Journal of Geophysical Research: Atmospheres

Nonstationary influence of the North Atlantic Oscillation on European precipitation

Authors


Abstract

[1] This paper analyzes the changing role of the North Atlantic Oscillation (NAO) in European precipitation in relation to interdecadal shifts in the atmospheric dynamic over the North Atlantic region. For this purpose, moving window correlations between precipitation and NAO have been related to the interdecadal variability of atmospheric pressure patterns. Analyses have been replicated using observed climate data for the twentieth century, paleoclimate reconstructions since 1785, and physically modeled climate by AOGCMs (CGCM3.1(T63), 20C3M). Results evidence that nonstationary relationship of the NAO across Europe is linked to interdecadal shifts in the location of the positions of the NAO pressure centers, which have been occurred at least since the last 2 centuries. Such displacement of the pressure NAO centers points to possible existence of different NAO patterns, which may help to explain why studies relating NAO to European climate behavior have not shown systematically strong correlations in the course of the twentieth century. A general trend toward a strengthening of the NAO-precipitation relationship over most of Europe has been detected for the twentieth century.

1. Introduction

[2] The North Atlantic Oscillation (NAO) is the main atmospheric circulation pattern in the North Atlantic sector, and affects the surface climate across large parts of Europe [Wanner et al., 2001; Trigo et al., 2002; Hurrell et al., 2003a]. The NAO is characterized by a spatial dipole with two pressure anomaly centers located near the Azores and Iceland [Hurrell, 1995]. During positive (negative) phases, both pressure centers are reinforced (weakened) and displaced to the south (north). Winter precipitation in Europe is largely determined by the phase of the NAO: positive (negative) phases lead to positive (negative) precipitation anomalies in north (south) Europe [Hurrell and van Loon, 1997].

[3] Nevertheless, the NAO influence cannot be considered stable since a nonstationary relationship between NAO and surface climate has been reported for Europe [Osborn et al., 1999; Slonosky et al., 2001; Walter and Graf, 2002; Lu and Greatbatch, 2002], including precipitation [Zveryaev, 2006; Beranová and Huth, 2008; Pauling et al., 2006]. Several hypotheses have been formulated to explain this nonstationarity, including modifications to the meridional pressure gradient [Zveryaev, 2006], North Atlantic air-sea dynamics and variability in thermohaline circulation [Walter and Graf, 2002], solar activity [Gimeno et al., 2003], and variability in the NAO [Haylock et al., 2007]. Moreover, neither the sign nor magnitude of the NAO nor the frequency of NAO extreme years explain the nonstationary relationship [Beranová and Huth, 2007; Haylock et al., 2007]. Jung et al. [2003] and Beranová and Huth [2008] have recently identified an intensification of the NAO influence on European climate since the late 1970s decade until the mid of the 90s, which coincides with an eastward shift in the location of the NAO pressure centers. This would suggest that the displacements of the NAO positions could help to explain nonstationary relationship between NAO and European precipitation. Nevertheless, it has not been previously analyzed whether other changes in the position of the NAO centers have been recorded in different periods that may help to explain nonstationary relationships.

[4] Here, we analyze the changing role of the NAO on European precipitation since 1785 and we propose an explanation of the nonstationary influence based on the interdecadal shifts in the location of the positions of the NAO pressure centers. In order to obtain robust and consistent results, analyses have been conducted by using three different data sources: (1) observed data during the twentieth century; (2) reconstructed precipitation and sea level pressure since 1785; and (3) physically modeled climate by Atmosphere-Ocean Global Climate Change Models (AOGCMs).

[5] Variability in the North Atlantic Oscillation (NAO) has significant implications in Europe for the economy, water resources availability, and many environmental processes. Thus, droughts, forest fires, water availability, and hydropower production have been shown to be associated with phases of the NAO [Hurrell et al., 2003a]. For those reasons, clarifying the mechanism that explains the changing influence of the NAO on surface climate is an important challenge, with also serious implications for climate downscaling, weather predictability, and projections of climate change.

2. Data and Methods

2.1. Databases

2.1.1. Monthly Precipitation Data Set

[6] We used a monthly gridded precipitation database at a latitude/longitude resolution of 10′ for the entire European continent, as compiled by the Climate Research Unit of the University of East Anglia (http://www.cru.uea.ac.uk/~timm/grid/TYN_SC_1_0.html [Mitchell and Jones, 2005]). The data set presents terrestrial surface climate for the 1901–2000 period, and has both higher spatial resolution than other data sets of similar temporal extent and a longer temporal coverage than other data sets of similar spatial resolution. The monthly gridded precipitation database is a revised and extended version of a previous data set [New et al., 1999, 2000]. The precipitation grids were interpolated directly from station observations, but before interpolating the database was checked for inhomogeneities in the station records using an automated method that is superior to previous methods by using incomplete and partially overlapping records and by detecting inhomogeneities with opposite signs in different seasons. Inhomogeneities were corrected so as to make the record consistent with its final values. Further confirmation of the reliability of this data set was found in the good agreement with satellite-based precipitation data from the Climate Prediction Center Merged Analysis of Precipitation (CMAP) data set [Xie and Arkin, 1997], in terms of both climatology and the leading modes of variability [Zveryaev, 2004].

