

(4)

$$ Var(S) = \frac{1}{{18}}\left[ {n\left( {n - 1} \right)\left( {2n + 5} \right) - \sum\limits_p {{t_p}\left( {{t_p} - 1} \right)\left( {2{t_p} + 5} \right)} } \right] $$

(5)

t _p :

number of identical data values in the observed values x _i

n :

number of annual values in the analysed data series and

x _j , x _k :

annual values in years j and k; j > k

The values of S and Var(S) were used to compute the test statistic Z which is a measure of the degree of the trend’s significance. The higher the absolute value of Z, the higher is the significance of the existing trend. Negative values of Z show negative trends and positive values show positive trends. The standard normal variable Z was applied to test the significance by means of the probability table of the normal distribution (Mann 1945; Kendall 1975)

$$ Z = \left\{ {\begin{array}{*{20}{c}} {\frac{{S - 1}}{{\sqrt {{Var(S)}} }}}{if\,S > 0} \\ 0{f\,S = 0} \\ {\frac{{S + 1}}{{\sqrt {{Var(S)}} }}}{f\,S < 0} \\ \end{array} } \right\} $$

(6)

In many studies the significance level was set to 95% with Z values > |1.96| (Yue and Hashino 2003; Schaefer and Domroes 2009) whereas for instance in KLIWA (2005a) or Franke et al. (2004) the level was set to 80% with Z values > |1.282|. The following trend analysis used the significance level of 80%. Mann-Kendall test results which were below the 80% significance level were labelled as “non-significant” but nevertheless were involved in the trend analysis and presented in the maps to support the common tendencies in the long-term time series.

4 Results

4.1 Summer temperature

Figure 3 shows the spatial distribution of the linear trend magnitudes calculated by least square fitting for the summer season. One single time series was not used for the summer trend analysis due to missing data (station 43). The values indicate a strong positive temperature trend at all 69 investigated stations in the period 1960–2006 with a mean temperature rise of 2.42°C. The Mann-Kendall test revealed a statistical significance on a very high level on most of the investigated time series (Table 1).

Fig. 3

Trend of mean summer temperatures [°C] at single meteorological stations during the investigation period 1960–2006

Table 1 Significance levels from the Mann-Kendall test and number of stations in different levels for the summer temperatures

In the investigated period, summer temperatures increased between 1.13°C at station 27 in the north-west up to 3.99°C at station 74 in the east of the investigation area (Table 2 and Fig. 3)

Table 2 Classified trend values of summer temperature trends

The spatial distribution pattern of summer temperature trends shows, that most of the extremely high temperature trends (>3°C) are situated in the eastern part of the investigation area. Sporadic high trend values of more than 3.0°C are observed in the remaining sub-regions. It is conspicuous that the highest trends (>3.0°C) occur in low altitudes below 500 m a.s.l. at six out of eight stations. Compared to the stations at lower altitudes (<1000 m a.s.l.) the alpine stations seem to exhibit smaller trend values. Only at three out of 14 alpine stations the trend is higher than 2.5°C whereas at the lower stations 24 out of 55 reach trend values of 2.5°C and even higher (Fig. 4).

Fig. 4

Trend values of summer temperature series, ordered by altitude (the black straight line illustrates the linear trend)

4.2 Autumn temperature

The trend magnitudes calculated by least square fitting for the autumn season reveal the following results: The mean autumn temperature rose by +0.66°C but at least 6 stations revealed minor negative non-significant trend values between −0.1 and −0.4°C. Most of the trend values revealed non-significant results after the Mann-Kendall test (Table 3).

Table 3 Significance levels from the Mann-Kendall test and number of stations in different levels for the autumn temperatures

In the investigated period, the positive autumn temperatures increased between 0.1°C and 1.39°C (Table 4). These values are quite small compared to the summer temperature trends but this finding corresponds with several comparable previous studies in Central Europe on seasonal temperature trends which revealed moderately positive or even negative temperature trends in the autumn season in varying time periods (Franke et al. 2004; Brázdil et al. 2009; KLIWA 2005a).

Table 4 Classified trend values of autumn temperature trends

The spatial analysis of the autumn temperature trends did not reveal any conspicuousness. Significant and non-significant trends are distributed all over the investigation area and also the different magnitudes of the trends are evenly distributed (Fig. 5). An analysis of the autumn temperature trends in different sea levels indicates a slight decrease of trend values with higher altitudes. However, this effect is nearly nullified when regarding only the significant trend values (Fig. 6).

