Owing to average temperature increases of at least twice the global mean, climate change is expected to have strong impacts on local hydrology and climatology in the Alps. Nevertheless, trend analyses of hydro-climatic station data rarely reveal clear patterns concerning climate change signals except in temperature observations. However, trend research has thus far mostly been based on analysing trends of averaged data such as yearly, seasonal or monthly averages and has therefore often not been able to detect the finer temporal dynamics. For this reason, we derived 30-day moving average trends, providing a daily resolution of the timing and magnitude of trends within the seasons. Results are validated by including different time periods. We studied daily observations of mean temperature, liquid and solid precipitation, snow height and runoff in the relatively dry central Alpine region in Tyrol, Austria. Our results indicate that the vast majority of changes are observed throughout spring to early summer, most likely triggered by the strong temperature increase during this season. Temperature, streamflow and snow trends have clearly amplified during recent decades. The overall results are consistent over the entire investigation area and different time periods.

INTRODUCTION

The pan-European importance of the Alpine region as a freshwater reservoir has been widely studied (European Environment Agency 2009). Climate change modifies the nature of this reservoir (Leipprand & Gerten 2006; Parry et al. 2007). This is driven by strong temperature increases, which have been roughly twice as high in the Alps compared with the global average (Brunetti et al. 2009; Keiler et al. 2010). In particular, the strong positive feedback processes related to albedo, snow cover and heat budget (snow–albedo feedback) are held responsible for this development (Fyfe & Flato 1999; Déry & Brown 2007; Vavrus 2007). For this reason, mountain areas are expected to be more prone to the impact of climate change than lowland areas (Viviroli et al. 2011).

Many climate impact studies provide the results of modelling efforts to predict future hydrological and climatological conditions in specific regions. Far less research has been conducted on analysing observed hydro-climatic variables to detect incipient changes and estimate future developments. One reason for this imbalance could be the fact that studying trends in hydro-climatic variables is challenging since there is a great deal of noise, and if there are signals, they tend to be very small (except in temperature records). Therefore, many of the annual trends found in literature, especially runoff trends, are not significant (cf. Pekarova et al. 2006; cf. Schimon et al. 2011), as they are masked by high variability (Morin 2011). Another problem is the fact that trend analyses depend strongly on the time window studied and the method used for trend detection (Viviroli et al. 2011). Few trend studies based on observational data provide the results of multiple hydro-climatic variables in one region, making comparison and especially interpretation difficult as the variables are highly interdependent: for example, snow conditions depend on temperature and precipitation, runoff depends on snow and temperature, and so on. Reviewing the literature, Kundzewicz (2004) strongly emphasises the necessity of joint detection and attribution studies for understanding the link between runoff trends and other variables.

In Austria and especially Tyrol, very few studies exist at all analysing even just single hydro-climatic variables. Fliri (1975) has studied the Tyrolean climate extensively, but in his time period, with a limited period of available data, and not in the context of a changing climate. Virtually no research exists on analysing climate change signals in Tyrol based on observational data except as part of large-scale studies concerning all of Austria (e.g. Schimon et al. 2011) or the Greater Alpine Region (e.g. Bard et al. 2011). Moreover, so far mostly temperature trends have been covered (e.g. Nemec et al. 2013).

To study whether a region has grown wetter or drier (warmer or colder, etc.) in general, it is necessary to look at interannual changes by studying yearly averages. This is often done when comparing trends of different regions in a country or a mountain range like the Alps, and most of the existing trend studies focus rather on these spatially large-scale analyses (Nayak et al. 2010). However, local stakeholders and industry, such as hydropower, irrigation and snow production for winter sports, are far more interested in looking at seasonal changes in water availability. These seasonal alterations have been studied mainly with the following two approaches. First, via annual indicators such as centre of volume (e.g. Stewart et al. 2005; Hodgkins & Dudley 2006), which are potentially able to indicate whether snowmelt occurs earlier in the year. However, several recent studies such as Whitfield (2013) and Kim & Jain (2010) discourage the use of these measures, as they can easily be affected by multiple factors such as winter rains, the numbers of dominant flow seasons or whether a hydrological year or a calendar year is considered. Second, seasonal streamflow changes have been analysed by breaking down the yearly averaged flows to quarterly or monthly averages and then testing them for trends (e.g. Gagnon & Gough 2002; Stahl et al. 2010; Schimon et al. 2011; Rangwala & Miller 2012). However, it seems that the yearly, seasonal or even monthly scale can only give limited information on trends, as the results, especially for runoff, are often heterogeneous and ambiguous (Radziejewski & Kundzewicz 2004). However, with a further reduction of the time span over which the values used for analysing trends are averaged, another problem is revealed: the smaller the temporal scale studied, the higher the variability of the data set and the smaller the ability of statistical tests (e.g. the Mann–Kendall (MK) test) to detect trends (Hensel & Hirsch 1992; Morin 2011). In addition, the probability of finding trends just by chance increases.

In this context, we applied a method to study daily trends with the stability of monthly values, which is based on an approach proposed by Kim & Jain (2010). They derived daily resolution trends based on 3-day averages for several watersheds in the western USA and found significant changes of seasonal streamflow. With this approach, it is possible to easily derive subseasonal changing patterns for any variable. The results are consistent, and combined with composites of results of different time periods, apply to the time of anthropogenic climate change. The present study integrates a number of hydro-climatic variables to better understand the interconnections between these variables. We do not go into detail with each one of the variables, but rather give a holistic picture of climate change impacts on the hydroclimatology and especially the hydrology of a selected region in the dry central valley of the European Alps. We focus on differences in altitude rather than location. In a topographically highly diverse region like the Alps, the local climate is usually far more influenced by altitude than longitude or latitude. Furthermore, we consider only trends of 30-day mean values; extreme values are not part of this study.

In summary, this study attempts to bridge the gap between: (1) resolving trend statistics on a daily resolution; (2) overcoming the instabilities of the MK-test; (3) emphasising ‘trend timing’ (the day in the calendar year of a trend occurrence) as a new variable for characterising trends next to trend magnitude; (4) providing an overview of the regional hydro-climatic trend patterns, their altitude dependences and their interconnections; and (5) interpreting the trends in the context of anthropogenic climate change.

METHODS

Study area

The area under investigation is largely situated in North Tyrol, Austria, and the adjacent federal states just north of the main Alpine ridge (Figure 1). The core study region is part of the relatively dry central alpine region, which is located in the rain shadow of the northern and southern Alpine border ranges. It has an extent of roughly 200 km in the east–west and 60 km in the north–south direction. Altitudes range from roughly 500 m above sea level (a.s.l.) in the valleys to 3,700 m a.s.l. at the highest peaks. The region is characterised by a temperate dry to temperate humid climate, where stations at high elevations generally receive more precipitation. A continental seasonal cycle with a distinct summer maximum is prevalent (Figure 2(e)). With regard to runoff, all of the watersheds studied show seasonal behaviour. The spring to summer runoff peaks reflect the regimes typically found in mountainous regions (Figure 2(a)). Median air temperatures are found within the range of about −5 to +15 °C (Figure 2(b)). Snow conditions are highly variable, with median snow heights (SH) of 40 cm and maximum SH up to 120 cm, all averaged over a period of 1 month (Figures 2(c) and (d)).

Figure 1

Topographic map of North Tyrol and the surrounding central Alps with the location of the hydro-climatic stations analysed in this study.

Figure 1

Topographic map of North Tyrol and the surrounding central Alps with the location of the hydro-climatic stations analysed in this study.

Figure 2

Mean annual cycles: monthly medians (specific Q, T, SH) resp. sums (NSH, P) of hydro-climatic variables for each station (dashed line) and median of all stations (solid black line) for the period 1980 to 2010.