2.1.2. North Atlantic Oscillation Index

[7] The NAO is quantified both in terms of spatial configuration and temporal intensity using NAO indices obtained for the period of instrumental information using SLP data [Hurrell et al., 2003b]. Two general procedures are followed in obtaining the NAO indices: (1) multivariate analysis (Principal Component Analysis) based on surface pressure grids or geopotential height fields, and (2) analysis of a time series of surface pressures, particularly those recorded at sea level [Hurrell, 1995; Jones et al., 1997]. As a consequence of the use of different procedures for obtaining NAO indices, with different NAO indices derived from instrumental records, the problem arose as to which NAO index to select. Wallace [2000] concluded that station-based indices do not provide an optimal representation of the NAO because of seasonal displacement of the Iceland low and the Azores anticyclone. Nevertheless, the use of PCA in defining the NAO index has a number of important shortcomings, because spatial patterns and corresponding PC time series are dependant on the size of the window used in the analysis. Osborn et al. [1999] showed that NAO indices derived from PCA or single stations show similar time evolutions, with correlations in the range of 0.84 to 0.96. Their analysis showed that the interannual variability of the winter NAO is unaffected by the choice of method used to calculate the NAO index.

[8] In the present study we used a NAO index based on station SLP time series. In this approach the NAO index was calculated from the gradient in surface pressure between observation points in Iceland and the area west of the Iberian Peninsula, including the Azores. These points were chosen because they are located close to the south and north pressure centers that characterize the NAO. Specifically, we used the NAO index developed by Jones et al. [1997], which is based on Gibraltar (southwest Iberian Peninsula) and Reykjavik (southwest Iceland) stations. The NAO index was compiled by the Climate Research Unit of the University of East Anglia (http://www.cru.uea.ac.uk/cru/data/nao.htm) for the period 1821 to the present. Although the stations at Ponta Delgada (the Azores) and Lisbon have also been used as the south node in calculating the NAO index [Hurrell, 1995], Gibraltar appears to better represent the southern part of the NAO dipole. Thus, Osborn et al. [1999] showed that the ratio of interdecadal to interannual variance is higher for those indices that use data from a southern node, such as Gibraltar, than for those that use information from a node located farther west, such as the Azores. Moreover, the use of a southwest European location in calculating the NAO provides a slightly better correlation with station and precipitation series across Europe than when data from the Azores [Hurrell, 1995; Jones et al., 2003] are used.

[9] The NAO is mainly active during the boreal winter, when the NAO determines the climate over extensive areas of Europe [Wanner et al., 2001] Therefore, in the present study we used winter data (December to March) to analyze the changing influence of the NAO on European winter precipitation. Interdecadal variability in the NAO appears to be most coherent when the December–March season is used [Osborn et al., 1999].

2.1.3. Pressure Data Set

[10] The pressure data set was provided by NCEP-NCAR (http://dss.ucar.edu/datasets/ds010.1/), with a spatial latitude/longitude resolution of 5°. The 5° latitude/longitude grids begin in 1899, and cover the North Atlantic region (20°N–80°N, 90°W–40°E). The few data gaps in the data set were filled by interpolation (splines with tension) using the data of neighboring grid points.

2.1.4. Climate Reconstruction

[11] To verify the results obtained using instrumental observations for the twentieth century, and for comparative purposes, we repeated the analysis for the nineteenth century. The lack of instrumental information for most regions made it necessary to use proxy data. We used reconstructions performed by the Climatology and Meteorology Research Group of the University of Bern (Switzerland). Among the available precipitation reconstructions for Europe, the unique data set that covers the whole continent is that developed by Pauling et al. [2006]. The database contains information from 1500, but we used data for the nineteenth century (or more specifically, from 1785 to 1900) because the uncertainty of reconstruction is much higher prior to this period [Pauling et al., 2006]. The winter season precipitation does not exactly coincide with that used for the twentieth century, since it covers the months of December to February. Among the reconstructions of the NAO index, we used the one developed by Luterbacher et al. [2002a], as it is highly reliable and contains monthly NAO reconstruction data, allowing matching with the December-to-February winter season of the reconstructed precipitation grids. Monthly SLPs of the winter season for the region 70°N–30°W, 30°N–40°E were also used [Luterbacher et al., 2002b]. The predictors used to reconstruct SLPs were independent from those used for the precipitation reconstructions (see details in work by Pauling et al. [2006]). The SLP database does not cover the same region as does the winter SLP data set for the twentieth century, but the database can be used to determine if the changes in the NAO-precipitation relationships are related to shifts in the positions of the SLP centers that characterize the NAO. All of these data are available from the World Data Center for Paleoclimatology (http://www.ncdc.noaa.gov/paleo/recons.html).

