Has streamflow changed in the Nordic countries? – Recent trends and comparisons to hydrological projections

Journal of Hydrology 394 (2010) 334–346

Contents lists available at ScienceDirect

Journal of Hydrology journal homepage: www.elsevier.com/locate/jhydrol

Has streamflow changed in the Nordic countries? – Recent trends and comparisons to hydrological projections Donna Wilson ⇑, Hege Hisdal, Deborah Lawrence Norwegian Water Resources and Energy Directorate (NVE), P.O. Box 5091, Majorstua, N-0301 Oslo, Norway

a r t i c l e

i n f o

Article history: Received 18 February 2010 Received in revised form 6 September 2010 Accepted 14 September 2010 This manuscript was handled by A. Bardossy, Editor in Chief, with the assistance of Haberlandt, Associate Editor Keywords: Streamflow trends Nordic region Climate change Flood Temporal autocorrelation Field significance

s u m m a r y A pan-Nordic dataset of 151 streamflow records was analysed to detect spatial and temporal changes in streamflow. Prior to undertaking analyses, all streamflow records with significant levels of autocorrelation were pre-whitened to remove the adverse effect of temporal autocorrelation on the test results. The Mann–Kendall trend test was applied to study changes in annual and seasonal streamflow as well as floods and droughts for three periods: 1920–2005, 1941–2005 and 1961–2000. Field significance was evaluated to determine the percentage of stations that are expected to show a trend due to the effect of cross-correlation. The period analysed and the selection of stations influenced the regional patterns found, but the overall picture was that trends of increased streamflow dominate annual values and the winter and spring seasons. Trends identified in summer flows differed between the three periods analysed, whereas no trend was found for the autumn season. In all three periods, a signal towards earlier snowmelt floods was clear, as was the tendency towards more severe summer droughts in southern and eastern Norway. These trends in streamflow result from changes in both temperature and precipitation, but the temperature induced signal is stronger than precipitation influences. This is evident because the observed trends in winter and spring, where snowmelt is the dominant process, are greater than the annual trends. A qualitative comparison of the findings with available streamflow projections for the region showed that the strongest trends found are generally consistent with future changes expected in the projection periods, for example increased winter discharge and earlier snowmelt floods. However, there are predicted changes that are not reflected in past trends, such as the expected increase in autumn discharge in Norway. Hence, the changes expected because of increased temperatures are reflected in the observed trends, whereas changes anticipated due to increases in precipitation are not. Ó 2010 Elsevier B.V. All rights reserved.

1. Introduction Climate change may cause changes in annual average streamflow, in the seasonal distribution of flow or in the magnitude and frequency of floods and droughts which could have extensive impacts on water management, agriculture and aquatic ecosystems. Countries such as Norway, Sweden, Iceland and Finland where much of their electricity production system is based on hydropower are especially sensitive to long-term variations in streamflow. Hydrological changes may offer new opportunities for hydropower development, but could require new investment, design, operation and management practices at new and existing hydroplants to take advantage of these opportunities. Conversely, increases in the magnitude of peak flows can have implications for flood management and dam safety.

⇑ Corresponding author. Tel.: +47 22 95 92 03; fax: +47 22 95 92 16. E-mail address: [email protected] (D. Wilson). 0022-1694/$ - see front matter Ó 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jhydrol.2010.09.010

This paper seeks to detect climate change signals in historical streamflow records and compares the results to published hydrological projections. Studies of climate change traditionally include both the development of possible future projections (as outlined below) and the analysis of possible climate change signals in historical data. Studies based on historical data evaluate the presence of trends or jumps over time relative to natural climate and hydrological variability. Changes might be detected in annual as well as seasonal or extreme values. In some cases, annual averages may remain unchanged although a change in seasonal or extreme values might be present. This paper focuses on detecting climate change signals in historical streamflow records, addressing the question of whether climate change is already impacting hydrology. The Mann–Kendall test, which is widely used in trend analyses (e.g. Mitosek, 1995; Hisdal et al., 2001) is used to detect trends in the streamflow records, after accounting for the adverse effect of spatial and temporal autocorrelation within the datasets. To evaluate if observed trends in streamflow are consistent with scenario projections, a qualitative comparison between the observed trends and published hydrologic projections for streamflow is made.

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

Climate projections for Europe and the Nordic countries have been developed as part of large international research activities, e.g. the PRUDENCE (Prediction of Regional scenarios and Uncertainties for Defining European Climate change risks and Effects, http://prudence.dmi.dk/; Christensen et al., 2007a), ENSEMBLES (http://ensembles-eu.metoffice.com/) and national regional climate modelling projects. PRUDENCE was a 3 years initiative (2001–2004) which produced a series of high-resolution climate change scenarios for the period 2071–2100 for Europe, and characterised the variability and level of confidence in these scenarios. Following this, ENSEMBLES ran for 5 years (2004–2009) and further developed and assessed a range of future climate projections for Europe to determine which are more probable than others (van der Linden, 2009). Parallel to these projects, regional climate modelling has been undertaken in numerous coordinated national projects (e.g. SWECLIM in Sweden, RegClim in Norway, FIGARE in Finland). The Regional Climate Modelling Programmes undertaken by individual countries developed regional climate change projections for the Nordic region (e.g. Rummukainen et al., 2004a). Several studies, often undertaken as part of the regional climate modelling projects, have simulated the potential impacts of climate change on hydrological processes for a range of basins in the Nordic region (e.g. Andréasson et al., 2004; Roald et al., 2006; Beldring et al., 2008). Reports of observed changes and the likely future impacts of climate change for individual Nordic countries are also available (e.g. Jørgensen et al., 2001; Rummukainen et al., 2004b; Tuomenvirta, 2004; Hanssen-Bauer et al., 2009). Information about observed and anticipated changes in the climate system are summarised in the Fourth Assessment Reports of the IPCC (Solomon et al., 2007; Parry et al., 2007). In Europe, temperatures increased throughout the 20th Century, with the greatest rates of change identified for the period 1979–2005 (Alcamo et al., 2007). Temperatures are also increasing at a faster rate in winter than summer (Jones and Moberg, 2003) and snow cover has decreased, particularly in the spring (Trenberth et al., 2007; Bates et al., 2008). Trends in precipitation across Europe are found to be more spatially variable, but the greatest increases have been observed in Northern Europe (Klein Tank et al., 2002; Trenberth et al., 2007). Recent studies of the climate in Finland (Hyvärinen, 2003) Sweden (Lindström and Alexandersson, 2004) and Norway (Hanssen-Bauer et al., 2009) found that precipitation in these countries increased during the last 100 years, particularly since the end of the 1970s. In Trenberth et al. (2007) observed changes regarding hydrology are summarised, but it is concluded that substantial uncertainty remains about trends in hydrological variables because the detected changes vary considerably over space and due to limitations with data quality and availability. Individual national studies for the Nordic countries (for example Hyvärinen, 1998; Førland et al., 2000; Ovesen et al., 2000; Lindström and Bergström, 2004; Hanna et al., 2004; Jónsdóttir et al., 2008; Hisdal et al., 2010) also show that trends in climatic and hydrological variables, either natural or human induced, vary considerably between regions. An overview of studies of long time series of precipitation, temperature and streamflow in the Nordic countries up to 2003 can be found in Hisdal et al. (2003). The national studies vary both regarding the period and variables analysed. To make a qualitative estimate of regional differences in the changes found, it is possible to compare the national studies. However, the results would be uncertain due to the various time periods analysed and the different methods used. To study the regional distribution of changes, there is a need to study data from several countries and to include a common time period and methodology. The latter is important because the trend or changes found will be strongly influenced by the time period studied. A previous Nordic study on regional differences in trends focused on annual and seasonal streamflow, and on the time period 1930–1980