Figure 2

Mean annual cycles: monthly medians (specific Q, T, SH) resp. sums (NSH, P) of hydro-climatic variables for each station (dashed line) and median of all stations (solid black line) for the period 1980 to 2010.

Data

We analysed daily observations of mean temperature (T), precipitation (P), new snow height (NSH), total SH and runoff (Q). The data were provided by Hydrographischer Dienst Tirol, AlpS GmbH and Tiroler Wasserkraft AG, and quality-checked by Austrian government authorities. For better orientation throughout our analyses we assigned an identification number to each station (Table 1). For T, NSH, SH and P, this ID corresponds to the rank of station height; that is, the lower the station, the higher the value of the ID (Figure 3(a)). With regard to runoff, the station ID matches the median watershed altitude, as gauge altitude provides far less information about the hydrologic conditions found in the watershed. Prior to the analysis, runoff was transformed to specific runoff (in flow rate per unit area).

Table 1

Database studied: station ID, altitude, location (WGS 84), initial year of data records, watershed area (Q), average watershed altitude (Q)

RunoffNSH, SH, Precipitation
Alt.AreaAv. alt.Alt.Year
ID(m)LatitudeLongitudeYear(km²)(m)ID(m)LatitudeLongitudeNSH, SHP
2,640 46.8678 10.8007 1973 11 3,127         33 980 47.3828 12.4269 1901 1896 
1,891 46.8665 10.8895 1967 90 2,884 2,290 46.9975 11.1400 1978 1978 34 950 47.3008 10.9236 1897 1895 
1,895 46.9112 10.7142 1983 55 2,880 1,970 47.2083 11.0069 1978 1979 35 880 47.0569 10.6586 1895 1895 
1,883 46.8717 10.9998 1966 73 2,792 1,960 47.1369 11.8717 1979 1979 36 860 47.2489 10.7411 1895 1896 
1,995 46.9978 13.3937 1961 2,686 1,940 46.8667 11.0244 1919 1895 37 815 47.3864 12.1386 1901 1896 
1,337 47.0796 10.8312 1961 167 2,613 1,906 46.8589 10.9161 1928 1895 38 760 47.2058 10.8869 1910 1920 
1,321 47.0026 12.3380 1951 107 2,600 1,695 47.2300 10.9186 1948 1911 39 695 47.2500 10.8836 1979 1972 
1,687 47.1099 12.4551 1951 39 2,590 1,640 47.1075 11.6036 1978 1979 40 675 47.2764 10.9817 1964 1964 
1,186 47.0508 10.9598 1976 517 2,558 1,630 46.9564 10.4450 1947 1911 41 618 47.3089 11.0878 1964 1964 
10 1,824 47.0225 11.6877 1976 24 2,499 1,620 46.9847 10.8667 1918 1918 42 610 47.2706 11.2072 1978 1979 
11 1,486 47.0106 12.6358 1951 47 2,473 10 1,550 47.1256 10.9653 1976 1976 43 570 47.3078 11.8631 1966 1966 
12 1,902 47.2124 10.9994 1981 2,448 11 1,520 47.2039 11.1053 1977 1977       
13 1,544 46.9988 10.1747 1966 98 2,411 12 1,380 46.9667 11.0111 1947 1911 Temperature    
14 882 47.2185 12.2508 1961 81 2,354 13 1,360 47.0183 11.4264 1947 1946 IDAlt.(m)LatitudeLongitudeYear
15 1,600 47.1492 10.8909 1984 13 2,336 14 1,350 47.0969 10.6492 1925 1911 1,942 46.8667 11.0244 1953  
16 924 47.1707 10.9031 1951 786 2,323 15 1,340 47.0181 12.3756 1935 1896 1,794 47.3517 12.3592 1950  
17 1,019 47.1051 10.4541 1951 385 2,263 16 1,335 47.0767 10.8367 1930 1907 1,445 47.0056 11.5133 1950  
18 1,504 47.1037 12.4990 1951 60 2,241 17 1,315 47.1364 10.6983 1978 1956 1,388 46.9169 12.3539 1950  
19 931 46.9798 12.5290 1975 285 2,216 18 1,314 47.0331 10.7506 1897 1896 1,352 47.0047 12.6464 1960  
20 1,095 47.1329 10.7711 1978 220 2,159 19 1,280 47.2269 12.0444 1901 1896 1,330 46.8911 10.4969 1958  
21 880 47.2322 12.3276 1980 45 2,117 20 1,270 47.1622 11.7439 1901 1897 1,041 47.1425 10.9289 1960  
22 1,174 46.9661 13.1835 1961 85 2,081 21 1,235 47.0844 11.4183 1895 1895 1,009 47.2292 12.1819 1960  
23 1,113 47.2643 10.2867 1951 248 1,951 22 1,200 47.1956 11.1567 1915 1907 934 47.2236 12.9925 1950  
24 837 47.1456 13.1184 1951 221 1,937 23 1,180 47.0764 10.9700 1895 1895 10 858 47.3853 13.4561 1950  
25 888 47.2377 12.4921 1951 75 1,915 24 1,130 47.0736 11.2597 1947 1947 11 796 47.1403 10.5636 1950  
26 917 47.2233 12.9999 1961 242 1,841 25 1,120 47.1225 10.7819 1911 1895 12 791 47.4597 12.3583 1957  
27 958 47.3842 10.5389 1975 64 1,726 26 1,070 47.1717 11.3617 1967 1957 13 770 47.3267 12.7953 1950  
28 943 47.3144 10.0416 1956 42 1,701 27 1,041 47.1403 10.9328 1953 1953 14 661 46.8256 12.8064 1950  
29 849 47.3102 13.3112 1951 91 1,594 28 1,040 47.3736 11.9656 1959 1959 15 643 47.1594 11.8506 1950  
30 861 47.3487 12.7448 1961 151 1,550 29 1,010 47.1875 11.4047 1926 1926 16 578 47.2600 11.3567 1951  
31 830 47.2956 9.7195 1956 33 1,475 30 1,003 47.0039 12.5442 1904 1896 17 578 47.2600 11.3842 1950  
32 673 47.3881 9.8790 1951 229 1,449 31 1,000 47.0914 11.7925 1949 1911 18 490 47.5753 12.1628 1950  
33 958 47.4150 10.9159 1980 88 1,440 32 990 47.1311 11.4556 1895 1895 19 430 47.8014 13.0017 1950  
RunoffNSH, SH, Precipitation
Alt.AreaAv. alt.Alt.Year
ID(m)LatitudeLongitudeYear(km²)(m)ID(m)LatitudeLongitudeNSH, SHP
2,640 46.8678 10.8007 1973 11 3,127         33 980 47.3828 12.4269 1901 1896 
1,891 46.8665 10.8895 1967 90 2,884 2,290 46.9975 11.1400 1978 1978 34 950 47.3008 10.9236 1897 1895 
1,895 46.9112 10.7142 1983 55 2,880 1,970 47.2083 11.0069 1978 1979 35 880 47.0569 10.6586 1895 1895 
1,883 46.8717 10.9998 1966 73 2,792 1,960 47.1369 11.8717 1979 1979 36 860 47.2489 10.7411 1895 1896 
1,995 46.9978 13.3937 1961 2,686 1,940 46.8667 11.0244 1919 1895 37 815 47.3864 12.1386 1901 1896 
1,337 47.0796 10.8312 1961 167 2,613 1,906 46.8589 10.9161 1928 1895 38 760 47.2058 10.8869 1910 1920 
1,321 47.0026 12.3380 1951 107 2,600 1,695 47.2300 10.9186 1948 1911 39 695 47.2500 10.8836 1979 1972 
1,687 47.1099 12.4551 1951 39 2,590 1,640 47.1075 11.6036 1978 1979 40 675 47.2764 10.9817 1964 1964 
1,186 47.0508 10.9598 1976 517 2,558 1,630 46.9564 10.4450 1947 1911 41 618 47.3089 11.0878 1964 1964 
10 1,824 47.0225 11.6877 1976 24 2,499 1,620 46.9847 10.8667 1918 1918 42 610 47.2706 11.2072 1978 1979 
11 1,486 47.0106 12.6358 1951 47 2,473 10 1,550 47.1256 10.9653 1976 1976 43 570 47.3078 11.8631 1966 1966 
12 1,902 47.2124 10.9994 1981 2,448 11 1,520 47.2039 11.1053 1977 1977       
13 1,544 46.9988 10.1747 1966 98 2,411 12 1,380 46.9667 11.0111 1947 1911 Temperature    
14 882 47.2185 12.2508 1961 81 2,354 13 1,360 47.0183 11.4264 1947 1946 IDAlt.(m)LatitudeLongitudeYear
15 1,600 47.1492 10.8909 1984 13 2,336 14 1,350 47.0969 10.6492 1925 1911 1,942 46.8667 11.0244 1953  
16 924 47.1707 10.9031 1951 786 2,323 15 1,340 47.0181 12.3756 1935 1896 1,794 47.3517 12.3592 1950  
17 1,019 47.1051 10.4541 1951 385 2,263 16 1,335 47.0767 10.8367 1930 1907 1,445 47.0056 11.5133 1950  
18 1,504 47.1037 12.4990 1951 60 2,241 17 1,315 47.1364 10.6983 1978 1956 1,388 46.9169 12.3539 1950  
19 931 46.9798 12.5290 1975 285 2,216 18 1,314 47.0331 10.7506 1897 1896 1,352 47.0047 12.6464 1960  
20 1,095 47.1329 10.7711 1978 220 2,159 19 1,280 47.2269 12.0444 1901 1896 1,330 46.8911 10.4969 1958  
21 880 47.2322 12.3276 1980 45 2,117 20 1,270 47.1622 11.7439 1901 1897 1,041 47.1425 10.9289 1960  
22 1,174 46.9661 13.1835 1961 85 2,081 21 1,235 47.0844 11.4183 1895 1895 1,009 47.2292 12.1819 1960  
23 1,113 47.2643 10.2867 1951 248 1,951 22 1,200 47.1956 11.1567 1915 1907 934 47.2236 12.9925 1950  
24 837 47.1456 13.1184 1951 221 1,937 23 1,180 47.0764 10.9700 1895 1895 10 858 47.3853 13.4561 1950  
25 888 47.2377 12.4921 1951 75 1,915 24 1,130 47.0736 11.2597 1947 1947 11 796 47.1403 10.5636 1950  
26 917 47.2233 12.9999 1961 242 1,841 25 1,120 47.1225 10.7819 1911 1895 12 791 47.4597 12.3583 1957  
27 958 47.3842 10.5389 1975 64 1,726 26 1,070 47.1717 11.3617 1967 1957 13 770 47.3267 12.7953 1950  
28 943 47.3144 10.0416 1956 42 1,701 27 1,041 47.1403 10.9328 1953 1953 14 661 46.8256 12.8064 1950  
29 849 47.3102 13.3112 1951 91 1,594 28 1,040 47.3736 11.9656 1959 1959 15 643 47.1594 11.8506 1950  
30 861 47.3487 12.7448 1961 151 1,550 29 1,010 47.1875 11.4047 1926 1926 16 578 47.2600 11.3567 1951  
31 830 47.2956 9.7195 1956 33 1,475 30 1,003 47.0039 12.5442 1904 1896 17 578 47.2600 11.3842 1950  
32 673 47.3881 9.8790 1951 229 1,449 31 1,000 47.0914 11.7925 1949 1911 18 490 47.5753 12.1628 1950  
33 958 47.4150 10.9159 1980 88 1,440 32 990 47.1311 11.4556 1895 1895 19 430 47.8014 13.0017 1950  
Figure 3