2.1.5. Atmosphere-Ocean Global Climate Change Model

[12] Grids of precipitation and SLP at a spatial resolution of 2.8°, simulated by the AOGCM CGCM3.1(T63) of the Canadian Climate Modeling and Analysis Centre (N. A. McFarlane et al., The CCCma third generation atmospheric general circulation model, 2005, CCCma internal report, http://www.cccma.ec.gc.ca/models/gcm3.shtml), were used to replicate analysis. In this study the output of the IPCC twentieth century experiment (20C3M) was used, in which initial conditions were taken from 1 January of year 1 (1870), and several forcing agents were varied during the 130-year duration of the model run (1870–1999). The simulation uses scenario 20C3M which starts from the midnineteenth century using historical (or estimated) series of greenhouse gases, sulphate aerosols direct effects, volcanoes and solar forcing. Data were obtained from the Coupled Model Intercomparison Project (CMIP3) database at https://esg.llnl.gov:8443/index.jsp. A NAO index was calculated from the winter SLP values (December–March) simulated by the CGCM3.1(T63) model, using the closest grid points to Iceland (62.78°N, 22.5°W) and Gibraltar (37.67°N, 5.62°W).

2.2. Identification of Nonstationarities of NAO on European Precipitation

[13] To analyze the nonstationary response of European precipitation to the NAO we calculated moving-window correlations between the NAO and winter precipitation using the Pearson correlation coefficient (r). A confidence level of p < 0.05 was chosen as indicative of significant correlations. The first calculation involved the initial 31 years (1902–1932), with the result (r) being assigned to the central year of this interval (1917). The second calculation was based on the years 1903–1933, and this process was repeated up to the years 1970–2000. The same procedure was applied to the climate reconstruction data set and the simulations from the CGCM3.1(T63) model.

2.3. Identification of Shifts in NAO Pressure Centers by Moving-Window PCA of Surface Pressure Series

[14] To identify the main spatial pattern of SLP anomalies in the North Atlantic region, and changes in the pattern over time, shifts in the positions and intensities of the surface pressure centers associated with the NAO dipole were analyzed using surface pressure grids and moving-window T-mode PCA (31-year periods). PCA attempts to synthesize in a small number of new uncorrelated variables most of the total variation of a large number of highly intercorrelated variables. The procedure has been widely applied for climatological studies [e.g., Richman, 1986; Jollife, 1986, 1990; Serrano et al., 1999; Huth, 2006]. The obtained uncorrelated variables are called principal components (PCs) and consist of linear combinations of the original variables. In terms of climate variables, two possible options can be applied to PCA: the S and T modes. The T-mode is the result of choosing the individual temporal observations as the variables, and the stations as the cases of those variables. T-mode identifies subgroups of observations with similar spatial patterns and it is the optimal procedure to obtain the most general spatial configuration of a climate variable. A correlation matrix was selected to provide an efficient representation of variance within the data set in PCA [Barry and Carleton, 2001]. Moving-window T-mode PCA was applied to SLP reconstructions for the nineteenth century, and observations and model simulations for the twentieth century. As the grids span from 20°N to 80°N, different grid surface areas were adjusted by a weighting coefficient of the cosine of latitude.

[15] The procedure is very similar to that used in analyzing the moving-window correlations between the winter NAO and precipitation series. In this case, the first T-mode PCA from the SLP anomaly series was performed using data from the interval 1902 to1932. T-mode PCA enables the leading mode of variability in atmospheric circulation to be obtained, which is then assigned to the year 1917. The following T-mode PCA was calculated for the years 1903 to 1932, with the leading mode assigned to 1918. The analysis was continued using the moving-window procedure until the period 1970–2000, with the dominant leading mode assigned to the year 1985. The percentage of variance explained by the leading mode of each PCA oscillated between 26.1 and 31.1%, always being the displayed spatial pattern identified as the NAO pattern. This implies that the NAO always appears in each 30-year subperiod as the leading mode of variability in atmospheric circulation.