335

(Hisdal et al., 1995). Over the last 15 years, streamflow records have been updated, and a similar study is possible for a longer period. Hisdal et al. (2010) recently investigated changes in annual, seasonal, floods and droughts in the Nordic countries up to 2002. In this paper the analyses undertaken by Hisdal et al. (2010) are further refined by considering the effect of spatial and temporal autocorrelation, investigating trend magnitude instead of statistical significance and updating the period of analysis to 2005. Enhanced climate change may result in more pronounced and more persistent changes in streamflow, so that the likelihood of detecting changes is expected to increase over time (Kundzewicz, 2004). This is an argument for continuing to examine updated streamflow records, including at the pan-Nordic scale, to identify larger scale regional differences in streamflow in terms of non-stationarity and climate variability. Furthermore, the majority of the studies detailed above exist only in national archives, which can be difficult to obtain and use and is a motivation for making these results in this paper more widely available. Most trend studies look at annual and seasonal values (e.g. Hyvärinen, 2003; Jónsdóttir et al., 2008). Less attention is given to extremes, especially droughts. Amongst the reasons are that extremes are especially prone to man-made environmental changes and also more vulnerable to measurement errors (Boiten, 2000). Detected trends are therefore more difficult to relate to changes in climate. In the latter half of the 20th century an increase in the frequency of heavy precipitation events has been observed in Northern Europe (Trenberth et al., 2007). This increase has been greater than the increase in mean annual precipitation (Groisman et al., 2005). Grønås et al. (2005) observed this tendency in the western part of Norway and state that this agrees with expected changes from human induced climate change. Although, even if changes in annual streamflow usually follow changes in annual precipitation, the change in floods does not relate that well to the increase in heavy precipitation. Recent studies of trends in annual maximum (Kundzewicz et al., 2005) and peak over threshold floods (Svensson et al., 2005) worldwide, and annual maximum floods in Europe (Kundzewicz, 2005) conclude that although more intense precipitation has been documented, a coherent and general increase in high river flows could not be detected. In these previous studies, streamflow records from Norway, Finland and Sweden covered different time periods, whereas, a common time period is analysed in the present study to enable a spatial comparison of trends. Annual maximum autumn and spring floods are also studied separately to distinguish between rain and snowmelt floods. In Hisdal et al. (2001) a pan–European study of trends in the severity and frequency of summer drought was carried out. For the period 1962–1990, which included a relatively high density of stations in Norway and Denmark, only one station within these two countries had a significant trend. However, a tendency towards more severe droughts was seen in southern and northern Norway and Denmark and less severe droughts in central Norway. The other Nordic countries had too few stations to draw any conclusions. Trends in the low flow indices 7-day annual minimum and 30-day annual minimum based on a global dataset were studied by Svensson et al. (2005). Only increases in low flows were found in Europe, but this was interpreted as resulting from a large number of reservoirs becoming operational during the time periods evaluated. In the present study, trends in droughts are only investigated for stations where reservoirs have insignificant influence. The Nordic countries have a good spatial coverage of streamflow records with a high level of quality control. Pristine basins with a minimum of human interventions such as urbanization, deforestation and changes in storage capacity are available for studies of extremes. Long-term data series enable documentation

336

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

of changes at inter-annual time scales and testing for trends in common time intervals across the dataset. Such data provides a unique opportunity to investigate changes in streamflow in the Nordic region. This paper aims to address the question ‘has streamflow changed in the Nordic countries?’ by means of an analysis of the temporal and spatial variations in annual and seasonal streamflow, floods and droughts. Results are compared to hydrological projections to investigate how observed trends relate to projected future changes. The next section describes data selection and the streamflow variables analysed, followed by details of the methods applied. Results are then presented and discussed with respect to the time period and region analysed. The detected trends are then compared to changes expected in the future, and final conclusions are proposed.

2. Data Daily data from 151 streamflow discharge gauging stations in Denmark, Finland, Iceland, Norway and Sweden were compiled. These data are stored in a common database, a Nordic version of the European Water Archive (EWA) of the Flow Regimes from International Experimental and Network Data (FRIEND) Project (Roald et al., 1993; Rees and Demuth, 2000). For information regarding data availability see the following link: http://ne-friend. bafg.de/servlet/is/7413/. Data were selected to cover the whole Nordic region with a common time period, from 1920 to 2005. The criteria for selecting series were that the records should be, as far as possible, unaffected by human induced changes in the basin, and that the records should be continuous and as long as possible. Even if most catchments in the Nordic region are only subject to minor land use changes and a minimum of water abstractions for irrigation, industry and public water supply, long pristine series are hard to find. The longest series are often affected by human activities in the basin, causing various forms of inhomogeneity in the series. The series were therefore classified into three categories: (i) series suitable for the analysis of annual values; (ii) series suitable for (i) and the analysis of monthly values, i.e. seasonal trends; (iii) series also suitable for (ii) and the analysis of daily values, i.e. trends in extremes. Only natural catchments were deemed suitable for the analysis of daily values. In addition, catchments were included for analysis of seasonal values if minor regulations in the catchment were not thought to influence mean seasonal flows or if flows have been naturalised to account for such regulations. For the analysis of annual values, catchments were also included if they contained small dams which were used only to store water within a single year. To obtain a good spatial resolution a reasonable and common time period should be selected. However, time series, as long as possible, are important to study long-term variability. A best possible Nordic coverage required a relatively short period to be selected (1961–2000). This period encompasses the total dataset (151 stations). Two additional sets of stations, 1941–2005 (111 stations) and 1920–2005 (68 stations) were chosen to investigate longer-term trends. The stations used for each of these three periods of analysis have data available for the full periods of analysis, with no periods of missing data. However, analyses of trends in seasonal streamflow and extremes further reduce the number of stations due to the data quality classification detailed above. A minimum number of stations, 42, were identified for analysis of data at a daily resolution for the period 1920–2005. The spatial

coverage of data is not uniform as a larger number of long records from pristine basins are available in Norway and Denmark compared to Sweden, Finland and Iceland. This is considered further in the methods and discussion sections. However, the dataset comprises a high-quality, long-term set of homogeneous series of adequate spatial resolution with a minimum of human influence for detecting trends caused by climate change, natural or human induced. Trend studies were performed for the following 11 hydrological variables:  mean annual streamflow (calendar years);  mean seasonal values (winter: December–February; spring: March–May; summer: June–August; autumn: September– November);  magnitude and timing (date) of the spring (1 March–15 July; see below) flood peak;  magnitude and timing of the autumn (16 July–11 November; see below) flood peak;  drought duration and deficit volume. Nordic river flow regimes vary considerably. In Denmark, along the western coast of Norway and much of southern Sweden there is no regular snow cover during the winter. In these areas flood events most commonly occur in the autumn, while the lowest flows occur during the summer months caused by lack of precipitation and high evaporation losses. In the north, the east and the mountainous regions, flood events tend to be induced by snowmelt and occur in the spring, while the lowest flows occur in winter due to precipitation being stored as snow. There are also transition regions where the annual maximum flood and the lowest flows can be found in any season. In the Nordic countries the dominant flood mechanism differs between autumn and spring. Flood events tend to be thermally driven (i.e. snowmelt) in the spring, whereas precipitation becomes the dominant driver in the autumn. A rough differentiation of the flood generation mechanism was achieved by considering spring and autumn floods separately. The spring flood period was defined to be 1 March–15 July, and the autumn flood period was 16 July–11 November. Timing was defined as the date of the flood peak. In Denmark, parts of southern Sweden and south-western Iceland the major floods usually occur during the winter season. For this region it could therefore be relevant to study floods for a season lasting from late autumn to early spring, but this study is limited to the two periods defined above. To perform a consistent analysis of droughts, it is necessary to distinguish between ‘summer droughts’ that are driven by a lack of precipitation and high evaporation and ‘winter droughts’ caused by precipitation being stored as snow. These two drought types are affected in different ways by climate change. Here only ‘summer droughts’ are considered. Most of the basins investigated in this study have a regular spring flood, with the exception of Denmark, southern Sweden and the groundwater-fed rivers in south-western Iceland. Plots of maximum, median and minimum values of streamflow were used to identify the average timing of the spring flood. The start of the summer season was defined to be at the end of the spring flood. The end of the summer season was defined to be the date when the average temperature drops below 0 °C. This method has previously been applied to define summer drought seasons for many basins across Europe and is further discussed in Hisdal et al. (2001). Finally, three seasons were identified and the stations were classified according to these. Season A (all year/no season) includes Denmark, the four western-most basins in Norway and the two south-western basins in Sweden. Season B (15 April–15 November) is applied to the coastal zone in Norway, two stations in southern Sweden and the three southern-most