Fig. 5

Trend of mean autumn temperatures [°C] at single meteorological stations during the investigation period 1960–2006 (non-significant trends in winter temperature trends are marked with a cross)

Fig. 6

Trend values of autumn temperature series, ordered by altitude (significant trends in winter temperature are illustrated with filled symbols, non-significant time series are displayed in white, the black straight line illustrates the linear trend)

4.3 Winter temperature

Even though temperature trends in the winter season are consistently positive as well, the linear trend values calculated by least square fitting for the winter season are markedly smaller than the summer trend values with a mean trend value of 1.68°C and the statistical significance is not as distinctive as for the summer trends (Fig. 7 and Table 5). At least 17 out of 70 investigated time series show no significance in trend at all although their computed trend values are positive between 0.11°C and 1.82°C. The remaining 53 significant stations show a positive temperature trend between 0.93°C and 2.44°C (Table 6). At least 17 of the significant stations show a relatively low significance level of only 80% whereas the main part of the time series achieves higher significance levels.

Fig. 7

Trend of mean winter temperatures [°C] at single meteorological stations during the investigation period 1960–2006 (non-significant trends in winter temperature trends are marked with a cross)

Table 5 Significance levels from the Mann-Kendall test and number of stations in different levels for the winter temperatures

Table 6 Classified trend values of winter temperature trends

It is remarkable that 7 out of 17 non-significant trends, which generally tend to show lower values compared to the significant trends, can be found in the alpine region and 5 of them above 1000 m. In the alpine region the trend of rising winter temperatures is not as pronounced as in the remaining investigation area. The highest trend values lie between 1.5 and 2.5°C for the winter temperatures and can be found especially in the northern part of the area in latitudes between 313 and 655 m a.s.l.. Similar to the summer temperature trends there is a slight tendency for lower increasing winter temperature trends in higher latitudes than in low latitudes (Fig. 8).

Fig. 8

Trend values of winter temperature series, ordered by altitude (significant trends in winter temperature are illustrated with filled symbols, non-significant time series are displayed in white, the black straight line illustrates the linear trend)

4.4 Spring temperature

The analysis of spring season from February to April results in a mean temperature rise of 1.88°C, which is very similar to the winter season. But the significance levels after the Mann-Kendall test show a different picture as most of the investigated time series show significant results at a very high level (Table 7).

Table 7 Significance levels from the Mann-Kendall test and number of stations in different levels for the spring temperatures

The classification of the computed spring trend values is shown in Table 8. The distribution of the number of stations in the different classes is very similar to the results of the winter trends (Table 6).

Table 8 Classified trend values of spring temperature trends

Higher trend values (>2.0°C) are found mainly in the southern part of the Upper Danube catchment (Fig. 9). Concerning the altitudinal range, most of the high trend values are located in the range of 300–1000 m a.s.l. and similar to the findings in the previous chapters on summer, autumn and winter temperature, the spring temperature trends tend to be higher in low altitudes and to be lower in high altitudes (Fig. 10)

Fig. 9

Trend of mean spring temperatures [°C] at single meteorological stations during the investigation period 1960–2006 (non-significant trends in winter temperature trends are marked with a cross)

Fig. 10

Trend values of spring temperature series, ordered by altitude (significant trends in winter temperature are illustrated with filled symbols, non-significant time series are displayed in white, the black straight line illustrates the linear trend)

4.5 Annual temperature

The annual mean temperature shows a strong positive trend during the investigation period. The computed trends show positive values at all stations with a temperature increase between 0.62°C in the north-west of the catchment area and 2.66°C in the very east (Fig. 11). The significance of the annual trend values is remarkably high and exceeds the 99.9% level at most of the stations (Table 9).

Fig. 11

Trend of mean annual temperatures [°C] at single meteorological stations during the investigation period 1960–2006

Table 9 Significance levels from the Mann-Kendall test and number of stations in different levels for the annual temperatures

The calculated absolute temperature trends exhibit a mean value of 1.65°C between 1960 and 2006. This value approximates the mean winter temperature trend. The fact, that in spite of the high summer trend values, the mean annual temperature trend is close to the winter value can be explained by the very low and even decreasing spring trend values and the values of the autumn trends which are close to the winter trends. More than half of the time series show an increase of annual temperatures between 1.5°C and 2.0°C (Table 10).