2.4. Identification of NAO General Patterns

[16] The general spatial configurations of the NAO, from observed and CGCM3.1(T63) modeled SLP values, were obtained by means of a rotated T-mode PCA, using the first leading modes derived from each moving-window 31-year T-mode PCA as variables, converted in absolute values to avoid that the positive and negative PC scores in different periods hide the general spatial pattern of the SLP anomalies, which is independent of the sign. Rotation was selected because this procedure redistributes the final explained variance, thus enabling a clearer separation of components but maintaining their orthogonality [Huth, 2006]. We used the Varimax rotation, which is the most widely applied option, as it produces more stable and physically robust patterns [Richman, 1986]. In the present analysis an objective N-rule was applied to select the number of components to retain. The N-rule divides the total variability into signal and noise components [Trigo and Palutikof, 2001]. Only the signal components (those PCs that are statistically significant at the 5% level) were retained. Temporal evolution of each pattern was identified from the time series of factorial loadings, thereby identifying periods in which the NAO pattern was representative.

3. Results and Discussion

[17] Figure 1 shows some examples of 31-year moving-window correlations between the winter NAO and precipitation instrumental observatories and neighbor selected points in the gridded database throughout the European continent for the period 1902–2000. Evolution of moving correlations inform about the markedly unstable response with time of European precipitation to NAO index. Moreover, similar patterns were identified using the time series of individual observatories, with no significant differences in correlations observed at stations and grid points. The unique statistical differences (t-test) are found in Bodo, Uppsala and Wien, although the temporal variability is close similar. This demonstrates that nonstationary relationships in the NAO-precipitation correlations were not the result of changes in the station network or inhomogeneity in the gridded data set.

Figure 1.

Example of 31-year moving-window correlations between the winter NAO and precipitation for points throughout the European continent. Red lines represent the temporal evolution of NAO-precipitation correlations for selected points in the gridded database (gray in the map). Blue lines represent moving-window correlations for instrumental observatories with complete records for the period 1902–2000 (black in the map). Horizontal dotted lines represent the significance threshold at 95%. Observatory data were obtained from the Global Historical Climatology Network (http://www.ncdc.noaa.gov/oa/climate/ghcn-monthly/).

[18] Figure 2 shows spatial distribution of correlations between winter precipitation and the NAO index during the twentieth century (Figure 2, right) and the leading principal components of winter SLPs obtained from T-mode PCA (Figure 2, left). Although the general pattern of negative (positive) correlations in the south (north) of Europe between winter precipitation and NAO index appears generally well defined for all subperiods, the spatial pattern and magnitude of the correlations varied noticeably over time.

Figure 2.

(right) Spatial distribution of correlations between winter precipitation and the NAO index during the twentieth century. The year indicated on each map represents the midpoint of each 31-year period. Units are Pearson coefficients of correlation (r). Dotted lines enclose areas with significant correlations (p < 0.05). (left) Leading principal components of winter SLPs obtained from T-mode PCA. The leading principal components were obtained using moving-window periods of 31 years, centered on the year indicated in each plot. The percentage of variance explained by the leading mode in each period is also shown. Since the sign of the PC pattern is arbitrary and to make easier the comparison, a common sign of the NAO was selecting for plotted (negative in the north and positive in the south).

[19] The spatial patterns of changes in correlations are in agreement with the shifts in position of the main sea level pressure (SLP) centers that characterized the NAO dipole throughout the twentieth century (Figure 2, left). An eastward shift in the SLP anomalies associated with the NAO has been proposed to explain the strengthening relationship between the NAO and surface climate in northern Europe in the last years of the twentieth century [Beranová and Huth, 2008; Jung et al., 2003; Jung and Hilmer, 2001], and some studies have suggested that the location of the NAO pressure centers undergoes secular changes, although presumably only infrequently under the current climatic conditions [Ulbrich and Christoph, 1999]. Nevertheless, using the 31-year moving-window T-mode principal component analysis (PCA) of SLP over the entire North Atlantic region, we found that the position and magnitude of the SLP centers of the NAO dipole show marked interdecadal variability. Thus, the documented recent eastward change is not an isolated phenomenon in terms of NAO behavior during the last century (Figure 2). The spatial pattern of the NAO in the first decades of the twentieth century explains the weak NAO-precipitation correlations obtained for southern Europe, and the strong positive correlations obtained for the area around the North Sea Basin. Thus, the location of the NAO dipole in the early decades of the twentieth century displaced the Southern Pressure Center (SPC) over North America with little influence on southern Europe, and the Northern Pressure Centre (NPC) was located over Greenland and extended toward the North Sea, affecting large areas of central Europe. In the 1930s and 1940s the correlations in southern Europe increased in terms of both magnitude and surface extent, which is in agreement with the increased influence of the SPC in this region. In contrast, the extent of surface area showing significant correlations decreased in northern regions, which coincides with a northward displacement of the NPC. A minimum positive correlation (in terms of both magnitude and spatial extent) was recorded in northern Europe between 1960 and 1965, in agreement with a northward displacement of the NPC. The weakening of NAO-precipitation correlations in southern Europe around 1970 is explained by reduced pressure anomalies and the split of the SPC into two centers, one located in the mid-Atlantic and the other in the eastern part of the Mediterranean Basin. During the last decades of the twentieth century, correlations strengthened markedly in northern (positive) and southern (negative) Europe, producing the strongest NAO-precipitation relationships for the entire century. The strengthening of these correlations coincides with an increase in the influence of both pressure centers in Europe. The SPC appears to have strengthened in the 1985 plot, where it is centered in the Mediterranean Basin, whereas the NPC shows a clear shift to the east.