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

stations in Finland. Season C (15 June–15 October) is used for the northern and inland catchments of Norway, Sweden and Finland and the Icelandic stations. River flow is considered to indicate drought conditions when the flow is below a specific threshold, Qthres. This method proposed by Yevjevich (1967) was originally based on the statistical theory of runs for analysing sequential time series with a time resolution of 1 month or longer. The approach has previously been used to select droughts from a daily hydrograph for trend studies (Hisdal et al., 2001). The method allows streamflow droughts to be characterised in terms of their duration, di, deficit volume (severity), vi, minimum flow, Qmin and time of occurrence (Fig. 1). During a prolonged dry period it is often observed that flow exceeds the threshold level for a short period of time, thereby dividing a large drought into a number of minor droughts that are mutually dependent. To, as far as possible, pool dependent droughts and remove minor droughts, an 11-day moving average procedure was adopted, as recommended by Tallaksen et al. (1997). The threshold level, Qthres, was determined for each station by using the 30th percentile (Q70), from the flow duration curve (i.e. the flow which is exceeded 70% of the time in the given period and season). A suitable drought threshold may be chosen in a number of ways. The choice is amongst others a function of the hydrological regime under study and the type of analysis to be carried out. Hisdal et al. (2002, 2004) discuss this and conclude that the thresholds in the range between the 10th and 30th percentile are reasonable choices for perennial rivers. When undertaking trend analysis it is an advantage to have one drought every year. Choosing the 30th percentile ensures this. Finally, the drought events were extracted for the three time periods for the available stations. Two drought characteristics were analysed:  annual maximum drought duration (days);  annual maximum deficit volume (m3). We chose not to analyse low flow data (i.e. annual daily minimum flows) because they are especially prone to measurement errors caused by factors such as altered cross-sections, weed growth and backwater effects. This might influence a trend study of daily minimum series, but drought as defined here, is more robust in this respect. 3. Methods

2002; Yue and Wang, 2002). High autocorrelation can also affect the estimate of trend magnitude (Zhang and Zwiers, 2004). Yue et al. (2003) state that positive serial correlation increases the possibility that the Mann–Kendall test rejects a hypothesis of no trend, when there is actually no trend. A basic assumption of the Mann– Kendall test is, therefore, that the data being analysed are a collection of random independent variables. Time series of hydrological variables may exhibit serial correlation, because of storage properties in the basin and frequently require pre-whitening to remove autocorrelation prior to trend analysis. The consistency of trends detected across a region is often studied by plotting results of at-site tests on a map. However, if data are spatially correlated and a trend is found at one site, it is likely that trends will also be found at other nearby sites. As the spatial coverage of stations is seldom uniform, a trend in a region with a relatively high density of stations will often result in a larger total percentage of trends being identified, compared to a trend being identified in a region with fewer stations. This non-uniformity can bias an assessment of regional trends. As for temporal autocorrelation, ignoring spatial correlation in a dataset can also result in a null hypothesis of no trend being rejected more frequently than it should be when considering overall trends across a region (Douglas et al., 2000; Yue et al., 2003; Burn and Hag Elnur, 2002; Renard et al., 2008). Pre-whitening, trend test and field significance procedures were applied in this work and are detailed in the following sub-sections. 3.1. Pre-whitening Autocorrelation coefficients were calculated for each of the 11 hydrological variables, at each station, for each of the three time periods. The hydrological variables at many stations have significant partial correlations (at the 5% level) at a time lag of 1 year, with data for some stations also possessing significant partial correlations over longer time lags. In particular, three Icelandic (Dynjandi- Brúará, Kljáfoss and Skálabrekka) and one Danish station (Lindenborg Bro) have high autocorrelation of annual values, possibly due to the buffering effect of groundwater aquifers. One Swedish catchment (Kallio) characterised by large lakes shows high autocorrelation of winter flows. Two Danish stations (Lindenborg Bro and Lindebjerg) have high autocorrelation of drought duration and drought volume likely to be due to the regulating effect of groundwater aquifers. The autocorrelations amounted to as much as 0.76, which suggests that they may not only influence significance levels, but also the estimate of trend magnitude (see Fig. 2 of Zhang and Zwiers (2004)). Pre-whitening of all 11 hydrological variables was undertaken to remove short-term persistence

3

Q (m /s)

Temporal autocorrelation within a time series and crosscorrelation between sites in a dataset can both affect the ability of a trend test to assess trend significance (Burn and Hag Elnur,

337

Q0

Time (days) Fig. 1. Definition of drought characteristics. (This article was published in Developments in Water Science, 48, Hisdal et al., Hydrological drought characteristics, Fig. 5.7. Copyright Elsevier, 2004.) The value of Q0 is set to Q70 in the present study.

338

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

following the iterative procedure described in Appendix A of Wang and Swail (2001). Pre-whitening can remove a proportion of the trend from a dataset, but this iterative procedure minimises this problem so that the pre-whitened time series possesses the same trend as the original series (Wang and Swail, 2001; Zhang and Zwiers, 2004; Liu et al., 2009). Following the pre-whitening procedure, partial correlations were recalculated (for lag = 1, 2, . . . , 20). This showed a notable reduction in the autocorrelation of each time series, with few partial correlations remaining significant at the 5% level. No further attempt was made to remove the remaining low levels of autocorrelation in the datasets. Birsan et al. (2005) state that where autocorrelation coefficients are low, the differences between the results of a trend test prior to and after pre-whitening are small. The Mann–Kendall test (Section 3.3) was performed on the pre-whitened time series. 3.2. Trend test According to Yue and Pilon (2004), even if the underlying distribution is known to be approximately normal, the power of a parametric test is only slightly higher than the power of a nonparametric alternative. For non-normal series, rank-based tests, including the Mann–Kendall test, are found to have an increased ability to detect trends compared to slope-based tests. The distribution of the streamflow variables being analysed will differ across space and can be highly skewed. This is particularly the case with floods and droughts which are typically characterised by many minor and few major events. Hence, a non-parametric test is favourable. In Kundzewicz and Robson (2004) resampling methods are pointed out as particularly suitable for hydrological data. Hisdal et al. (2001) applied two tests, a resampling test and the widely used non-parametric Mann–Kendall test, to investigate trends in streamflow droughts in Europe. A strong agreement between the two tests was found. The Mann–Kendall test searches for a trend in a time series without specifying whether the trend is linear or nonlinear, but does assume that the trend is monotonic. An example of an application of the test can be found in Mitosek (1995), who compared different trend tests to detect signals of climate variability and change in monthly and annual discharges of 176 series from all over the world. Further references of application to hydroclimatological time series can be found in Yue and Pilon (2004). A description of the test can be found in Salas (1993). The Mann– Kendall S statistic is given by (Burn and Hag Elnur, 2002):

S¼

n1 X n X

SgnðX j  X i Þ

ð1Þ

i¼1 j¼i1

where Xi and Xj are the sequential data values, n is the dataset record length, and

8 > < þ1 Sgn ðhÞ ¼ 0 if > : 1

h>0 h¼0 h<0

3.3. Significance testing In the present study, trends are discussed in terms of their trend magnitude and direction. Critical values of the Mann–Kendall S statistic (Smax; Table 1) are used to identify strong and weak trends. The Smax values are defined using p-values (two-sided) of 0.05 and 0.3, which indicate the magnitude of the trend is likely to be in the upper or lower 2.5% and 15%, respectively, of the statistical distribution. Because the data series within each study period are all of the same length, and pre-whitening has been undertaken to preserve the original trend magnitude, analysing trends in terms of either magnitude or significance levels should nominally be roughly equivalent. However, this paper focuses on trend magnitude because long-term persistence (as opposed to short-term persistence removed by pre-whitening) of many hydroclimatic time series can cause trend test significance to be highly sensitive to the trend test used (Cohn and Lins, 2005). In particular, Cohn and Lins found that statistical significance is difficult to assess because it depends on assumptions being made about the underlying stochastic process. They illustrate using the Northern Hemisphere temperature record that trend significance can vary by up to 25 orders of magnitude depending on the trend test used. Statistical significance is often cited to bolster scientific argument, but as Cohn and Lins, and more recently Clarke (2010) acknowledge, the concept of statistical significance is meaningless when discussing poorly understood systems and simply reporting the identification of significant trends does little to advance hydrological understanding. 3.4. Field significance testing Field significance testing determines the percentage of stations expected to show a trend, for critical values of the Mann–Kendall S statistic (see Section 3.3), purely by chance, due to the effect of cross-correlation. Renard et al. (2008) compared several methods for assessing field significance using synthetic datasets and found a bootstrap method to be an adequate and robust tool. A bootstrap procedure described by Burn and Hag Elnur (2002) was used to determine the percentage of stations expected to show a trend due to the effect of cross-correlation. Where the number of observed trends is greater than the number of trends expected, the results can be considered field significant. In this paper, 2000 bootstrap samples were selected to establish the percentage of stations expected to show a trend, which is the upper limit of the number recommended by Davison and Hinkley (1997). 4. Results and discussion The results are assessed in two different ways: summary plots of the stations exhibiting positive and negative trends, and maps of the spatial variability of the trends (all periods). The results are discussed in terms of strong and weak trends as defined in Section 3.3. Because of the quality classification, the results are as far as possible not influenced by human interference in the