(a) Station altitude (or watershed median altitude for Q, respectively), P stations are equal to NSH and SH stations; (b) data availability.

Figure 3

(a) Station altitude (or watershed median altitude for Q, respectively), P stations are equal to NSH and SH stations; (b) data availability.

Like in all mountain regions, station density is higher at lower altitudes and vice versa (Figure 3 and Table 1). This applies especially for the temperature variables, where general low station availability in the area under investigation adds to the problem.

The MK test for trend detection

The calculation of the trend significance followed the methods proposed by Mann and Kendall (Mann 1945; Kendall 1975; Hensel & Hirsch 1992). The non-parametric rank-based MK test is a standard test in hydroclimatology to detect whether a trend is significant (e.g. Gagnon & Gough 2002; Birsan et al. 2005; Schimon et al. 2011). Its popularity in hydrological sciences stems from the robustness against outliers, the absence of requirements concerning the distribution of the data and its high statistical power compared with other trend tests. This means that there is a high probability that a trend is correctly detected in the case one is present (Hensel & Hirsch 1992; Hess et al. 2001). The MK test was criticised in recent publications (e.g. Hamed & Rao 1998; Yue et al. 2002): the presence of autocorrelation in time series results in the overestimation of the significance of a trend. For this reason, the MK test procedure was adjusted to account for autocorrelation using the pre-whitening method published by Wang & Swail (2001). However, as serial correlation coefficients of the data sets considered were mostly small, there were hardly any differences between the original and the adjusted MK test results.

The Sen's Slope Estimator for trend magnitude

If the MK test revealed a significant trend at the α = 0.1 level, we calculated the Sen's Slope Estimator, a linear regression method that is robust against non-normal data and outliers (Sen 1968; Hensel & Hirsch 1992). First, it is necessary to compute the slope between all possible pairs s of data points. Afterwards, the Sen's Slope is calculated as the median of all slopes. Following the common language, when speaking of positive (negative) trends in the present study, this implies that there is a significant trend of the mean with a positive (negative) value of the Sen's Slope Estimator. Similar to Nayak et al. (2010), for example, the Sen's Slope is also referred to as the ‘trend magnitude’ throughout this study.

30-day moving average trend statistics

The results of trend analyses highly depend on the time period that is used for aggregating the values. If we take the commonly used yearly or seasonal averages, we lose a lot of the information contained in the data. Monthly values improve this issue, but are still very coarse. However, the shorter the aggregation time period chosen, the higher the variability of the aggregated data. For example, if we take daily time steps, we study a time series that consists only of, for example, January 1st values. As the interannual variability of daily values is a lot higher than that of monthly averages, the MK test is less capable of detecting trends. For this reason we applied the method of 30-day moving average (30DMA) trends, where a 30-day window is moved over the data set and tested daily on significance and trend magnitude. The approach is based on the study of Kim & Jain (2010), who derived daily resolution streamflow trends on the basis of 3-day averages for the western USA. We decided to increase this window to 30 days, to further lower the variability of the averaged data sets. The 30-day moving average is calculated as follows: 
formula
1
with m = 1, …, 365 (or 366 in case of a leap year) and x as the data value on the day of year (DOY) m for which the 30DMA is calculated. The 30DMA is assessed for all 365 days of a year and centred on the day of reference. Following the calculation of the 30-day moving averages, the MK test and the Sen's Slope method are applied to the filtered time series by analysing each day of the year separately. Finally, for each station of each variable, a 365-day data set with the results of the 30DMA trend tests is available. This indicates whether a 30-day trend exists on one of the days. If a trend exists, the time series provides the slope of the trend. Positive values represent trends in the upward direction, negative ones trends in the downward direction. The day in the calendar year in which a specific significant 30DMA trend occurs is referred to as ‘trend timing’ throughout this study.

Period selection and composite

The first part of our analysis was the moving average MK statistics; the second part was the investigation of the trend stability within different periods.