[20] Thus, the spatial pattern revealed a general strengthening of the influence of the NAO throughout most of Europe. Figure 3 shows the spatial distribution of trends in the moving window correlation series between wintertime NAO and precipitation. In north and northeast Europe, correlations between NAO and precipitation have become increasingly positive, whereas in South and Southwest Europe the correlations are more negative. Nevertheless, this approach does not enable to determine the strength of NAO-precipitation correlations. We also calculated the trend with the absolute magnitude values of the moving window correlation (negative values converted to positive values). The obtained trend in absolute values indicates that the magnitude of correlations (independent of sign) shows a general trend toward a strengthening of the NAO-precipitation relationship during the twentieth century over most of Europe. The exceptions are those areas in Northwest Europe most open to the west and northwest flows (western British Isles, the Netherlands, and western Scandinavia), as the recent displacement of the north pressure center to the east has weakened the NAO influence over these regions.

Figure 3.

Spatial distribution of trends in the moving window correlation series between wintertime NAO and precipitation. Values are the nonparametric coefficient of correlation (Rho-Spearman) between the series of moving window correlations and the series of years. Before calculating trends series were corrected for autocorrelation according to the procedure proposed by Yue et al. [2002]. Dotted lines enclose areas with significant trends (p < 0.05). (a) Original sign of the moving window correlation series. (b) Moving window correlation series in absolute values.

[21] Interdecadal shifts in the locations of NAO pressure centers were not only unique for the twentieth century, and they also contribute to most of the changes recorded in the NAO-precipitation relationship since at least the late eighteenth century, analyzed by means of climate reconstructions (Figure 4). The variance explained by NAO between the observed and reconstructed data was quite different (30% and 45%, respectively). Reconstructed SLP data was created for a number of predictors that cannot account for the spatial variability of SLP over Europe neither the local features recorded by observations. For this reason, the first PC from reconstructed SLP groups a higher percentage of variance than observations since the most specific features commonly are not recorded in climate reconstructions. Noticeable differences in magnitude and spatial pattern are evident among the correlations in the four periods, related to shifts in the positions of the NAO SLP centers. For example, the 1870 and 1880 plots shows a reinforcement of NAO-precipitation correlations in southeastern Europe related to a general displacement of the SPC to the east Mediterranean area. Also, for northern Europe the decreased correlation in these plots is clearly related to displacement of the NPC to Greenland.

Figure 4.

As for Figure 2, but using nineteenth century SLP and precipitation reconstructions.

[22] Moreover, similar results were obtained from climate data under a twentieth century emissions scenario (20C3M) modeled using the CGCM3.1(T63) model. Although some differences between the NAO spatial pattern obtained from observations and the model can be observed: mainly the split of the southern center and the southern shift of the eastern part of the southern center; the model records the main features, although more simplified than observations, of the spatial distribution of the NAO-precipitation correlations and also the NAO north-south dipole over the entire analyzed period (Figure 5). From physical-based simulation, noticeable changes in the spatial patterns and magnitude of correlations between the NAO and European winter precipitation were also found (Figure 6). Latitudinal and longitudinal movements of the NPC contribute to most of the changes in the NAO-precipitation correlations in the modeled period. The western displacement of the NPC between 1920 and 1950 contributes to the decrease in magnitude and surface extent of significant correlations in Scandinavia and the British Isles. The increase in correlations in southern Scandinavia, Finland, and large areas of continental northern Europe since 1960 is related to an eastern displacement of the NPC, but with a more southerly component relative to the 1890–1910 period, which contributes to the lack of significant NAO-precipitation correlations in northern Scandinavia. In southern Europe the main feature is the latitudinal movement of the limit of negative and significant NAO-precipitation correlations. The maximum area with significant correlations is found in the first decades of the century. This coincides with the SPC being predominantly located in the southwest of the Iberian Peninsula. The surface extent and magnitude of significant correlations decreased from 1940 to 1960, mainly in southwestern Europe, coinciding with displacement of the SPC to the east. The 1970 and 1980 figures indicate a new westward displacement of the SPC, consistent with the increased correlation magnitude in the Iberian Peninsula and north Africa, and a decrease in the area with significant correlations in the Balkans and southeast Europe.