Table 10 Classified trend values of annual temperature trends

Similarly to the seasonal temperature trend values the highest annual temperature trends tend to occur in lower altitudes (Fig. 12). Considering the time series which show trend values of more than 2.0°C, most of them are located below 450 m a.s.l.. Furthermore, the time series with the highest values are situated predominantly in the eastern part of the catchment area. Most stations with lower temperature trends can be found in the region of the main Alpine divide and in the pre-Alps.

Fig. 12

Trend values of annual temperature series, ordered by altitude (the black straight line illustrates the linear trend

Figure 13 illustrates the temporal development of the mean annual temperatures of various selected time series in combination with the mean of the annual temperatures from 1961 to 1990 as a reference value for different sea levels (which represent different temperature levels) to show the total range of mean temperatures in the Upper Danube catchment. The course of the single graphs is very similar but strongly depends on the altitude of the climatic stations. In general, a higher location of a meteorological station implies lower mean annual temperatures.

Fig. 13

Temporal development of selected mean annual temperatures from 1960–2006 compared to the mean temperatures on this stations from 1961 to 1990

The single graphs can reflect observed mean global temperature developments to a certain extent. For example, one of the warmest years in the investigated time period on a global scale was the year 1994 and a sequence of warm years could be observed since 2000 (Lean & Rind 2009). At least 30 out of the 70 stations in the study show their warmest annual temperature in this year and all selected temperature graphs show a peak in this year. Furthermore, an obvious sequence of warm years could be observed since the year of 2000 in the investigated time series. Figure 13 illustrates, that the mean temperature of the selected stations from 1961 to 1990 is exceeded in nearly all years and time series since 2000. More than half of the time series (38 out of 70) showed their warmest year in 2000 or later.

However, this comparison cannot be conducted for all extreme years. One of the coldest years in the time period 1960–2006 on a global scale has been 1985 and we can find indications that the year 1985 is a relatively cold year in the Upper Danube catchment as well. Indeed one can find many years in the investigated time period, which are much colder than the year 1985 and therefore it cannot be concluded from the investigated time series to a global behaviour or vice versa.

4.6 Summer precipitation

To emphasise considerable changes, the intervals for the relative precipitation trend were determined to 10% for all further explanations. Nevertheless all the time series with values between −10% and +10% are illustrated in the maps and figures to show the general tendencies of the precipitation.

The summer precipitation trend values in the investigation area during the period 1960–2006 show positive as well as negative precipitation trends (Fig. 14) but not all of the time series were classified as significant (Table 11)

Fig. 14

Relative trend of summer precipitation sums [%] at single meteorological stations during the investigation period 1960–2006 (significant trends in summer precipitation are illustrated with black symbols, non-significant trends in summer precipitation are illustrated with white symbols)

Table 11 Significance levels from the Mann-Kendall test and number of stations in different levels for the summer precipitation

At least 44 out of the 76 time series show values between −10% and +10%. Time series with positive trends are located predominantly in the Alps and the significant positive trend values are located exclusively in the eastern part of the Alps. The highest positive trend value was computed at station 64 (Kitzbühel in Austria) with a summer precipitation increase of 17%. The remaining investigation area is mainly characterised by negative summer precipitation trends. 18 out of 25 noticeable negative trends reveal significance levels of 80% or higher with a mean relative precipitation trend of −12.1%. Most of the significant time series with very high trend values of a precipitation decrease are located in the western part of the investigation area. In this region the summer precipitation reached a decline up to −31% between 1960 and 2006.

Considering the computed summer precipitation trends combined with their altitudinal location it is remarkable that both positive as well as negative trends can be found in altitudes up to 1800 m a.s.l. whereas in altitudes of more than 1800 m a.s.l. only positive trends occur. Negative trend values can be found mainly in altitudes between 300 and 980 m a.s.l., only two time series with negative summer precipitation trends appear above this range. However, time series with positive trend values can be found at all altitudes of the meteorological stations from 300 to 3105 m a.s.l.. Thus it appears that there exists a general tendency of increasing summer precipitation in alpine altitudes, whereas lower altitudes of less than 1000 m a.s.l. reveal a decreasing tendency in the summer precipitation characteristic (Fig. 15).

Fig. 15

Trend values of summer precipitation series, ordered by altitude (significant trends in summer precipitation are illustrated with filled symbols, non-significant time series are displayed in white, the black straight line illustrates the linear trend)

4.7 Autumn precipitation

The autumn precipitation trend values in the investigation area during the period 1960–2006 show mostly positive precipitation trends, but most of the time series were classified as non-significant (Table 12). At least 55 out of the total of 76 time series show values of more than +10% but only 20 of them turned out to be significant. The mean precipitation increase in the autumn years of 1960–2006 is 17.7%.