To separate climate signals from time-series variability, it is of major importance to point out the approximate time when anthropogenically driven climate change began, as trend analyses only reveal global relations in a data set. This point is crucial, since it means that the results of trend analyses highly depend on the time span chosen for study, and comparing trends derived from different periods can lead to misleading results (Laternser & Schneebeli 2003; Schönwiese & Janoschitz 2005). This issue has received only limited attention and is still the reason for some contradictory trends in the literature (Zhang et al. 2010).

Owing to the fact that three-quarters of the human-induced increase in CO2 concentration has occurred between the middle of the last century and today (Steffen et al. 2007), we chose 1950 as a starting point. The exponential character of the rise in atmospheric CO2 since 1950 made us believe that, if there were any signals in our time series, we would find them in a period that starts in 1950 or later. In this context, Kundzewicz (2004) underlines that the probability of change detection increases with intensifying climate change, because the impacts may be greater and persist longer.

The number of periods of different length to test for the existence of a trend was limited by the time span that allows feasible trend analyses. The literature recommends periods of more than 25 to 30 years; shorter time spans cause instability of the resulting trend magnitudes (Burn & Elnur 2002; Schönwiese & Janoschitz 2005). For this reason, we set the start of the shortest period to 1980. So the original time series was arranged into four data sets of different length (periods chosen: 1950–2010, 1960–2010, 1970–2010 and 1980–2010). The difference of 10 years between the initial years of the periods was arbitrarily chosen but reflects the literature (e.g. Zhang et al. 2001) and our experience concerning trend changes. Similar to Burn & Elnur (2002), if the length of all missing data points of a station data set was more than 5 years, the station affected was not further considered for this period. For each one of these new data sets, we calculated the 30-day moving average trend statistics.

Then, a composite procedure was chosen to filter just the trend signals that persisted throughout all periods studied. This was especially important for the snow variables, as there was a lot of noise in the trend results. The final values in the diagram are mean values of the four different periods. This ensured that the composites only featured trends that tested significant in all periods. In some cases, not enough stations existed back to the 1950s for a clear picture of trend variety. In other cases, trends were not visible in the 1950 to 2010 period but in the 1970 to 2010 one, for example, indicating that trends occurred only in the more recent periods. For this reason, a second composite was produced, including only the last two periods (1970–2010 and 1980–2010). For comparison we included both the four- and the two-period composites in the results section.

Effect of altitude on trend timing and magnitude

To understand the processes that cause trends in hydro-climatic variables, it is helpful to study the effect of altitude on these changes. First, we inspected 30DMA seasonal trend timing: when analysed as a function of the DOY, 30DMA trends occur in conglomerations, referred to as ‘trend clusters’ throughout this study. As demonstrated in the results section, these trend clusters appear not only at one station but often at various stations analysed. These corresponding clusters will be labelled as ‘trend cluster families’. Sometimes these clusters display a shift; that is, they occur earlier or later at stations with different altitudes. To compare the timing of these different trend occurrences, we calculated the central moment (‘centre of gravity’; unit: DOY) of the trend clusters under investigation. This central moment of trend timing is a more robust measure for characterising trend occurrences than the DOY of the maximum trend magnitude, because it includes all trend magnitudes in a cluster 
formula
2
p(q) is the DOY when the first (last) significant trend in a trend cluster turns up. As an illustrative example, Figure 4 displays schematically two runoff trend clusters (cf. Figure 6). The timing of their central moments is marked.
Figure 4

Schematic illustration of two trend clusters and the central moment of their timing within the calendar year versus trend magnitude.

Figure 4

Schematic illustration of two trend clusters and the central moment of their timing within the calendar year versus trend magnitude.

The hypothesis of no correlation was tested to analyse whether the correlation coefficients between station height (or mean basin height) and the central moment of trend timing were significant or not. To analyse the effect of altitude on trend magnitude, a similar procedure was carried out: the 0.95 quantile trend magnitudes (a more robust measure than maximum trend magnitude) of the cluster family under investigation was computed and finally analysed on correlation to the corresponding station or basin altitudes.

RESULTS

MK test based on yearly average values

The results of the standard method for detecting trends – the MK test based on yearly average values – is shown for selected stations and varying time periods in Figure 5: the initial year of analysis is constantly raised by 1 year while the final year 2010 remains constant. Despite being the same time series, this evidently affects magnitude and significance of trends.

Figure 5

Trend magnitude of yearly averaged runoff for selected stations as a function of the initial year of the data set analysed (final year consistently 2010); solid (dashed) line where trend is (not) significant (MK test, α = 0.1); colour bar indicates runoff station ID. Please refer to the online version of this paper to see the figure in colour: http://www.iwaponline.com/jwc/toc.htm.

Figure 5

Trend magnitude of yearly averaged runoff for selected stations as a function of the initial year of the data set analysed (final year consistently 2010); solid (dashed) line where trend is (not) significant (MK test, α = 0.1); colour bar indicates runoff station ID. Please refer to the online version of this paper to see the figure in colour: http://www.iwaponline.com/jwc/toc.htm.

The analysis provides a very heterogeneous picture of the runoff trends found in the study area. There are only a few stations with continuous significant trends. Most stations exhibit trends in the positive direction, but absolute trend magnitudes are mostly unstable. In the 1980 to 2010 period, only seven of the 33 available stations (1970 to 2010 period: six of the 19 available stations) depict a significant trend (not shown).

30-day moving average trend statistics

The results of the 30DMA trend analysis are plotted in Figure 6 for the 1980–2010 runoff time series at gauging station Vent, showing the trend behaviour of runoff in a strongly glaciated head watershed in the Oetztal Alps. A non-zero value on, for example, January 15th means that for the 30 days on either side of this value (January 1st to January 30th) a trend has been detected. This value indicates the magnitude and the direction (positive/negative) of the slope of this trend.

Figure 6

30-day moving average MK trend statistic compared with conventional, monthly MK trend statistic for the 1980–2010 runoff time series at gauging station Vent.

Figure 6

30-day moving average MK trend statistic compared with conventional, monthly MK trend statistic for the 1980–2010 runoff time series at gauging station Vent.

The 30DMA trend statistic reveals additional information compared with the conventional one (Figure 6). Instead of mere monthly information (indicated as triangles in Figure 6), there is daily information on trend presence and magnitude. The magnitude of the 30DMA trend on the 15th of a month is approximately the same as the one calculated by conventional monthly MK statistics. The 30DMA trends do not appear on single days but, as mentioned before, in clusters. These clusters are sometimes too short to be detected as significant by conventional monthly trend analyses (e.g. the cluster in September). The trend cluster in spring has a clear maximum in May, showing that positive runoff trends in Vent are strongest in late spring.

Maximum trends of 0.0025 m3 (s km2)−1 per year equal about 5% of the average annual specific runoff of 0.0495 m3 (s km2)−1. However, the 30DMA trend magnitudes have to be interpreted carefully, as they just reflect the situation on one specific 30-day period in the calendar year.

Period composite

Elevation above sea level is one of the key factors that influence hydro-climatic variables in mountainous areas. For this reason we visualised the 30DMA MK statistics for all stations in the study area ordered by height or mean basin height. Figure 7 shows an example of the four time periods that were used for the production of the runoff composites. For the other variables, only the final results of the period composites are provided. Colour bars in all following figures correspond to the 30DMA trend magnitude.

Figure 7

Annual distribution of runoff trend magnitude; x-axis: month; y-axis: station ID (ordered by mean basin height, descending); z-axis (colour): trend magnitude of 30-day means (only where significant trend detected); grey lines mark stations without enough available data. Please refer to the online version of this paper to see the figure in colour: http://www.iwaponline.com/jwc/toc.htm.

Figure 7

Annual distribution of runoff trend magnitude; x-axis: month; y-axis: station ID (ordered by mean basin height, descending); z-axis (colour): trend magnitude of 30-day means (only where significant trend detected); grey lines mark stations without enough available data. Please refer to the online version of this paper to see the figure in colour: http://www.iwaponline.com/jwc/toc.htm.