Figure 5.

(a) (right) Spatial distribution of correlations between winter precipitation and the NAO index obtained from observations between 1902 and 2000. (left) Leading principal component of winter SLPs form observations. (b) (right) Spatial distribution of correlations between winter precipitation and the NAO index obtained from CGCM3.1(T63) model simulations between 1870 and 1999. (left) Leading principal component of winter SLPs simulated from the CGCM3.1(T63) model under the twentieth century emissions scenario. The percentage of variance explained by the leading modes is also shown.

Figure 6.

As for Figure 2, but using climate simulations of the CGCM3.1(T63) model for the IPCC twentieth century emissions experiment.

[23] The strong agreement of results obtained using three different sources (climate reconstructions, observations and the AOGCM model) suggest that the spatial pattern of the NAO is not stable over decadal timescales, and that this causes noticeable differences in the NAO-precipitation relationships. It also points to the possible existence of different NAO types related to the location and surface extent of main pressure centers, which lead to very different climatic impacts over the European continent.

[24] The rotated T-mode PCA from the first leading modes derived from each moving-window 31-year T-mode PCA show four significant NAO patterns (Figure 7), accounting for 93.9% and 89.4% of the NAO spatial variability for observations and the CGCM3.1(T63) model. Grouping different NAO spatial patterns was carried out using the factorial loading values of each component obtained, and subsequently applying the maximum loading rule. Each year was assigned to the component with the highest loading value. This procedure has been applied in many climatic classifications elsewhere [e.g., Karl and Koscielny, 1982; Comrie and Glenn, 1998]. The patterns show that the dominant NAO configuration changes noticeably over decadal timescales. Each of the four patterns appears as the dominant NAO configuration during several time slides. In the four patterns the spatial structure, based on a pressure dipole, is clearly evident and is always associated with the NAO pattern. The dominant NAO pattern in the first two decades of the twentieth century shows the SPC displaced to the east and the NPC over the North Sea (Rotated-PC 3). This pattern also appears from the CCC-GCMII model analysis, corresponding to the spatial configuration of the Rotated-PC 4. Relationship between both patterns is strong (r = 0.71). Between 1930 and 1960 the NAO shows the NPC over Greenland and Iceland whereas the SPC shows maximum intensity around 36°N, covering a large latitudinal band between the North American eastern coastland and the West Mediterranean basin (Rotated-PC 1). This pattern is not so evident by means of the CCC-GCMII model, although maximum correlation is found with the Rotated-PC 2 (r = 0.43 for the whole North Atlantic region). The 1960 and 1970 decades show a NAO pattern in which both, the NPC and the SPC, are displaced to the South regarding previous decades. The SPC is split in two SLP centers, showing this pattern a tripole structure with the third SLP center over the Balkans. This spatial configuration is also identified from the CCC-GCMII model analysis, corresponding to the Rotated-PC 1 (r = 0.60). The last years are dominated by a NAO pattern characterized by a dipole structure but also with a clear displacement, both for the NPC and the SPC, to the East (Rotated-PC 4). This pattern is also identified from the CCC-GCMII model (Rotated-PC 3), and it represents the 20.7% of the total variance.

Figure 7.

(a) Spatial distribution of rotated principal component scores obtained from leading PCs in the SLP moving-window T-mode PCA procedure for twentieth century observations. Time series of loadings are also shown for each component to identify the periods for which the NAO pattern is representative. (b) As in Figure 7a, but using climate simulations of the CGCM3.1(T63) model for the IPCC twentieth century emissions experiment.