Table 1 Critical values of the Mann–Kendall statistic for the identification of strong and weak trends.

a

Period of analysis

Number of annual data values in each station time seriesa

Range of S

Smax (strong trends)

Smax (weak trends)

1920–2005 1941–2005 1961–2000

85 64 39

3750 to 3750 2016 to 2016 741 to 741

6518 or P518 6339 or P339 6163 or P163

6274 or P274 6180 or P180 687 or P87

The number of annual data values in each series is equal to the number of years minus one, as each time series has been pre-whitened.

339

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

30

Number of stations (%)

1920-2005

1941-2005

1961-2000

25

20

15

10

5

Strong positive trend

autumn

summer

spring

winter

annual

autumn

summer

spring

winter

annual

autumn

summer

spring

winter

annual

0

Strong negative trend

Fig. 2. Summary of strong trends for annual and seasonal flow in all time intervals. Positive trends indicate a tendency towards increased flows, whereas negative trends indicate a tendency towards reduced flows.

catchment. Hence, the detected changes are caused either by natural climate variability or by climate change. 4.1. Summary statistics The percentages of stations with strong positive and negative trends are summarised in Fig. 2 (annual and seasonal flow) and Fig. 3 (extremes). All results are field significant, i.e. for both the strong and weak trends the percentage of stations showing a trend is greater than that expected by chance, with the following exceptions:  Autumn flows (1941–2005 strong and weak trends, 1961–2000 weak trends).  Timing of the autumn flood (1961–2000 weak trends).  Drought duration (1961–2000 weak trends).  Drought volume (1961–2000 weak trends). Three issues should be kept in mind when discussing the results; (i) only few records are available for the period 1920–2005, (ii) in the trend analysis observations in the first and last few years are more important than the period of analysis (Hisdal et al., 2001), and (iii) although field significance has been calculated, this procedure only identifies whether results are field significant at the Nordic scale and does not identify which stations are spatially correlated or remove spatial correlation from the datasets. A further study would be required to conclude whether stations at the regional scale are correlated, but this is beyond the scope of this paper. More than 6% of stations show strong trends in all annual and seasonal variables for all periods, except for autumn flow (Fig. 2). For the extremes, fewer strong trends are evident, particularly during the 1961–2000 period, but the trend in the magnitude of the spring and winter peaks is clearest with more than 7% of stations showing a trend in all three periods (Fig. 3). In general a larger proportion of trends are found in annual and seasonal values than the extremes. Overall positive trends, towards increasing streamflow, dominate annual and seasonal flows. Regarding annual discharge, a positive trend is found for about 17% of the stations in the period 1961–2000. For the two longer periods, 1920–2005 and 1941– 2005, this number is reduced to between 7% and 8%. The winter and spring seasons are both dominated by positive trends in all

periods. The clearest trends are identified for winter flows, with more than 12% of stations showing a strong trend in each period. For the period 1920–2005, over 26% of stations show strong trends in spring flows, but for the two shorter periods fewer strong trends are identified (7–9%). For the summer season strong trends are identified at between 6% and 14% of stations, but the positive dominance in the period 1961–2000 is replaced by a negative dominance in the two longer periods, 1920–2005 and 1941–2005. Less than 4% of the stations show a strong trend for the autumn season. Concerning floods, the picture is less clear (Fig. 3). For flood peaks (spring and autumn), roughly equal numbers of positive and negative trends are identified for the two shorter periods, 1941–2005 and 1961–2000. For the 1920–2005 period, a trend towards decreasing streamflow dominates. For the longer two periods, 1920–2005 and 1941–2005, there is a clear dominance of negative trends for the timing of the spring flood, indicating a trend towards earlier spring floods. For the shorter 1961–2000 period only 2% of stations show a strong trend in the timing of the spring flood. However, regardless of the period analysed, there is a clear dominance of positive trends for the timing of the autumn flood, meaning a trend towards later autumn floods. For droughts, trends towards increasing drought durations and volumes dominate all periods, indicating a tendency towards more severe droughts, although only between 3% and 7% of stations show a strong trend in any of the periods.

4.2. Trends in annual discharge Trends in annual streamflow are presented in Fig. 4. Strong trends are indicated with large blue circles (strong positive trend) and large red circles (strong negative trend). Weak trends are indicated with smaller circles, light blue if positive, orange if negative. Green circles indicate no trend was found. Positive trends are identified for stations in south-western Norway in all three periods. A positive trend is identified for a few stations in northern Sweden in the period 1921–2005, but a greater number of positive trends are identified for this region in 1941–2005 and 1961–2005, with strong trends for a large number of stations in the later period. Positive trends are also identified for southern Sweden for the period 1961–2000, and western Finland

340

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

Number of stations (%)

25

1920-2005

1941-2005

1961-2000

20

15

10

5

Strong positive trend

drought volume

drought duration

spring timing

spring peak

autumn timing

autumn peak

drought volume

drought duration

spring timing

spring peak

autumn timing

autumn peak

drought volume

drought duration

spring peak

spring timing

autumn timing

autumn peak

0

Strong negative trend

Fig. 3. Summary of strong trends for flood and drought in all time intervals. Positive trends indicate a tendency towards more severe floods and droughts, whereas negative trends indicate a tendency towards less severe floods and droughts.

Fig. 4. Trends in annual streamflow for the periods 1920–2005 (left), 1941–2005 (middle) and 1961–2000 (right).

in the periods 1941–2005 and 1961–2000. In Denmark, positive trends are identified at a limited number of stations in all periods, but most stations show no trend in the periods 1941–2005 and 1961–2000. Generally, in a zone from central Norway through eastern Norway and into central Sweden there are no trends. A limited number of stations, especially in southern Norway and southern Sweden have a negative trend in the period 1920–2005. Almost all stations in Iceland show no trend in streamflow during the period 1961–2000. These results are in agreement with the findings of Jónsdóttir et al. (2005, 2008). Weak trends in Iceland during the period 1941–2005 are apparent but only two stations are of that length. A station to the north (Reykjafoss) has a weak positive trend, while a station to the south-west (Ellidaarstøø) has a weak negative trend. By examining individual series (non-pre-whitened), temporal characteristics of annual streamflow can be identified. There were several relatively dry years from the late 1960s to around 1980 at

many stations (e.g. Lindström and Bergström, 2004; HanssenBauer et al., 2009). The early 1940s were also dry across much of the Nordic region, but with some wet years in central Norway (Hanssen-Bauer et al., 1995). This explains why central Norway is without trends, and why many series have strong positive trends in the two shorter periods. The 1920s were cool, but with abundant precipitation in many regions (e.g. Lindström and Bergström, 2004; Førland et al., 2000), which helps explain why most series are without any trend in the 1920–2005 period. Precipitation in western Norway has increased by 20% over the last 100 years, with the increase being most marked since the end of the 1970s (Hanssen-Bauer et al., 2009). Glacier streams in western Norway peaked in the warm 1930s and have had increasing discharge since the mid 1960s because of rising rainfall and temperatures, the later of which has increased melt-induced streamflow (Førland et al., 2000; Stranden and Skaugen, 2009). Precipitation in northern Sweden has also increased, particularly from the late 1940s (Lindström and Alexandersson, 2004), and explains why a positive