Runoff

Generally, the trend timing is in agreement among most of the stations and periods studied (Figure 7). Positive springtime trend clusters exist for nearly all stations, reflected by the yellow-reddish hues. Depending on the mean altitude of the basin, they appear earlier or later in the year: The higher the altitude, the higher the DOY and the stronger the trends that occur. Negative trends follow in summer (bluish hues), with a higher total amount of trend clusters and greater magnitudes at lower stations. The timing of these trends seems to depend on basin height as well. Afterwards, a less pronounced positive trend cluster family emerges in autumn; however, this seems to be less dependent on basin height. All of the trend cluster families mentioned are persistent throughout all of the periods.

Only very few runoff records date back to the 1950s. For this reason, only three periods (1960–2010, 1970–2010, 1980–2010) were used for the production of Figure 8(a). For all other variables, four-period composites are shown supplementary to the two-period ones.

Figure 8

Composites of runoff trends in different periods: (a) 1960–2010, 1970–2010 and 1980–2010 (‘long-term trends’); (b) 1970–2010 and 1980–2010 (‘recent trends’).

Figure 8

Composites of runoff trends in different periods: (a) 1960–2010, 1970–2010 and 1980–2010 (‘long-term trends’); (b) 1970–2010 and 1980–2010 (‘recent trends’).

As expected, these composites show a similar picture compared with the analyses of the single periods above: a positive springtime trend cluster family is followed by a negative one in summer, both altitude dependent. Autumn trends tend not to depend on altitude and only appear at some stations. Comparing the two- and four-period composites, there are small differences in trend timing, but general patterns stay the same. However, trend magnitude is roughly twice as great in the two-period composite compared with the three-period one.

Snow

The NSH and SH composites show a decreasing trend cluster family in springtime that is clearly associated with altitude (Figures 9 and 10). Both variables reveal very similar patterns concerning trend timing. The different periods studied are in good agreement with each other. Magnitude differences between the two- and four-period composites explicitly indicate stronger trends in the last two periods studied. Comparing both variables, NSH trends are roughly half the magnitude of the SH trends. At a few stations there is an indication that negative autumn trends in both NSH and SH do exist. However, springtime trends are much more abundant.

Figure 9

Composites of ‘NSH’ trends in different periods: (a) 1950–2010, 1960–2010, 1970–2010 and 1980–2010; (b) 1970–2010 and 1980–2010.

Figure 9

Composites of ‘NSH’ trends in different periods: (a) 1950–2010, 1960–2010, 1970–2010 and 1980–2010; (b) 1970–2010 and 1980–2010.

Figure 10

Composites of ‘total SH’ trends in different periods: (a) 1950–2010, 1960–2010, 1970–2010 and 1980–2010; (b) 1970–2010 and 1980–2010.

Figure 10

Composites of ‘total SH’ trends in different periods: (a) 1950–2010, 1960–2010, 1970–2010 and 1980–2010; (b) 1970–2010 and 1980–2010.

Temperature

Concerning temperature data series, a concentration of positive trends is apparent in the spring to early summer months (Figure 11). It seems that the trend cluster family has two peaks, one in April and one in May to June. The timing is uniform over all stations analysed; that is, it is not dependent on altitude. Differences between the two- and four-period composites are merely the different trend magnitudes. The averaged temperature trend is about twice as strong in the two-period composite compared with the four-period composite, similar to what was observed for Q, NSH and SH trends. Furthermore, a weak positive trend signal is visible at lower stations in autumn in the ‘recent time’ composite.

Figure 11

Composites of temperature trends in different periods: (a) 1950–2010, 1960–2010, 1970–2010 and 1980–2010; (b) 1970–2010 and 1980–2010.

Figure 11

Composites of temperature trends in different periods: (a) 1950–2010, 1960–2010, 1970–2010 and 1980–2010; (b) 1970–2010 and 1980–2010.

Total precipitation

Precipitation does not show any clear signals (Figure 12). Thus, it clearly differs from the other analysed variables. Throughout the year there are mostly small positive trend clusters that do not indicate any pattern. Compared with the two-period composite, the four-period one provides a reduced picture of heterogeneous trends.

Figure 12

Composites of precipitation trends in different periods: (a) 1950–2010, 1960–2010, 1970–2010 and 1980–2010; (b) 1970–2010 and 1980–2010.

Figure 12

Composites of precipitation trends in different periods: (a) 1950–2010, 1960–2010, 1970–2010 and 1980–2010; (b) 1970–2010 and 1980–2010.

Height dependence

Illustrative of the positive runoff trend cluster family in spring, Figure 13 shows the steps required for analysing the height dependence of trends. First, a composite and a trend cluster family have to be chosen (Figure 13(a)). The more periods considered in a composite, the fewer data sets are available, so for some variables it was necessary to take composites of fewer periods to cover a larger data set. Furthermore, some trends only appear in the more recent periods, so every single composite was analysed in order to avoid missing any cluster family. Herewith it was possible to identify the composite with the strongest relationship of altitude to trend timing as well as magnitude.

Figure 13

(a) Composite of two runoff periods; (b) only positive trends in DOYs 1–200; (c) trend timing and trend magnitude vs. mean basin height; line: best fit line where correlation significant.

Figure 13

(a) Composite of two runoff periods; (b) only positive trends in DOYs 1–200; (c) trend timing and trend magnitude vs. mean basin height; line: best fit line where correlation significant.

In a following step, clusters with similar timing and trend direction (positive/negative) are filtered to find the trend cluster families. This is done by: (1) assessing the DOY of the whole family with the possible shift at other stations; and (2) considering only positive or negative trends (Figure 13(b)). For each trend cluster of each station, both the central moment of trend timing and the trend magnitude, characterised by the 0.95 quantile, are calculated in a final step (Figure 13(c)).

Both Table 2 and Figure 14 provide information on the specific trend cluster families and their altitude dependence. For this purpose, it was tested whether the correlation coefficients between station height (or mean basin height) and the central moment of trend timing were significant or not. The lower the p-value, the higher the probability that the relationship between trend timing and station altitude or mean basin height is not just due to coincidence. For this reason, we selected the composites with the lowest p-values for the representation in Table 2 and Figure 14. Concerning precipitation, it was not possible to identify any trend cluster families.

Table 2

Summary of the 30DMA trend statistics for trend timing

VariableQNSHSHT
Timing of trend occurrence 
Season Spring Summer Autumn Spring Spring Spring 
DOY 1–200 100–220 240–360 70–130 70–130 60–200 
Direction Positive Negative Positive Negative Negative Positive 
Composites: Correlation of trend timing to station height and mean basin height 
No. of periods 
Slope 0.069 0.042 0.025 0.028 0.019 −0.0006 
Corr_coeff 0.84 0.8 0.53 0.91 0.57 −0.055 
p-valuea 5.20 × 10−7 0.0019 0.079 3.20 × 10−6 0.026 0.8 
VariableQNSHSHT
Timing of trend occurrence 
Season Spring Summer Autumn Spring Spring Spring 
DOY 1–200 100–220 240–360 70–130 70–130 60–200 
Direction Positive Negative Positive Negative Negative Positive 
Composites: Correlation of trend timing to station height and mean basin height 
No. of periods 
Slope 0.069 0.042 0.025 0.028 0.019 −0.0006 
Corr_coeff 0.84 0.8 0.53 0.91 0.57 −0.055 
p-valuea 5.20 × 10−7 0.0019 0.079 3.20 × 10−6 0.026 0.8 

aBold if significant at the α = 0.1 level

Figure 14

Relationship between station height (for NSH, SH and T) and mean basin height (for Q) to trend timing; best-fit line: solid (dashed) line in case of (non-) significant correlation; diagram title indicates: 1. variable, 2. initial and final DOY, 3. trend direction.