[25] Some studies have suggested that shifts in the position of the NAO centers might be related to the dominant sign of the NAO index. Cassou et al. [2004] suggested a significant eastward displacement or expansion toward Europe for the NAO+ climate regime compared to the NAO- regime. Peterson et al. [2003] also revealed a nonlinear dependence of the spatial pattern of the NAO on the sign of the NAO index, with shifts to the east corresponding to positive NAO index, whereas negative NAO index showed predominance to a westward displacement. Using T-mode PCA, nonsignificant differences have been found in the NAO index values corresponding to the NAO spatial pattern. Figure 8 shows the probability density functions for the NAO index corresponding to the years of each dominant NAO pattern identified in Figure 7 for the twentieth century observations and the CGCM3.1(T63) model for the IPCC twentieth century emissions experiment. The probability density functions of the NAO index in the different NAO patterns show similar shape and positive or negative values are unmarkedly presented as a function of the NAO pattern. Thus, the years in which the NAO shows a pattern with a dominant eastward shift, both for observations (Rotated PC-4) and model (Rotated PC-3), the NAO index does not show any trend toward positive and negative values. These results seem to agree with the findings of Beranová and Huth [2007] who tested whether the relationship between the NAO index and the precipitation shows differences as a function of the sign of the NAO and found a complex and non-well-defined behavior. It is not expected to have changes in the magnitude of the NAO-precipitation correlations related to trends in the NAO index when the sign of the NAO index is not related to the spatial pattern of the NAO. Using detrended NAO and precipitation series, Beranová and Huth [2007] observed that correlations are very close to the original series. They concluded that the existence of significant correlations between the NAO index and their nonstationarity do not result from the presence of trends to positive or negative phases in the NAO index. It is remarkable that similar shifting NAO patterns have been obtained using both observations and climate modeling, as this rules out changes in station network and/or inhomogeneities in the data sets as possible causes of either the nonstationary relationship between the NAO and European precipitation, or shifts in the NAO spatial patterns. Commonly, similar NAO patterns are detected in the model outputs and in observations [Osborn, 2004], but similar shifting patterns have not been previously identified. Our results confirm that the common assumption of a constant NAO spatial structure has several limitations in explaining the role of the NAO in European surface climate variability, in agreement with the findings of Beranová and Huth [2007] and Jung et al. [2003] for the second half of the twentieth century, and it may helps to explain why studies relating NAO to European climate behavior have not shown systematically strong correlations over the twentieth century [e.g., Slonosky et al., 2001; Polyakova et al., 2006;Zveryaev, 2006].

Figure 8.

Probability density functions for the NAO index corresponding to the years in which each NAO pattern identified in Figure 7 is dominant for the twentieth century observations and the CGCM3.1(T63) model for the IPCC twentieth century emissions experiment. The name of each curve corresponds to the rotated patterns in the Figure 7.

[26] Moreover, these limitations do not depend on the method used to obtain the NAO index. Figure 9 shows the temporal evolution of three different NAO indices in the twentieth century. The index based on Gibraltar SLP data was derived from Jones et al. [1997], the index based on the Azores SLP data was obtained using data from the Ponta Delgada station, and the third NAO index corresponds to the PC time series of the leading empirical orthogonal function (EOF) of SLP anomalies over the Atlantic sector (20N–80°N, 90°W–40°E). The latter two indices are available at http://www.cgd.ucar.edu/cas/jhurrell/indices.info.html#naopcdjfm. The evolution of NAO indices is rather similar whether they are obtained from pattern-based methods or station/grid point-based approaches; the data show similar nonstationary NAO-precipitation relationships. The correlation coefficient for the Gibraltar–Reykjavik NAO index and the PC index is 0.87, meanwhile with the Ponta Delgada–Reykjavic NAO index is 0.88 (1902–2000). The close relationships observed among the different NAO indices imply that the results obtained in the present study regarding the changing relationship between NAO and precipitation are not peculiar to the selected NAO index. As an example, Figure 9 shows the spatial distribution of correlations between the NAO and winter precipitation for 31-year periods centered on the years 1920, 1950, and 1985, as obtained using the Gibraltar (A), the Azores (B), and PC (C) NAO indices. The changing nature of the correlations over time is evident regardless of the NAO index employed, and the three sets of spatial patterns are similar and in all cases demonstrate the strengthening of NAO-precipitation correlations in the later decades of the twentieth century. Therefore, current NAO indices do not have the capacity to take account of the moving NAO spatial configuration and its possible impacts on surface climate. Other indicators developed to take into account the seasonality [Portis et al., 2001] and the interannual shifts in the position of the NAO centers [Hu and Wu, 2004] could be more convenient to quantify the NAO effects on the European climate.

Figure 9.

(a) Temporal evolution of three different NAO indices. GIBRALTAR shows the NAO index calculated as the difference between the monthly standardized surface pressures at Gibraltar (southwest Iberian Peninsula) and Reykjavik (southwest Iceland). AZORES was obtained from the Ponta Delgada station (the Azores), and the third NAO index corresponds to the principal component (PRINC. COMPONENT) time series of the leading PC of SLP anomalies over the Atlantic sector (20°–80°N, 90°W–40°E). (b–d) Maps of the correlation between the precipitation series and the GIBRALTAR, AZORES, and PRINC. COMPONENT NAO indices, respectively. Units are Pearson coefficients of correlation (r). The correlations were calculated for 31-year periods, of which the central years are indicated in each plot. Dotted lines enclose areas with significant correlations (p < 0.05).