Table 12 Significance levels from the Mann-Kendall test and number of stations in different levels for the autumn precipitation

The stronger and significant trends of the autumn precipitation (>20%) can be found in the peripheral areas of the investigation area, mainly in the south-east (Fig. 16). Considering the computed autumn precipitation trends combined with their altitudinal location it is remarkable that the few negative precipitation trends are located in regions below 1500 m a.s.l. whereas the positive trends occur all over the altitudinal range (Fig. 17). As this distribution is quite similar to the summer trends, it can be confirmed that negative trend values can be found mainly in low altitudes whereas increasing autumn precipitation values can be found in all altitudinal levels.

Fig. 16

Relative trend of autumn precipitation sums [%] at single meteorological stations during the investigation period 1960–2006 (significant trends in summer precipitation are illustrated with black symbols, non-significant trends in summer precipitation are illustrated with white symbols)

Fig. 17

Trend values of autumn precipitation series, ordered by altitude (significant trends in summer precipitation are illustrated with filled symbols, non-significant time series are displayed in white, the black straight line illustrates the linear trend)

4.8 Winter precipitation

The winter precipitation trend is negative in 29 time series whereas 47 time series show a positive trend (Fig. 18 and Table 13). Compared to the summer precipitation trends, the trend values of the winter precipitation generally show higher values in the positive as well as in the negative range. Indeed many of the investigated station data show very low changes between −10% and +10%. Only 17 time series show a significance level of 80% or more and the mean relative trend of these time series reaches almost +10%.

Fig. 18

Relative trend of winter precipitation sums [%] at single meteorological stations during the investigation period 1960–2006 (significant trends in winter precipitation are illustrated with black symbols, non-significant trends in summer precipitation are illustrated with white symbols)

Table 13 Significance levels from the Mann-Kendall test and number of stations in different levels for the winter precipitation

Five of 29 negative trend values reveal significant changes between −14% and −52% in the investigation period. These stations are situated in the Alps or Pre-Alps (Garmisch, Innsbruck, Saalbach, Mooserboden) and accordingly at the Hohenpeißenberg, which is the highest altitude in the Swabian-Bavarian plateau with an elevation of 989 m a.s.l.. The non-significant negative winter precipitation trends are situated predominantly in the Alpine region as well.

12 out of 47 positive trends show a significance level of 80% or more with precipitation increases of up to 47% at station 50 in the very east of the investigation area. Significant as well as non-significant positive trends occur mainly in the northern part of the investigation area whereas only two noticeably positive trends are situated in the Alpine region.

On the one hand the altitudinal distribution of the winter precipitation trends reveals an accumulation of positive trends with very high positive values in altitudes up to 1000 m a.s.l. and only very few (5 out of 47) positive trends in altitudes which are located above 1000 m a.s.l.. On the other hand the negative precipitation trends are distributed at altitudes between 356 and 2036 m a.s.l. with most stations located below 1000 m a.s.l.. Since 8 out of 29 decreasing trend values are situated above 1000 m a.s.l. the average linear trend of winter precipitation becomes negative (Fig. 19).

Fig. 19

Trend values of winter precipitation series, ordered by altitude (significant trends in winter precipitation are illustrated with filled symbols, non-significant time series are displayed in white, the black straight line illustrates the linear trend)

4.9 Spring precipitation

The spring precipitation trends show very low significance. Only four out of 64 time series, which were analysed, show any significance at all (Fig. 20, Table 14). 49 time series reveal values between −10% and +10%, so they are not meaningful to consider them in the interpretation of the results. 14 of the meaningful trends show positive values, 13 of them are negative. The mean spring precipitation trend is −0.4% due to the fact that the positive and negative trend values are balanced out.

Fig. 20

Relative trend of spring precipitation sums [%] at single meteorological stations during the investigation period 1960–2006 (significant trends in winter precipitation are illustrated with black symbols, non-significant trends in summer precipitation are illustrated with white symbols)

Table 14 Significance levels from the Mann-Kendall test and number of stations in different levels for the spring precipitation

Overall, the relation between all trends (positive and negative) is very balanced. This shows only marginal changes in temperature during the last 47 years. Due to the small number of only four significant spring precipitation trends and therefore the too little force of expression the authors waive the analysis of the trend behavior in different altitudes.