Figure 14

Relationship between station height (for NSH, SH and T) and mean basin height (for Q) to trend timing; best-fit line: solid (dashed) line in case of (non-) significant correlation; diagram title indicates: 1. variable, 2. initial and final DOY, 3. trend direction.

An altitude dependency was identified for most of the trends in hydro-climatic variables. Mainly the trend timing rather than the trend magnitude was found to be altitude dependent. One exception exists: the springtime runoff trend magnitude of a two-period composite was found to be altitude dependent, showing that trends in higher basins are stronger than in lower ones (Figure 13(c)). However, this relationship is not very strong.

The timing of all Q trend cluster families studied depends on the mean basin altitude. However, the relationship between trend timing and autumn trends is very weak, and significance would not have been detected with a lower, more conservative α. With regard to NSH and SH, timing of springtime trends depends on station altitude as well; both trends were found during the same periods of the calendar year. No altitude dependence was detected for T trends.

The persistence of trend timings throughout different periods was quite strong in all variables. Unfortunately, data availability limited the informative value of the analyses of the longer periods.

DISCUSSION

30-day moving average trend statistics

Owing to the high seasonal trend variability (negative next to positive, significant next to non-significant trends), it is difficult to identify trends on the basis of yearly, seasonal or even monthly averages. Consistent with the observations, trends of hydro-climatic variables are caused by a variety of nonlinear processes and are not restricted to fixed periods. Contradictory trends found with these methods due to different data set lengths are not exceptional (Viviroli et al. 2011). In the example provided above, trend magnitude and significance based on analyses of yearly runoff data sets are highly variable.

It is obvious that conventional trend analyses of monthly averages provide a finer resolution than trends of yearly totals. The critical reader might ask whether these trends in monthly resolution are not enough and if it is not possible to achieve similar findings to those of 30DMA-analyses. On the one hand, trend tests based on daily resolution are able to detect trends that could be missed by monthly tests: for instance, the negative runoff trends in autumn shown above were missed by conventional monthly trend statistics. On the other hand, in most cases the monthly statistics are not capable of capturing the height dependence of a trend cluster family, as the trend timing varies in some variables (e.g. NSH) just about a few days.

Compared to monthly averages, the widely used trend tests of 3-month averages could result in even smaller magnitudes and less significance. In addition, one runs the risk of possibly balancing trend magnitudes because of opposing trends found over a period of 3 months.

Trend timing as a more stable measure to characterise trends in addition to trend magnitude could reveal valuable information that has not been found with conventional trend analyses. The additional information can help elucidate the processes that cause these trends.

Note that one must keep in mind that we conducted, for example for runoff, 33 times 365 tests in total. With an α-level of 0.1, we run the risk that 10% of the results presented could appear also by chance. For this reason, we carried out additional tests with lower significance levels (not shown). In these plots, the trend cluster families found earlier still remained, but were less obvious. However, these coherent patterns confirmed our hypothesis that the structures found are not just due to chance.

Runoff

Especially in mountainous regions, runoff integrates a variety of processes driving the hydrology of a watershed: the lower the mean altitude of a watershed, the more the runoff regime is influenced by snow melt processes and precipitation compared with glacial melting. The higher the mean altitude of a watershed, the more the runoff regime is driven by glacial melt and the later in the year the annual runoff peak is observed (cf. Figure 2). Due to rising temperatures, this peak is moving towards earlier DOYs, producing a forward shift in both the rising and the falling limbs of the mean annual hydrograph, as observed throughout our analysis. Similar patterns in daily resolution streamflow trends have been observed by Kim & Jain (2010) for the western USA; however, no glacially influenced watersheds were analysed. Trends in the rising limb are most likely caused by increased glacial melt, earlier snow melt and less precipitation falling as snow. Trend magnitude increases with mean basin altitude, demonstrating that the hydrology of high elevation watersheds changes the most dramatically. The trends in the falling limb are possibly caused by reductions in the snow water content; thus an earlier depletion of the snow reservoir, and earlier and more intense evapotranspiration because of earlier snow-free conditions and higher temperatures. At the more glacially influenced gauges (lower station IDs), trends in the falling limb are mostly missing, possibly because there is enough glacial melt water to compensate for the water missing due to earlier melting (contrary to the snow reservoir, which is limited to the resources of at most one winter's snowfall). However, in the 1980–2010 period, trends in the falling limb occur at some of these gauges as well. This is possibly caused by a reduction of glacial melt water because glacier mass at some glaciers is already reduced in such a way that increased temperature will not result in additional runoff (cf. Braun et al. 2000).

Positive runoff trends in autumn display a different behaviour: maximum trend magnitude is smaller by more than a power of ten than that of spring or summer Q trends. Furthermore, trend timing depends considerably less on mean basin height. For this reason, we assume that these trends are caused by reasons other than the spring and summertime trends. Possibly they could be attributed to melting processes of freshly fallen snow that happen often during this time of year. These SH are generally smaller and more similar over different altitudes than the SH of a spring snow cover that has grown over the whole winter season. The drivers of these processes are most probably the trends in temperature, which occur during this season.

Reviewing the literature on runoff trends, there is an agreement on earlier spring freshet in mountain watersheds due to earlier snowmelt and more precipitation falling as rain instead of snow, but most of the studies were carried out for catchments in North America (Hamlet et al. 2005; Mote et al. 2005; Knowles et al. 2006; Viviroli et al. 2011). Less research has been conducted on observed runoff data from the European Alps: Birsan et al. (2005) extensively analysed mean runoff in Switzerland and found increasing trends in winter, spring and autumn. Schimon et al. (2011) detected only few significant trends and homogeneous patterns for streamflow trends in Austria.

Snow

Most previous research on changes in SH in the Alps has not considered seasonal differences, but rather interannual or spatial differences of yearly averages, extreme values, number of snow and snowfall days or single characteristic days of the year (Föhn 1990; Laternser & Schneebeli 2003; Mote et al. 2005; Marty 2008). Renard et al. (2008) found that snow melt starts noticeably earlier in the French Alps. For the western USA, more studies came to similar conclusions.

Our analyses confirm that decreases in SH and solid precipitation in the study area occur predominantly in spring, similar to analyses in the continental USA (American Meteorological Society 2012), and that there is almost no significant trend apparent in autumn. We assume that the trends in NSH are due to proportionately more liquid and less solid precipitation. The changes in SH most likely originate both from the negative trends in NSH and from earlier melting. The drivers of these developments are most likely the temperature increases and the resulting snow–albedo feedback processes during the same period (Groisman et al. 1994). These are especially strong in springtime as the incident solar radiation is much stronger than that in autumn (Hall et al. 2008).

Temperature

The present study showed that the mean temperature trend magnitude has increased about twice between the two- and four-period composites. This goes along with the average temperature increase, which amplified with intensifying climate change in recent decades (Schönwiese & Janoschitz 2005). Changes in temperature are most likely the main driver for the trends of most other variables studied. Most of the NSH, SH and Q trends occur during similar DOYs as the T trends and have intensified in recent decades as well. The timing of the T trends in spring and early summer corresponds to other observations (Ceppi et al. 2012) and climate projections (e.g. Déqué et al. 2007) and is a result of the snow–albedo feedback processes and its subsequent effects on local climate (Hall et al. 2008).