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

trend is apparent at so many Swedish stations in the two shorter periods. In Denmark, the positive trends identified at a limited number of stations correspond to a general increase in precipitation throughout the study period (Ovesen et al., 2000). In Finland, a strong positive trend is found at only one station (Niskakosi) for the period 1941–2005. This is partially caused by the occurrence of a sequence of two very dry years in the early 1940s. At this station, flows in 1941 and 1942 were 26% and 52%, respectively, of the 1941–2005 long-term mean. 4.3. Trends in seasonal discharge For analyses of seasonal trends, 128 stations are available for the period 1961–2000, 94 for 1941–2005 and 56 for the longest period 1920–2005. For Finland and Sweden the number of stations is almost doubled when using data for the period 1941–2005, compared to that available for the period 1920–2005. For some stations, winter values may be less accurate due to typically very small discharge values. Small deviations from the

341

‘true’ value might lead to large percentage deviations. Also, ice jamming and ice cover might affect the stage–discharge relationship. Attention must therefore be given when interpreting results for the winter season. Fig. 5a shows a clear positive trend for the winter season in many regions and for all periods. Exceptions are the west coast of central and northern regions of Norway, the eastern part of Finland, northern and eastern Denmark and Iceland where there are no strong positive trends in any of the periods. The trend toward increased winter flow diminishes for the northern parts of Norway, Sweden and Finland and for Denmark for the period 1961–2000. Positive trends in winter runoff might be caused by a general increase in temperature during recent decades in combination with increased winter precipitation in many regions (Førland et al., 2000; Moberg et al., 2006). The main pattern can also be observed for the period 1920–2005 which starts with several cold and wet years in the 1920s. A marked increase in streamflow has also taken place in the Nordic countries during spring (Fig. 5b), likely caused by an increase in both precipitation and temperature. This corresponds

Fig. 5. Trends in seasonal streamflow for the periods 1920–2005 (left), 1941–2005 (middle) and 1961–2000 (right).

342

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

Fig. 5 (continued)

with an earlier snowmelt and spring flood for many catchments in Norway, Sweden and Finland, and Jutland in Denmark (Fig. 6). There are only a few stations with positive trends which are situated in the coastal region of south-west Norway, south-east Sweden and southern Finland. Different seasons often have opposite trends (Hisdal et al., 1995). This is apparent for the periods 1920–2005 and 1941– 2005, when comparing the summer season to winter and spring. During summer (Fig. 5c) there is a tendency for reduced streamflow in the central, southern western and eastern regions of Norway in the periods 1920–2005 and 1941–2005. A change in the timing of the spring flood could be the main reason. In Denmark, central Sweden and Finland there are also a limited number of stations with negative trends during at least one of the three periods being analysed. In contrast, an increase in summer streamflow is identified for central, western and northern Norway, northern Sweden, southern Finland and parts of western Iceland for the period 1961–2000. In Iceland, a negative trend in spring temperature (Jónsdóttir et al., 2005) has delayed the snowmelt in some basins

from the spring to the summer period, generating an increase in summer discharge. For the autumn season there are few strong trends, and most stations show no trend at all (Fig. 5d). In the period 1920–2005, some stations along the Norwegian coast show a weak positive trend, while in the period 1961–2000 stations in the same region show a weak negative trend. 4.4. Trends in floods Spatio-temporal patterns in the magnitude and timing of autumn floods are less clear than those for mean seasonal and annual values. For the longest period, 1920–2005, most stations (57%) in central, southern, eastern and western regions of Norway have a tendency towards reduced autumn flood magnitudes, with 19% of these stations showing a strong trend. For the periods 1941– 2005 and 1961–2000 many of the negative trends disappear, but a tendency towards decreased autumn floods is found along the west coast of Norway. Positive trends are found at a limited

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

343

Fig. 6. Trends in the timing of the spring flood peak for the periods 1920–2005 (left), 1941–2005 (middle) and 1961–2000 (right).

number of stations in Northern Sweden during the period 1941– 2005, but in the 1961–2000 period positive trends dominate. At inland stations in Norway and western Sweden there is a tendency for autumn floods to occur later during the periods 1920– 2005 and 1941–2005. In the shorter period, 1961–2000, this trend almost disappears except in southern Norway. In Denmark, western Norway and eastern Sweden there are a limited number of stations with weak negative trends, but the stations with trends differ between time periods. The magnitude of the spring flood shows no systematic trends either in time or in space. However, there is a clear tendency towards an earlier spring flood in each of the Nordic countries, except for Iceland (Fig. 6). This tendency is found both in regions having a snowmelt flood (inland of Norway, Sweden and Finland) and in regions dominated by rain floods (Denmark and southwestern Sweden). This picture is repeated for all three periods, but the proportion of strong trends identified in Denmark, southern Norway and southern Sweden for the period 1961–2000 is notably higher than during the other periods and is difficult to explain. Ovesen et al. (2000) found that precipitation has increased in

Denmark throughout the period 1920–2000, but seasonal changes in the timing of precipitation were not investigated. The change in timing in inland catchments is likely to be caused by increased temperatures which increase melt-induced streamflow. In Finland, the spring temperature has increased in recent decades as shown by Tuomenvirta (2004) who found that the mean temperature in 1963–2002 was 1.8 °C higher than in 1847–1876. The Icelandic stations are the exception to this tendency. In Iceland, most stations show no trend in the timing of the spring flood, but some stations in the period 1961–2000 show a weak trend towards larger and later floods. This is caused by a negative trend in spring temperature. During a cold spring the spring flood is delayed, more snow accumulates and the probability of a sudden warming followed by a large spring flood becomes higher. 4.5. Trends in drought The spatial trend patterns in drought duration and deficit volume are consistent, although the number of strong trends is

Fig. 7. Trends in drought deficit volume for the periods 1920–2005 (left), 1941–2005 (middle) and 1961–2000 (right).

344

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

slightly greater for drought deficit volume. Therefore, only drought deficit volume trends are shown (Fig. 7). In contrast to the other variables, strong positive trends are indicated in red indicating a trend towards more severe droughts and strong negative trends are indicated in dark blue, which indicates a tendency towards less severe droughts. For all time periods there is a tendency towards more severe summer droughts in southern and eastern Norway. Although, the percentage of stations exhibiting trends for these drought variables is not as large as for summer season flows. For the periods 1941–2005 and 1961–2005, western Norway has trends indicating less severe summer droughts, but this pattern is less evident for the period 1920–2005. For the period 1961– 2000 western Iceland and southern Finland have more negative than positive trends, which indicates that droughts have become less severe in these regions.

5. Comparison with available projections To assess the effects of future climate changes on basin hydrology, detailed climate information in both time and space are needed. Hanssen-Bauer et al. (2005) provide a review of climate projections for Scandinavia (Denmark, Norway and Sweden) based on statistically downscaled climate scenarios. Although the results vary, there are several features in common. Temperature projections for the 21st century predict that the greatest increases in temperature will occur with increasing distance from the coast, at higher latitudes and in the winter season. Precipitation projections are less consistent, as the projected changes are linked to projected changes in the atmospheric circulation, which differs considerably between climate models. However, annual precipitation is expected to increase, with the increase most pronounced for the winter season. For the summer period, the projections show an increase in northern Scandinavia, with several projections indicating a reduction in parts of southern Scandinavia. These findings are consistent with the recent IPCC Fourth Assessment report (Christensen et al., 2007b), the PRUDENCE (http://prudence.dmi.dk/) and ENSEMBLES (van der Linden and Mitchell, 2009) projects. Scenarios also predict more days with intense precipitation and fewer days with light precipitation although regional differences are evident (Skaugen et al., 2003; Boberg et al., 2007; Caroletti and Barstad, 2009). For Norway, these changes are reflected in streamflow projections based on the HadAM3H (Emission scenario A2 and B2) and the ECHAM4/OPYC3 (Emission scenario B2) Global Climate Models. Possible future streamflow projections for 2071–2100 (Engen-Skaugen et al., 2005; Roald et al., 2004, 2006; Beldring et al., 2008; Hanssen-Bauer et al., 2009) are simulated by a Gridded Water Balance Model (Beldring et al., 2003). The projections vary, but there are some common features indicating increased annual discharge in most of Norway, but a slight decrease for some basins in southern, south-eastern and parts of northern Norway. This differs from our findings of observed trends, as increased flows were mainly found in south-western Norway. However, the increase in winter and spring discharge found in the projections agrees with the detected trends. The projections show a reduction in summer flow in most of the country, whereas a decreasing summer discharge trend is only found for the central, southern, western and eastern regions of Norway, and not northern Norway. The earlier timing of the spring flood, as found in the observed streamflow, is also indicated by the projections. The major differences between the trends and the projections are found for the autumn season. The projections indicate an increase in flow for the whole country whereas few trends in this direction were found in each of the periods analysed. The predicted