4.10 Annual precipitation

The trend of the annual precipitation based on the daily 24-hour rainfall totals of the 76 meteorological stations shows negative values in 31 time series and positive values in 45 time series with a mean relative precipitation change of +2.6% (Fig. 21). Indeed more than 74% of the time series (56 out of 76) show no annual precipitation trend at all with inconsiderable trend magnitudes between −10% and +10% (Table 15). Four of the remaining 20 time series exhibit negative trend values which are larger than −10%, but only two of them are significant according to the Mann-Kendall test. Highest significance levels can be found in the 16 time series which show an annual precipitation increase of more than 10%. 13 out of 16 time series with positive trend values of more than +10% are significant with trend values up to +21% (Fig. 21).

Fig. 21

Relative trend of annual precipitation sums [%] at single meteorological stations during the investigation period 1960–2006 (significant trends in annual precipitation are illustrated with black symbols, non-significant trends in summer precipitation are illustrated with white symbols)

Table 15 Significance levels from the Mann-Kendall test and number of stations in different levels for the annual precipitation

Negative as well as positive precipitation trends are spread over all altitude levels in the investigation area so that we were not able to identify any clear tendencies with altitude (Fig. 22). On the one hand, negative precipitation trends can be found especially in the south and south-east of the Upper Danube basin with two out of three significant trends in the alpine region. On the other hand, the positive trends are distributed all over the investigation area except in the western region.

Fig. 22

Trend values of annual precipitation series, ordered by altitude (significant trends in annual precipitation are illustrated with filled symbols, non-significant time series are displayed in white, the black line illustrates the linear trend)

4.11 Correlation between temperature and precipitation trends

A statistical correlation analysis was conducted with all pairs of precipitation/temperature trends for the four investigated seasons and the annual trends. The results show only very low correlation coefficients between 0.08 and 0.31 and they show no statistical significance. Nevertheless, a visual comparison of the temperature and precipitation trends revealed a possible connection between the trends of temperature and of precipitation even though temperature trends show spatially uniform behaviour (generally higher temperatures in seasonal as well as annual time series), whereas precipitation trends show positive as well as negative values in the seasonal and annual time series. Considering the positive precipitation trends for summer and winter trends in Fig. 23a–e it appears that larger positive temperature trends imply larger positive precipitation trends. This statement cannot be made for annual, spring and autumn trends. The negative winter precipitation trends as well as the negative annual precipitation trends tend to become more moderate with larger temperature trends. However, the negative summer, spring and autumn precipitation trends show no relation at all to larger values in the temperature trends (Table 16).

Fig. 23

Relation between precipitation trends (significant and non-significant) and temperature trends of summer (a), autumn (b), winter (c), spring (d) and annual data sets (e)

Table 16 Location and elevation (m a.s.l.) of meteorological stations (T: used for temperature analysis; P: used for precipitation analysis)

5 Summary and discussion

The annual as well as the seasonal trends for temperature and precipitation of meteorological stations in the Upper Danube region observed during the period 1960 and 2006 enhance the affection of the region by Climate Change. Particularly the temperature trends in summer reflect the change because of their considerably high significance and their markedly high trend values. The results indicate that Climate Change apparently show much stronger effects on the Upper Danube region than it is predicted by the global view published in the latest IPCC report (Solomon et al. 2007).

The results of the temperature trend analysis show smaller temperature increases in the alpine region compared to the whole investigated area. These results disagree with the common findings that a Climate Change induced temperature increase is generally higher in alpine regions than in lowlands (Beniston et al. 1994). One reason for this might be a reduced effect of Climate Change impacts in regions where the climate is already characterized by strong variations as it is proved for alpine regions. The investigated temperature trends in this study (especially the summer, winter, spring and annual trends) seem to be higher than in most other studies in Central European available so far. However, the chosen period 1960–2006 in the present study is 6 years longer than the time period in almost all comparable studies (KLIWA 2005a, Franke et al. 2004). Comparably high temperature trend values are documented for the Czech Republic in Brázdil et al. (2009), where the investigation period was 1961–2005. This supports the observation that temperature increase has taken place particularly in the first years of the 21^st century. In this context it is necessary to allude to the fact that computed trend values are always affected by the chosen period, so results of different studies can only be compared in the strict sense, if the investigation periods agree completely. It remains to be seen whether similarly large temperature trends are observed in other Central European regions once comparable studies of similar time series have been published.