Temperature trends have been widely studied, with an obvious altitude dependence of trend magnitude in climate projections or reanalyses. But in situ measurements often show a different, sometimes opposite picture (Ceppi et al. 2012; Rangwala & Miller 2012). In our analyses, we did not find any altitude dependence of trend magnitude or trend timing. This might be due to the fact that the snow–albedo feedback processes do not have significant impacts at the local scale: The higher the station, the later in the year the mean daily temperatures pass the freezing point. The snow–albedo feedback processes are effective especially during this period. So if these processes would take place on a local scale, at least trend timing as a more stable measure than trend magnitude would appear with a temporal offset and hence would be altitude dependent. For this reason we assume that the snow–albedo feedback effects take place over a greater region during a similar time of year. These results have crucial consequences for hydrological and climatological modelling, as the altitude dependence of temperature trends is a key aspect for climate change projections (Ceppi et al. 2012). However, for a more general conclusion, it would be advisable to include more stations into this analysis, especially high-altitude ones. Moreover, air mass exchanges and orographic effects such as aspect and shading may override possible altitude effects.

Precipitation

Contrary to all other variables studied, precipitation observations do not exhibit any clear trend signal. We expected to find at least positive springtime trends in total precipitation due to an artefact effect caused by a smaller measurement error because of proportionately more rain and less snow, such as that detected in the arctic by Forland & Hanssen-Bauer (2000).

Frei & Schär (2001), Bates et al. (2008) and Morin (2011) pointed out the problem of minimum detectability: a trend below a certain threshold might be undetectable due to the high variability of precipitation. In the winter months, NSH trend behaviour can be expected to be similar to P trend behaviour. But while NSH trends are obviously over the threshold to be detected, P trends are not.

It is possible that detectability poses no problem and in reality there is no trend at all in precipitation. The studies of Bates et al. (2008) and Dobler et al. (2010) support this assumption: in the majority of climate scenarios, the Alpine region is the transition region between the projected precipitation decreases in southern Europe and increases in northern Europe, which would make the absence of a trend plausible.

CONCLUSIONS

Especially with regard to the European Alps, the literature lacks trend studies of observational data with consistent results and clear explanations for the impulses that triggered these trends. Most of the studies on this subject derive trends of data sets containing yearly, seasonal or monthly totals following the conventional method. This standard way of calculating trend significance and magnitude of averaged values did not prove to be a robust method for detecting climate change signals. One reason for the many insignificant runoff trends found in the literature is the fact that positive and negative subseasonal trends compensate for and balance each other. This can happen not only for yearly but also for seasonal averages.

We found that it is important to explicitly identify the subseasonal timing of trends to get an idea of the reasons for these changes, and that trend timing is a more stable measure for characterising trends than trend magnitude. Our results are mostly consistent over all of the stations analysed. The study assimilates trend analyses of different variables for a holistic perspective of the changes in Alpine hydroclimatology. To cope with the constraint of analysing only global trends in a data set, we found it helpful to consecutively shorten the data set studied, calculate trend statistics and afterwards produce a composite of these statistics. Compared with many other studies on hydro-climatic trends, this paper deals with trends in one single region. Due to the complexity of our results, we put our focus on the altitude dependence, not on the spatial variability. This could be a task for subsequent research.

The majority of the trends found do occur during spring and early summer (NSH, SH, Q, T). It is not certain whether the trends were caused by human greenhouse gas emissions or by natural variability. But it is safe to say that these trends occur during the period of anthropogenic climate change and that they are mostly consistent and explainable by the timing of trends in temperature, one of the main drivers of climate variability. With regard to temperature, runoff and snow trends, we found a clear amplification of the trend signal in recent decades.

Concerning runoff, the multitude of processes involved in runoff generation is reflected by the seasonal differences in trend signals. Most of the previous studies compare rivers of different regions without having a closer look at their mean basin altitude, resulting in heterogeneous or insignificant trends. We found that the timing of runoff trends depends strongly on the altitude, and with it the runoff regime of the river studied. This means that, when comparing rivers from different regions, it is necessary to study only rivers of a similar mean basin altitude. If not, one might run the risk of obtaining results that declare, for example, that in one season one region shows a different tendency than another region, maybe just because of a general lower/higher mean altitude of the first region, and thereby different runoff regimes studied compared with those in the second region. And even when only comparing trends of one single region, the 3-month averaged trend results might give incoherent results just due to different runoff regimes studied. With regard to SH and snowfall, we emphasise that trends were only found in spring and not in autumn, underlining the impacts of springtime snow–albedo effects. Concerning temperature, our study implies that the effects of climate change on temperature do not depend on station altitude in our study area. For temperature, neither trend timing (as a more stable measure) nor trend magnitude show a clear relationship with altitude. With regard to precipitation, it is uncertain whether consistent precipitation trends could not be found due to undetectability or due to the absence of persistent trends. At least we can state that there is no seasonal balancing of positive and negative trends that hinders trend detection in precipitation records.

Although there is no guarantee that the trends observed in the past will continue for the next decades as well, we want to identify some of the expected implications for the main stakeholders. First, the ski season will end earlier because of a reduced amount of snow and less snowfall. Snow production in spring is getting more difficult owing to the higher temperatures. Second, hydropower producers have to face the effects of alternating runoff regimes. The overall amount of water available does not change significantly (cf. Schimon et al. 2011), but retreating glaciers, earlier snowmelt and less precipitation falling as snow lead to a loss of storage and seasonal retention capacities especially in springtime. These results are in agreement with the future runoff projections of various modelling approaches for the alpine region (e.g. Leonhardt et al. 2009; Magnusson et al. 2010; Tecklenburg et al. 2012). Third, as the region under investigation is a part of the relatively dry central Alpine region, agricultural irrigation during the summer season is common practice. Since less water will be available in summer, water scarcity might be an issue in dry years, especially after the disappearance of glaciers. With its extensive agricultural sector, the region just south of our study area (South Tyrol) will probably be affected even more severely in that context.

The annual water availability in northern Tyrol is characterised by a precipitation maximum in summer, a runoff maximum in spring to summer and low flows in winter. Thus, the runoff changes detected result in a seasonal balancing of the water budget. From a water resource manager's point of view, water availability is changing beneficially. However, the economic (tourism, agriculture) and ecological impacts of the changing Alpine hydro-climate are a different matter.

ACKNOWLEDGEMENTS

The authors would like to thank in particular Hydrographischer Dienst Tirol, AlpS GmbH and Tiroler Wasserkraft AG for providing local station data and for many stimulating discussions. This research was supported by the Potsdam Research Cluster for Georisk Analysis, Environmental Change and Sustainability (PROGRESS), founded by the German Federal Ministry of Education and Research.