reduction in spring flood magnitude in regimes with a dominant snowmelt flood and increased autumn floods close to the coast, which corresponds to the expected increase in intense precipitation events, is not reflected in this study of past trends. However, projections of future flood magnitudes are very uncertain due large uncertainties in how rare climate events which generate floods are likely to change in the future. Drought projections as defined in this study are not available. The most systematic change in observed streamflow in Sweden is an increase in winter and spring streamflow, and a trend towards an earlier spring flood. This is a consequence of unusually high temperatures and precipitation in recent years (Lindström and Alexandersson, 2004). The increased temperatures have led to a shorter period for snow accumulation and to earlier snowmelt. These observations agree with streamflow projections (e.g. Bergström et al., 2001; Andréasson et al., 2004). Andréasson et al. (2004) presented hydrological projections based on the A2 and B2 scenarios for 2071–2100, compared to a control climate of 1961–1990. The projections generally suggest that in the future we will experience decreasing streamflow during summer in south-eastern Sweden, decreasing spring flood peaks and increasing autumn flood peaks. These patterns are not, however, evident in observed trends. For Finland, Denmark and Iceland there are no publications on streamflow projections. The discussion for these countries is therefore based on precipitation and temperature projections. Climate projections for Finland point to an increase in precipitation, particularly in the winter and in the spring (http://prudence.dmi.dk/; van der Linden and Mitchell, 2009). Consequently, mean annual flow is expected to increase, but due to the predicted milder winters, spring floods may decrease in central and southern parts of the country. Summer and autumn floods are expected to become more severe; on the other hand, summer droughts might become more frequent and intense. The analysis of the Finnish time series included in this study only reveals some indication of these changes in southern Finland. For Denmark, temperature is expected to increase in all seasons (http://prudence.dmi.dk/; Kjellström, 2004; Räisänen et al., 2004; van der Linden and Mitchell, 2009). Precipitation increases are predicted for annual, winter and to a less extent spring and autumn values (van der Linden and Mitchell, 2009). Summer precipitation projections indicate that rainfall will decrease in this season. As a result, increased winter and decreased summer flow, in particular, are expected. The trends show an increase in winter flow for the periods 1920–2005 and 1961–2000, but not for 1941–2005. No clear trends are identified for the summer period. The trends observed in the Icelandic 1961–2000 streamflow and precipitation series show some similarities, but also some differences with the trends predicted for the future by the multimodel ensemble mean of RCM simulations (A1B scenario) in the ENSEMBLES project (van der Linden and Mitchell, 2009). The projections predict only small annual changes in precipitation in Iceland up to 2050, which is consistent with the absence of trends in observed streamflow over the periods analysed. However, the climate projections indicate that annual precipitation may increase substantially in north-eastern Iceland from 1961–1990 to 2071– 2100 (van der Linden and Mitchell, 2009). This would continue the trend observed for historical precipitation (Jónsdóttir et al., 2005, 2008), but not streamflow.

6. Conclusions An analysis of trends in Nordic streamflow is described based on annual as well as seasonal and extreme values. In some catchments, there are no changes, but distinct regional patterns are

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

found for the periods observed. Therefore, it can be said that streamflow in the Nordic countries has changed, even though some of the trends found depend on the period analysed. There are large areas with increased annual discharge for the periods 1941–2005 and 1961–2000. This increase is mostly absent in 1920–2005 due to some wet, but cold years in the beginning of this period. It should be noted, however, that few stations show a strong trend towards decreased annual flows. This is partially a result of all periods ending in 2000/2005, where the last 10/15 years includes many mild and wet years. Studying other periods could have changed this picture as the trends found are a result of the period analysed. During the winter and spring seasons there has been a notable increase in streamflow in large parts of the Nordic region. This is also a result of the warm and wet period after about 1990. However, the positive trend is also present for the longest period which includes the cool and wet 1920s. This can be explained by the low temperatures leading to precipitation being stored as snow in these seasons in the 1920s; hence even if the annual discharge was high, the winter and spring discharge was not. The summer discharge in the 1920s was high, either due to snowmelt, abundant precipitation or a combination of both, leading to a distinct negative trend in summer streamflow for the period 1920–2005. A decrease in summer flow was also identified for the period 1941–2005, whereas there was an increase in central, western and northern Norway and northern Sweden for the period 1961–2000. For the autumn period very few trends are identified. Recent flood and drought events are often thought of as a consequence of climate change. Our findings do not support a theory of increasing rain floods and decreasing snowmelt floods as, for example, found in the projections for Norway (Roald et al., 2006), as no clear patterns are found related to the magnitude of the flood peaks. The fact that the 1920s were cool is also reflected in the tendency for spring flood peaks to occur earlier over the period 1920– 2005, since snowmelt is likely to occur earlier in warmer years. These trends in annual and seasonal streamflow, floods and droughts result from changes in both temperature and precipitation, but we can conclude that the temperature induced signal is more clearly reflected in streamflow than precipitation influences. For instance, the relative increase in streamflow in the winter and spring, is larger than the increase in annual flows. This is a consequence of temperature affecting the timing of snowmelt and, thus, the seasonal distribution of flows rather than annual totals. Similarly, a change in the timing of the spring flood (except for Iceland) has been observed due to changes in the timing of snowmelt. In addition, a tendency towards more severe summer droughts in southern and western Norway was found, which can be explained by the gradually increasing summer temperatures (Hanssen-Bauer et al., 2009). Acknowledgements The research described in this paper was supported by the Climate and Energy and Climate and Energy Systems projects funded by Nordic Energy Research, the Nordic energy sector and the national hydrometeorological institutions. The organisations that contributed data to the database; the National Environmental Research Institute – Denmark, the Norwegian Water Resources and Energy Directorate, the Finnish Environment Institute, the Swedish Meteorological and Hydrological Institute and the National Energy Authority in Iceland, are gratefully acknowledged. We thank T. Fjeldstad and E. Klausen for loading the data into the database, and are grateful to an earlier reviewer who acknowledged the issue of significance against long-term persistence.