[27] The present study has been focused on precipitation, nevertheless a changing relationship between the NAO and other climatic variables such as temperature [Slonosky and Yiou, 2002; Jones et al., 2003; Gimeno et al., 2003; Haylock et al., 2007], Sea Surface Temperature [Polyakova et al., 2006] and Arctic sea ice export [Hilmer and Jung, 2000; Jung and Hilmer, 2001] have been found as well. The nonstable relationships between NAO and surface climate variables, including precipitation, can be attributed to changes in the frequency of synoptic storms related to the NAO [Lu and Greatbatch, 2002]. Therefore, Jung et al. [2003] showed a higher NAO control on the occurrence of deep cyclones in Scandinavia and the Mediterranean region in agreement with the eastward change of the NAO centers recorded in the last decades of the twentieth century.

[28] More research is needed to clarify the physical causes of decadal shifts in the NAO spatial patterns. It has recently been suggested that the changing role of the NAO influence on northern hemisphere temperature may be linked to changes in solar activity [Gimeno et al., 2003], because different spatial configurations of the NAO have been obtained at solar maxima and minima [Huth et al., 2006]. Also the mean intensity of westerly flows have been proposed to explain westward and eastward shifts of the NAO action centers [Luo and Gong, 2006]. Climate change could also explain shifts in the NAO configurations, since anthropogenic forcing may cause changes in the frequencies of some natural atmospheric circulation regimes [Corti et al., 1999], and thus changes in the spatial patterns of the north hemisphere leading modes of SLP variability [Ulbrich and Christoph, 1999; Brandefelt, 2006].

[29] A failure to take account of shifts in NAO pressure centers, and the use of NAO indices based only on the sign and magnitude of the dipole, may explain why NAO trends are not key contributors in current climate simulations seeking to model predicted changes in winter precipitation over Europe and the Mediterranean region [Stephenson et al., 2006]. Moreover, climate forecasting based on NAO conditions may have limitations if such attempts are uniquely based on the sign and magnitude of NAO indices. Some current research focuses on improving European climate forecasting and NAO prediction from seasonal to multidecadal timescales (see review in work by Bojariu and Gimeno [2003]), on the basis of lagged Atlantic Ocean sea surface temperature-European climate relationships [Rodwell et al., 1999; Sutton and Hodson, 2005], but also on snow cover [Walland and Simmonds, 1997] and sea ice [Deser et al., 2000]. Nevertheless, these models aim only to predict the magnitude and sign of the winter NAO. We have also shown the need to predict the possible spatial configuration of the NAO, because this is a key factor in understanding the real effect of the NAO on European precipitation, and thus in obtaining reliable forecasting of climatic conditions. The practical implications of such predictions extend beyond the winter months, since annual water resource availability in many European regions is highly dependent on the winter NAO [López-Moreno and Vicente-Serrano, 2008].

4. Summary and Conclusions

[30] In this paper, we present evidence that the nonstationary relationship of the NAO is linked to interdecadal variability in the positions of the NAO pressure centers. The spatial configuration of the NAO changes substantially prior to marked shifts in the magnitude and spatial influences of the NAO on precipitation patterns in Europe. The reliability of the results obtained is supported by evidence from the consistency of analyses conducted with observed data during the twentieth century, reconstructed precipitation and sea level pressure since 1785, and physically modeled climate by Atmosphere-Ocean Global Climate Models (AOGCMs). Our findings suggest that climate prediction and climate change scenarios based on the NAO must take interdecadal shifts in the position of NAO pressure centers into account.

Acknowledgments

[31] We would like to thank the Tyndall Centre for Climate Change Research (UK) and Tim Mitchell for providing the precipitation database used in this study. We would also like to thank the Climate Research Unit of the University of East Anglia (UK) for providing the North Atlantic Oscillation Index, the Climatology and Meteorology Research Group of the University of Bern (Switzerland) for providing nineteenth century reconstructions, and the National Center for Atmospheric Research (U.S.) for providing the sea level pressure grids of the Northern Hemisphere. We acknowledge the modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI), and the WCRP's Working Group on Coupled Modeling (WGCM) for their roles in making available the WCRP CMIP3 multimodel data set. Support of this data set is provided by the Office of Science, U.S. Department of Energy. This work was supported by projects financed by the Spanish Commission of Science and Technology (CGL2005-04508/BOS and CGL2008-01189/BTE), the 7th framework programme of the European Commission (projects ACQWA and EUROGEOSS), and the Aragón Government.