REFERENCES

REFERENCES
American Meteorological Society
2012
Climate Change – An information statement of the American Meteorological Society
.
Bull. Amer. Met. Soc.
.
Bard
A.
Renard
B.
Lang
M.
2011
Understanding trends in hydrologic regimes. AdaptAlp WP 4 Report
,
Lyon, France
.
Bates
B. C.
Kundzewicz
Z. W.
Wu
S.
Palutikof
J. P.
(eds)
2008
Climate Change and Water. Technical Paper of the Intergovernmental Panel on Climate Change
.
IPCC Secretariat
,
Geneva
.
Birsan
M. V.
Molnar
P.
Pfaundler
M.
Burlando
P.
2005
Streamflow trends in Switzerland
.
J. Hydrol.
314
(
1–4
),
312
329
.
Braun
L. N.
Weber
M.
Schulz
M.
2000
Consequences of climate change for runoff from Alpine regions
.
Ann. Glaciol.
31
,
19
25
.
Brunetti
M.
Lentini
G.
Maugeri
M.
Nanni
T.
Auer
I.
Böhm
R.
Schöner
W.
2009
Climate variability and change in the Greater Alpine Region over the last two centuries based on multi-variable analysis
.
Int. J. Climatol.
29
,
2197
2225
.
Burn
H. B.
Elnur
M. A. H.
2002
Detection of hydrologic trends and variability
.
J. Hydrol.
255
,
107
122
.
Ceppi
P.
Scherrer
S. C.
Fischer
A. M.
Appenzeller
C.
2012
Revisiting Swiss temperature trends 1959–2008
.
Int. J. Climatol.
32
(
2
),
203
213
.
Déqué
M.
Rowell
D. P.
Lüthi
D.
Giorgi
F.
Christensen
J. H.
Rockel
B.
Jacob
D.
Kjellström
E.
de Castro
M.
van den Hurk
B.
2007
An intercomparison of regional climate simulations for Europe: assessing uncertainties in model projections
.
Clim. Change
81
,
53
70
.
Dobler
C.
Stötter
J.
Schöberl
F.
2010
Assessment of climate change impacts on the hydrology of the Lech Valley in northern Alps
.
J. Water Clim. Change
1
(
3
),
207
218
.
European Environment Agency
2009
Regional climate change and adaptation. The Alps facing the challenge of changing water resources. 8/2009. EEA Report
,
Copenhagen, Denmark
.
Fliri
F.
1975
Das Klima der Alpen im Raume von Tirol Monographien zur Landeskunde Tirols
.
Universitatsverlag Wagner
,
Innsbruck, Austria
.
Föhn
P.
1990
Schnee und Lawinen
. In:
Schnee, Eis und Wasser der Alpen in einer wärmeren Atmosphäre
(
Vishcher
D. D.
, ed.),
Mitteilungen 108
.
Versuchsanstalt für Wasserbau, Hydrologie und Glaziologie
,
Zürich
, pp.
33
48
.
Gagnon
A. S.
Gough
W. A.
2002
Hydroclimatic trends in the Hudson bay region, Canada
.
Can. Water Resour. J.
27
,
245
262
.
Groisman
P. Ya
Karl
T. R.
Knight
R. W.
Stenchikov
G. L.
1994
Changes of snow cover, temperature, and radiative heat-balance over the Northern Hemisphere
.
J. Clim.
7
,
1633
1656
.
Hall
A.
Qu
X.
Neelin
J. D.
2008
Improving predictions of summer climate change in the United States
.
Geophys. Res. Lett.
35
,
L01702
.
Hensel
D. R.
Hirsch
R. M.
1992
Statistical Methods in Water Resources
.
Elsevier Science
,
Amsterdam
.
Hess
A.
Iyer
H.
Malm
W.
2001
Linear trend analysis: a comparison of methods
.
Atmos. Environ.
35
,
5211
5222
.
Keiler
M.
Knight
J.
Harrison
S.
2010
Climate change and geomorphological hazards in the eastern European Alps
.
Phil. Trans. R. Soc. A
368
,
2461
2479
.
Kendall
M. G.
1975
Rank Correlation Methods
.
Charles Griffin
,
London
.
Knowles
N.
Dettinger
M. D.
Cayan
D. R.
2006
Trends in snowfall versus rainfall in the Western United States
.
J. Clim.
19
,
4545
4559
.
Kundzewicz
Z. W.
2004
Searching for changes in hydrological data
.
Hydrol. Sci. J.
49
(
1
),
3
6
.
Laternser
M.
Schneebeli
M.
2003
Long-term snow climate trends of the Swiss Alps (1931–99)
.
Int. J. Climatol.
23
,
733
750
.
Leonhardt
G.
Olefs
M.
Neubarth
J.
Thieken
A.
Schönlaub
H.
Schöberl
F.
Kuhn
M.
2009
Auswirkungen einer möglichen Klimaänderung auf ein alpines Speicherkraftwerk
.
Dresdener Wasserbauliche Mitteilungen
, p.
39
.
Magnusson
J.
Jonas
T.
Lopéz-Moreno
I.
Lehning
M.
2010
Snow cover response to climate change in a high alpine and half-glacierized basin in Switzerland
.
Hydrol. Res.
41
(
3–4
),
230
240
.
Mann
H. B.
1945
Non-parametric test against trend
.
Econometrika
13
,
245
259
.
Marty
C.
2008
Regime shift of snow days in Switzerland
.
Geophys. Res. Lett.
35
,
L12501
.
Mote
P. W.
Hamlet
A. F.
Clark
M. P.
Lettenmaier
D. P.
2005
Declining mountain snowpack in western north America
.
Bull. Am. Meteorol. Soc.
86
,
39
49
.
Nemec
J.
Gruber
C.
Chimani
B.
Auer
I.
2013
Trends in extreme temperature indices in Austria based on a new homogenised dataset of daily minimum and maximum temperature series
.
Int. J. Climatol.
33
,
1538
1550
.
Parry
M. L.
Canziani
O. F.
Palutikof
J. P.
van der Linden
P. J.
Hanson
C. E.
(eds)
2007
Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change
.
Cambridge University Press
,
Cambridge
().
Pekarova
P.
Miklanek
P.
Pekar
J.
2006
Long-term trends and runoff fluctuations of European rivers
. In:
Climate Variability and Change: Hydrological Impacts
.
IAHS
,
UK
, pp.
520
525
.
Radziejewski
M.
Kundzewicz
Z. W.
2004
Detectability of changes in hydrological records
.
Hydrol. Sci. J.
49
,
39
51
.
Renard
B.
Lang
M.
Bois
P.
Dupeyrat
A.
Mestre
O.
Niel
H.
Sauquet
E.
Prudhomme
C.
Parey
S.
Paquet
E.
Neppel
L.
Gailhard
J.
2008
Regional methods for trend detection: assessing field significance and regional consistency
.
Water Resour. Res.
44
,
W08419
.
Schimon
W.
Schöner
W.
Böhm
R.
Haslinger
K.
Blöschl
G.
Merz
R.
Blaschke
A. P.
Viglione
A.
Parajka
J.
Kroiß
H.
Kreuzinger
N.
Hörhan
T.
2011
Anpassungsstrategien an den Klimawandel für Österreichs Wasserwirtschaft
.
Bundesministerium für Land- und Forstwirtschaft, Umwelt und Wasserwirtschaft
,
Vienna, Austria
.
Schönwiese
C.-D.
Janoschitz
R.
2005
Klima-Trendatlas Deutschland 1901–2000
.
Berichte des Instituts für Atmosphäre und Umwelt der Universität Frankfurt/Main
,
Frankfurt, Germany
.
Stahl
K.
Hisdal
H.
Hannaford
J.
Tallaksen
L. M.
van Lanen
H. A. J.
Sauquet
E.
Demuth
S.
Fendekova
M.
Jódar
J.
2010
Streamflow trends in Europe: evidence from a dataset of near-natural catchments
.
Hydrol. Earth Syst. Sci.
14
,
2367
2382
.
Stewart
I. T.
Cayan
D. R.
Dettinger
M. D.
2005
Changes toward earlier streamflow timing across western North America
.
J. Clim.
18
,
1136
1155
.
Viviroli
D.
Archer
D. R.
Buytaert
W.
Fowler
H. J.
Greenwood
G. B.
Hamlet
A. F.
Huang
Y.
Koboltschnig
G.
Litaor
I.
López-Moreno
J. I.
Lorentz
S.
Schädler
B.
Schreier
H.
Schwaiger
K.
Vuille
M.
Woods
R.
2011
Climate change and mountain water resources: overview and recommendations for research, management and policy
.
Hydrol. Earth Syst. Sci.
15
,
471
504
.
Yue
S.
Pilon
P.
Phinney
B.
Cavadias
G.
2002
The influence of autocorrelation on the ability to detect trend in hydrological series
.
Hydrol. Processes
16
,
1807
1829
.
Zhang
X.
Harvey
K. D.
Hogg
W. D.
Yuzyk
T. R.
2001
Trends in Canadian streamflow
.
Water Resour. Res.
37
(
4
),
987
998
.
Zhang
Z.
Dehoff
A. D.
Pody
R. D.
Balay
J. W.
2010
Detection of streamflow change in the Susquehanna River Basin
.
Water Resour. Manag.
24
,
1947
1964
.