345

References Alcamo, J., Moreno, J.M., Nováky, B., Bindi, M., Corobov, R., Devoy, R.J.N., Giannakopoulos, C., Martin, E., Olesen, J.E., Shvidenko, A., 2007. Europe. In: Parry, M.L., Canziani, O.F., Palutikof, J.P., van der Linden, P.J., Hanson, C.E. (Eds.), 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, pp. 541–580. Andréasson, J., Bergström, S., Carlsson, B., Graham, L.P., Lindström, G., 2004. Hydrological change – climate change impact simulations for Sweden. Ambio XXXIII (4–5), 228–234. Bates, B.C., Kundzewicz, Z.W., Wu, S., Palutikof, J.P., 2008. Climate Change and Water. Technical Paper of the Intergovernmental Panel on Climate Change, IPCC Secretariat, Geneva. Beldring, S., Engeland, K., Roald, L.A., Sælthun, N.R., Voksø, A., 2003. Estimation of parameters in a distributed precipitation-runoff model for Norway. HESS 7, 304–316. Beldring, S., Engen-Skaugen, T., Førland, E.J., Roald, L.A., 2008. Climate change impacts on hydrological processes in Norway based on two methods for transferring regional climate model results to meteorological station sites. Tellus 60A, 439–450. Bergström, S., Carlsson, B., Gardelin, M., Lindström, G., Pettersson, A., Rummukainen, M., 2001. Climate change impacts on runoff in Sweden – assessments by global climate models, dynamical downscaling and hydrological modelling. Clim. Res. 16 (2), 101–112. Birsan, M.-V., Molnar, P., Burlando, P., Pfaundler, M., 2005. Streamflow trends in Switzerland. J. Hydrol. 314, 312–329. Boberg, F., Berg, P., Thejll, P., Christensen, J.H., 2007. Analysis of Temporal Changes in Precipitation Frequencies using PRUDENCE Data. Danish Climate Centre Report 07-03. Danish Meteorological Institute. . Boiten, W., 2000. Hydrometry. Balkema, Rotterdam. Burn, D.H., Hag Elnur, M.A., 2002. Detection of trends and hydrologic variability. J. Hydrol. 255, 107–122. Caroletti, G.N., Barstad, I., 2009. Future extreme precipitation assessment in Western Norway – using a linear model approach. Hydrol. Earth Syst. Sci. Discuss. 6, 7539–7579. Christensen, J.H., Carter, T.R., Rummukainen, M., Amanatidis, G., 2007a. Evaluating the performance and utility of regional climate models: the PRUDENCE project. Clim. Change 81 (Suppl. 1), 1–6. doi:10.1007/s10584-006-9211-6. Christensen, J.H., Hewitson, B., Busuioc, A., Chen, A., Gao, X., Held, I., Jones, R., Kolli, R.K., Kwon, W.-T., Laprise, R., Magaña Rueda, V., Mearns, L., Menéndez, C.G., Räisänen, J., Rinke, A., Sarr, A., Whetton, P., 2007b. Regional climate projections. In: Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B., Tignor, M., Miller, H.L. (Eds.), Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge and New York, pp. 849–940. Clarke, R., 2010. On the (mis) use of statistical methods in hydro-climatological research. Hydrol. Sci. J. 55, 139–144. Cohn, T.A., Lins, H.F., 2005. Nature’s style: naturally trendy. Geophys. Res. Lett. 32 (23). doi:10.1029/2005GL024476. Art. No. L23402. Davison, A.C., Hinkley, D.V., 1997. Bootstrap Methods and their Application. Cambridge University Press, Cambridge. Douglas, E.M., Vogel, R.M., Kroll, C.N., 2000. Trends in floods and low flows in the United States: impact of spatial correlation. J. Hydrol. 240, 90–105. Engen-Skaugen, T., Roald, L.A., Beldring, S., Førland, E.J., Tveito, O.E., Engeland, K., Benestad, R., 2005. Climate Change Impacts on Water Balance in Norway. Met. Report No. 1/2005, Climate, Oslo. Førland, E., Roald, L.A., Tveito, O.E., Hanssen-Bauer, I., 2000. Past and Future Variations in Climate and Runoff in Norway. DNMI-Report 19/2000, Climate, Oslo. Groisman, P.Y., Knight, R.W., Easterling, D.R., Karl, T.R., Hegerl, G.C., Razuvaev, V.N., 2005. Trends in intense precipitation in the climate record. J. Clim. 18, 1326– 1350. Grønås, S., Kvamme, D., Teigen, R., 2005. Og regnet strømmet ned på jorden (And it Poured over the Earth, in Norwegian), Cicerone, vol. 14, issue 1. Centre for International Climate and Environmental Research, Oslo, pp. 24–26. Hanna, E., Jónsson, T., Box, J.E., 2004. An analysis of Icelandic climate since the nineteenth century. Int. J. Climatol. 24, 1193–1210. Hanssen-Bauer, I., Førland, E.J., Tveito, O.E., 1995. Trends and Variability in Annual Precipitation in Norway, Klima-Report 27/95. Norwegian Meteorological Institute. Hanssen-Bauer, I., Achberger, C., Benestad, R.E., Chen, D., Førland, E.J., 2005. Review: statistical downscaling of climate scenarios over Scandinavia. Clim. Res. 29, 255–268. Hanssen-Bauer, I., Drange, H., Førland, E.J., Roald, L.A., Børsheim, K.Y., Hisdal, H., Lawrence, D., Nesje, A., Sandven, S., Sorteberg, A., Sundby, S., Vasskog, K., Ådlandsvik, B., 2009. Klima i Norge 2100 Bakgrunnsmateriale til NOU Klimatilpassing (Climate in Norway 2100 Background Material to the NOU Climate Adaption). Norsk klimasenter, Oslo. Hisdal, H., Erup, J., Gudmundsson, K., Hiltunen, T., Jutman, T., Ovesen, N.B., Roald, L.A., 1995. Historical Runoff Variations in the Nordic Countries. NHP-Report No. 37. Norwegian Hydrological Council, Oslo.

346

D. Wilson et al. / Journal of Hydrology 394 (2010) 334–346

Hisdal, H., Stahl, K., Tallaksen, L.M., Demuth, S., 2001. Have streamflow droughts in Europe become more severe or frequent? Int. J. Climatol. 21, 317–333. Hisdal, H., Tallaksen, L.M., Frigessi, A., 2002. Handling non-extreme events in extreme value modelling of streamflow droughts. In: van Lanen, H.A.J., Demuth, S. (Eds.), FRIEND 2002 – Regional Hydrology: Bridging the Gap between Research and Practice. IAHS Publ. No. 274, pp. 281–288. Hisdal, H., Holmqvist, E., Hyvärinen, V., Jónsson, P., Kuusisto, E., Larsen, S.E., Lindström, G., Ovesen, N.B., Roald, L.A., 2003. Long Time Series – a Review of Nordic Studies, Climate, Water and Energy Project. Report No. 2, Reykjavik. Hisdal, H., Tallaksen, L.M., Clausen, B., Peters, E., Gustard, A., 2004. Hydrological drought characteristics. In: Tallaksen, L.M., van Lanen, H.A.J. (Eds.), Hydrological Drought – Processes and Estimation Methods for Streamflow and Groundwater, Developments in Water Science, vol. 48. Elsevier Science B.V, Amsterdam, pp. 139–198. Hisdal, H., Holmqvist, E., Jónsdóttir, J.F., Jónsson, P., Kuusisto, E., Lindström, G., Roald, L.A., 2010. Has Streamflow Changed in the Nordic Countries? Report 12010. Norwegian Water Resources and Energy Directorate, Oslo. Hyvärinen, V., 1998. Observed trends and fluctuations in hydrological time series in Finland – a review. In: Proceedings of the Second International Conference on Climate and Water, Espoo, Finland, pp. 1064–1070. Hyvärinen, V., 2003. Trends and characteristics of hydrological time series in Finland. Nordic Hydrol. 34 (1/2), 71–90. Jones, P.D., Moberg, A., 2003. Hemispheric and large scale surface air temperature variations: an extensive revision and an update to 2001. J. Clim. 16, 206–223. Jónsdóttir, J.F., Jónsson, P., Uvo, C.B., 2005. Trend analysis of Icelandic discharge, precipitation and temperature series. Nordic Hydrol. 37 (4–5), 365–376. Jónsdóttir, J.F., Uvo, C.B., Clarke, R.T., 2008. Trend analysis in Icelandic discharge, temperature and precipitation series by parametric methods. Hydrol. Res. 39, 425–436. Jørgensen, A.M.K., Fenger, J., Halsnæs, K., 2001. Climate Change Research – Danish Contributions. Gads Forlag, Copenhagen. Kjellström, E., 2004. Recent and future signatures of climate change in Europe. Ambio 33 (4–5), 193–198. Klein Tank, A.M.G., Wijngarrd, J.B., Konnen, G.P., Bohm, R., Demaree, A., Gocheve, M., Mileta, S., Pashiardis, L., Hejkrlik, C., Kern-Hansen, R., Heino, P., Bessemoulin, G., Muller-Westermeier, G., Tzanakou, M., Szalai, S., Palsdottir, D., Fitzgerald, D., Rubin, S., Capaldo, M., Maugeri, M., Leitass, A., Bukantis, A., Aberfeld, R., van Engelen, A.F.V., Forland, E., Mietus, M., Coelho, F., Mares, C., Razuvaev, V., Nieplova, E., Cegnar, T., López, J.A., Dahlstrom, B., Moberg, A., Kirchhofer, W., Ceylan, A., Pachaliuk, O., Alexander, L.V., Petrovic, P., 2002. Daily dataset of 20th-century surface air temperature and precipitation series for the European Climate Assessment. Int. J. Climatol. 22, 1441–1453. Kundzewicz, Z.W., 2004. Searching for change in hydrological data – editorial. Hydrol. Sci. J. 49 (1), 3–6. Kundzewicz, Z.W., 2005. Change detection in high river flows in Europe. In: Franks, S., Wagener, T., Bøgh, B., Gupta, H.V., Bastidas, L., Nobre, C., de Oliveira Gãlvao, C. (Eds.), Regional Hydrological Impacts of Climate Change – Hydroclimatic Variability. IAHS Publ. 296, pp. 71–80. Kundzewicz, Z.W., Robson, A.J., 2004. Change detection in hydrological data – a review of the methodology. Hydrol. Sci. J. 49 (1), 7–19. Kundzewicz, Z.W., Graczyk, D., Maurer, T., Pin´skwar, I., Radziejewski, M., Svensson, C., Szwed, M., 2005. Trend detection in river flow series: 1. Annual maximum flow. Hydrol. Sci. J. 50 (5), 797–810. Lindström, G., Alexandersson, H., 2004. Recent mild and wet years in relation to long observation records and future climate change in Sweden. Ambio XXXIII (4–5), 183–186. Lindström, G., Bergström, S., 2004. Runoff trends in Sweden 1807–2002. Hydrol. Sci. J. 49 (1), 69–83. Liu, S., Mo, X., Zhao, W., Naeimi, V., Dai, D., Shu, C., Mao, L., 2009. Temporal variations of soil moisture over the Wuding River basin assessed with an ecohydrological model, in-situ observations and remote sensing. Hydrol. Earth Syst. Sci. 13, 1375–1398. Mitosek, H.T., 1995. Climate variability and change within the discharge time series: a statistical approach. Clim. Change 29, 101–116. Moberg, A., Jones, P.D., Lister, D., Walther, A., Brunet, M., Jacobeit, J., Alexander, L.V., Della-Marta, P.M., Luterbacher, J., Yiou, P., Chen, D., Klein Tank, A.M.G., Saladié, O., Sigró, J., Aguilar, E., Alexandersson, H., Almarza, C., Auer, I., Barriendos, M., Begert, M., Bergström, H., Böhm, R., Butler, C.J., Caesar, J., Drebs, A., Founda, D., Gerstengarbe, F.-W., Micela, G., Maugeri, M., Österle, H., Pandzic, K., Petrakis, M., Srnec, L., Tolasz, R., Tuomenvirta, H., Werner, P.C., Linderholm, H., Philipp, A., Wanner, H., Xoplaki, E., 2006. Indices for daily temperature and precipitation extremes in Europe analysed for the period 1901–2000. J. Geophys. Res. 111. doi:10.1029/2006JD007103. Art No. D22106. Ovesen, N.B., Iversen, H.L., Larsen, S.E., Müller-Wohlfeil, D-I., Svendsen, L.M., Blicher, A.S., Jensen, P.M., 2000. Afstrømning i danske vandløb. (Runoff in Danish Watercourses, in Danish). Faglig rapport fra DMU, Nr. 340. Danmarks Miljøundersøgelser, Silkeborg, Denmark. Parry, M.L., Canziani, O.F., Palutikof, J.P., van der Linden, P.J., Hanson, C.E. (Eds.), . Climate Change 2007: Impacts, Adaption and Vulnerability. Contribution of

Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge. Räisänen, J., Hansson, U., Ullerstig, A., Döscher, R., Graham, L.P., Jones, C., Meier, H.E.M., Samuelsson, P., Willén, U., 2004. European climate in the late twentyfirst century: regional simulations with two driving global models and two forcing scenarios. Clim. Dynam. 22, 13–31. Rees, G., Demuth, S., 2000. The application of modern information system technology in the European FRIEND project. Moderne hydrologische informationssysteme. Wasser Boden 52 (13), 9–13. Renard, B., Lang, M., Pois, 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. doi:10.1029/2007WR006268. Art. No. W08419 2008. Roald, L.A., Wesselink, A.J., Arnell, N.W., Dixon, J.M., Rees, H.G., Andrews, A.J., 1993. European water archive. In: Gustard, A. (Ed.), Flow Regimes from International Experimental and Network Data (FRIEND). Institute of Hydrology, Wallingford, pp. 7–20. Roald, L.A., Beldring, S., Skaugen, T., 2004. Scenarios of annual and seasonal runoff in Norway. In: Järvet, A. (Ed.), XXIII Nordic Hydrological Conference – Selected Articles NHP Report 48, vol. I, pp. 71–76. Roald, L.A., Beldring, S., Engen-Skaugen, T., Førland, E.J., Benestad, R., 2006. Climate Change Impacts on Streamflow in Norway. Norwegian Water Resources and Energy Directorate, Report A1-2006. Rummukainen, M., Bergström, S., Persson, G., Rodhe, J., Tjernström, M., 2004a. The Swedish regional climate modelling programme, SWECLIM: a review. Ambio XXXIII (4–5), 176–182. Rummukainen, M., Bergström, S., Persson, G., Rodhe, J., Tjernström, M., (Eds.), 2004b. Ambio Special Issue: Swedish Regional Climate Modelling Programme. Ambio XXXIII (4–5), 175–274. Salas, J.D., 1993. Analysis and modelling of hydrologic time series. In: Maidment, D.R. (Ed.), Handbook of Hydrology. McGraw-Hill, New York, pp. 19.1–19.72. Skaugen, T., Astrup, M., Roald, L.A., Førland, E.J., 2003. Scenarios of extreme daily precipitation for Norway under climate change. Nordic Hydrol. 15 (1), 1–13. Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B., Tignor, M., Miller, H.L. (Eds.), 2007. Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge and New York. Stranden, H.B., Skaugen, T., 2009. Trends in Annual Maximum Snow Water Equivalent in Norway (1914–2008). Norwegian Water Resources and Energy Directorate, Report 3-2009, Oslo. Svensson, C., Kundzewicz, Z.W., Maurer, T., 2005. Trend detection in river flow series: 2. Flood and low-flow index series. Hydrol. Sci. J. 50 (5), 811–824. Tallaksen, L.M., Madsen, H., Clausen, B., 1997. On the definition and modelling of streamflow drought duration and deficit volume. Hydrol. Sci. J. 42 (1), 15–33. Trenberth, K.E., Jones, P.D., Ambenje, P., Bojariu, R., Easterling, D., Klein Tank, A., Parker, D., Rahimzadeh, F., Renwick, J.A., Rusticucci, M., Soden, B., Zhai, P., 2007. Observations: surface and atmospheric climate change. In: Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B., Tignor, M., Miller, H.L. (Eds.), Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge and New York, pp. 237–336. Tuomenvirta, H., 2004. Reliable Estimation of Climatic Variations in Finland. Finnish Meteorological Institute Contributions no. 43. Finnish Meteorological Institute, Helsinki. van der Linden, P., 2009. Introduction to the ENSEMBLES Project. ENSEMBLES: Climate Change and its Impacts: Summary of Research and Results from the ENSEMBLES Project. Met Office Hadley Centre, Exeter, UK. pp. 3–6. van der Linden, P., Mitchell, J.F.B. (Eds.), 2009. ENSEMBLES: Climate Change and its Impacts: Summary of Research and Results from the ENSEMBLES Project. Met Office Hadley Centre, Exeter, UK. Wang, X.L.L., Swail, V.R., 2001. Changes of extreme wave heights in Northern Hemisphere oceans and related atmospheric circulation regimes. J. Clim. 14 (10), 2204–2221. Yevjevich, V., 1967. An Objective Approach to Definition and Investigations of Continental Hydrologic Droughts. Hydrology Papers 23. Colorado State University, Fort Collins. Yue, S., Pilon, P., 2004. A comparison of the power of the t test, Mann–Kendall and bootstrap tests for trend detection. Hydrol. Sci. J. 49 (1), 21–37. Yue, S., Wang, C.Y., 2002. Regional streamflow trend detection with consideration of both temporal and spatial correlation. Int. J. Climatol. 22, 933–946. Yue, S., Pilon, P., Phinney, B., 2003. Canadian streamflow trend detection: impact of serial and cross-correlation. Hydrol. Sci. J. 48 (1), 51–63. Zhang, X., Zwiers, F.W., 2004. Comment on ‘‘Applicability of prewhitening to eliminate the influence of serial correlation on the Mann–Kendall test” by Sheng Yue and Chun Yuan Wang. Water Resour. Res. 40 (3). doi:10.1029/ 2003WR002073. Art. No. W03805.