The precipitation data generally shows lower significance values than the temperature data due to its higher spatial and temporal variability. This variability is reflected particularly by the appearance of positive as well as negative trends both in seasonal and annual time series. Because of the high variability of precipitation in alpine regions the trend is not unique. Therefore, we cannot make a universally valid statement for the spatial development of the precipitation in the investigation area. The annual precipitation trends are moderate, ranging between −20% and +21%. The mean annual precipitation trend of all investigated time series of more than +10% or rather less than −10% is only +2.6%. For this reason, we conclude that a general trend in the mean annual precipitation for the investigation area cannot be identified from the existing data.

Nevertheless, the seasonal precipitation shows significant changes. We observe a tendency of negative winter precipitation trends in the southern part of the investigation area and a slightly positive trend for winter precipitation in the northern investigation area. These tendencies are intensified towards the East of the Upper Danube basin. Mean winter precipitation shows a positive trend of almost +10% during the last 47 years. The autumn precipitation revealed a quite similar intensification of precipitation in the eastern part of the investigation area and the trend values are almost entirely positive with a mean precipitation trend of +17.8%. However, summer precipitation trends reveal a quite different picture. Due to a precipitation decrease in most time series the mean summer precipitation shows a negative trend of more than 12% over the last 47 years. The spring precipitation trends are mainly non-significant with a mean precipitation trend of only −0.4% and they do not reveal any spatial pattern of distribution.

However, the general tendency of increasing precipitation in the winter season and decreasing precipitation in the summer season does not apply for the alpine region. Here we observe predominantly no change of summer precipitation and a decrease of winter precipitation at several alpine stations. These results are comparable with the study of long-term temperature changes from KLIWA (2005b) in Bavaria and Baden-Württemberg.

The presented results demonstrate that, based on homogeneous station data from the past 47 years, regional Climate Change in the Upper Danube basin shows significantly larger temperature effects than the global average. In the Upper Danube region the measured temperature trends of the past 47 years even exceeded the temperature trends assumed in the IPCC report for this region until the end of the 21 century based on the SRES-A1B scenario (see Solomon et al. Fig. 11.5, p.875). The general tendency derived by the IPCC from the analysis of the results of 20 global climate models, that winter precipitation in the region will increase and summer precipitation will decrease already shows up as a weak signal in the measured station data. The analysis of precipitation data also shows that the regional precipitation change signals are generally still weak, show large uncertainties and do not allow a solid statistical interpretation that goes beyond a confirmation of IPCCs general seasonal tendency. The recorded changes in climate as well as their regional deviation from the global picture drawn by IPCC in turn have already influenced the hydrological cycle, the availability of water resources, and the temporal and spatial distribution of water. Because of the high statistical significance of the past temperature signal the most pronounced and statistically most robust regional influence of Climate Change on the hydrology of the Upper Danube basin is on the dynamics of snowfall and snow storage. Past Climate Change has already resulted in a decline of snow cover since the 70s of the last century, which strongly affected winter tourism and has widely resulted in the installation and use of equipment to produce artificial snow.

The future impacts for the hydrological cycle in the Upper Danube region in case of ongoing seasonal and annual trends are documented in detail within the GLOWA-Danube project (GLOWA-Danube Projekt 2010). The analyses of different climate scenarios give the following results for the period 2011 until 2060: due to increasing air temperature coupled with a strong increase of evapotranspiration the water availability will decrease but will not become sparse. In combination with decreasing summer precipitation, the river discharge in the Upper Danube catchment will be reduced in the future. The reduction will be least in the Alps and be strongest along the Danube River. The water delivery at gauge Achleiten as well as the low flow of the Danube River will be reduced. This might lead to limitations for shipping in summer as well as to changes in the reservoir management for the energy production from hydropower in the whole catchment. Increasing temperatures lead to a strong reduction in the height of the snow cover, to a reduction of snow cover duration and to a diminution of the snow cover at lower elevations, which affects the winter tourism. The glaciers will have disappeared almost completely until 2060. The results also show an increase in crop yields due to higher temperatures and increasing atmospheric CO₂ concentration except during dry years as a consequence of soil drying, especially along the Danube River. Due to less rainfall, higher temperatures and the higher evapotranspiration an increase in forest fire risk is expected in the spring and summer months.

Future studies on the regional impact of Climate Change in the Upper Danube basin as well as on the identification of adaptation strategies should take care that regional Climate Change scenarios, which are the basis for scenario analyses, draw their credibility from the fact that they are able to realistically produce the regional Climate Change signal of the past. This can be achieved through either improving Regional Climate Models especially in mountainous regions or through the use of sophisticated stochastic climate generators, which produce the meteorological drivers of the future climate scenarios through rearranging past records of existing climate stations.