SEARCH

SEARCH BY CITATION

Keywords:

  • European temperatures;
  • Atlantic SST;
  • SSA;
  • NAO;
  • AMO

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data
  5. 3. Methodology
  6. 4. Analysis
  7. 5. Common oscillatory modes
  8. 6. Nonlinear trend analysis
  9. 7. Discussion and concluding remarks
  10. Acknowledgements
  11. References

The relationships between the main patterns of variability of the Atlantic sea-surface temperatures (SSTs) and the European land-surface temperatures (LSTs) at interannual-to-multidecadal time scales are investigated along the period 1872–2004. Principal component analysis (PCA) is used firstly to obtain the main spatio-temporal patterns of variability of the LSTs and SSTs. Singular Spectral Analysis (SSA) is then used to decompose the time series associated with these patterns into nonlinear trends and quasi-periodic oscillations, searching for common oscillatory modes to the SSTs and LSTs. The potential predictability of the LSTs based on the SSTs is also analysed. Regarding the SSA results, three robust oscillations of periods around 13.7, 7.5 and 5.2 years, present both in the Atlantic SSTs and north-western European LSTs, were isolated. These oscillations were found to be associated mainly with a quadripolar SST pattern in the North Atlantic region, usually related to the North Atlantic Oscillation (NAO) atmospheric mode of variability. The predictability study revealed that the SSTs of the Atlantic Ocean are able to account for about 12% of the north-western European LSTs variance. Additionally, an oscillatory component with period around 3.6 years was identified, but no significant connection between SST and LST was found for this mode. In addition, at this time scale, we find that the El Niño-Southern Oscillation (ENSO) is leading the Atlantic SST quadripolar pattern by 6 months. Finally, the analysis of the nonlinear trends showed the presence of oscillations with periods around 60–100 years, both in the SSTs and LSTs. At these later time scales, our results reveal that the multidecadal behaviour of the southern European LSTs is related to Atlantic Multidecadal Oscillation (AMO) during the period 1872–1940, unlike the northern and eastern European LSTs, while, during the period 1941–2004, the AMO sign seems to be present in whole Europe. Copyright © 2010 Royal Meteorological Society


1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data
  5. 3. Methodology
  6. 4. Analysis
  7. 5. Common oscillatory modes
  8. 6. Nonlinear trend analysis
  9. 7. Discussion and concluding remarks
  10. Acknowledgements
  11. References

Because seasonal variability is the most remarkable and regular component of climatic variability in mid-latitudes, seasonal forecasting in Europe has received renewed interest in recent years (Carson, 1998; Goddard et al., 2001; Mathieu et al., 2004; Rodriguez-Fonseca et al., 2006; Gámiz-Fortis et al., 2008a,b). However, less attention has been given to nonseasonal variability, especially interannual variability, due mainly to the complexity it presents for not being as regular as seasonal variability. At present, we know that the main potential predictors for the European climate are given by the large-scale circulation patterns (Hurrell, 1995), the advection of North Atlantic sea-surface temperature (SST) anomalies (Sutton and Allen, 1997) and the memory of the European climate itself (Blender et al., 2003). The influence of these predictor variables can be viewed from the perspective of different timescales. On monthly timescales, there is strong evidence that much of the observed ocean variability is caused by atmospheric forcing (Wallace and Jiang, 1987; Cayan, 1992; Deser and Timlin, 1997; Paeth et al., 2003) whereas, on interdecadal timescales, it is well established that the ocean dynamics could be responsible for sustaining or modulating the atmospheric variability modes (Sutton and Allen, 1997; Terray and Cassou, 2002; Paeth et al., 2003).

It is worthwhile investigating quasi-periodic oscillation of surface climatic fields because periodic components are easier to predict by statistical techniques than others are, although the percentages of variance explained by the periodic oscillations are not very large in most cases. In fact, many quasi-periodic oscillation signals have been detected in atmospheric–oceanic circulation and surface climatic fields such as El Niño-Southern Oscillation (ENSO) (Walker, 1924; Bjerknes, 1969; Horel and Wallace, 1981; van Loon and Madden, 1981; Rasmusson and Carpenter, 1982; Deser and Wallace, 1987; Philander, 1990; Moron et al., 1998), the North Atlantic Oscillation (NAO) (Walker and Bliss, 1932; van Loon and Rogers, 1978; Rogers and van Loon, 1979; Wallace and Gutzler, 1981; Barnston and Livezey, 1987; Lamb and Peppler, 1987; Rogers, 1990; Tourre et al., 1999) and the Atlantic multidecadal oscillation (AMO) (Delworth and Mann, 2000; Kerr, 2000; Enfield et al., 2001; Hurrell and Folland, 2002; McCabe et al., 2004; Sutton and Hodson, 2005). The climatic effects of the NAO in the North Atlantic region have been widely studied in recent years, revealing it to be the most important source of climate variability in Europe (Hurrell, 1995; Osborn et al., 1999; Castro-Díez et al., 2002). Particularly, the interannual temperature variability in central and northern Europe is substantially induced by changes in the NAO.

Other works have speculated that variations in Atlantic SST may determine part of the long-term (decadal) variability observed in the European climate (Sutton and Allen, 1997; Rodwell et al., 1999). Evidence from observations and atmospheric model simulations suggest that the Atlantic Ocean exerts a significant influence on the winter climate of the North Atlantic/European region, influencing both interannual variability and longer timescale variability (Rodwell et al., 1999; Latif et al., 2000; Mehta et al., 2000; Robertson et al., 2000; Hoerling et al., 2001; Sutton and Hodson, 2003). Atlantic SSTs regulate surface air-pressure patterns, the strength and location of westerlies and storm tracks passing over the Atlantic. All of these have a direct impact on surface climate over Europe. The evidence suggests that the atmospheric response to variability in the Atlantic Ocean has a significant projection on the NAO pattern. Rodwell et al. (1999); Latif et al. (2000); Mehta et al. (2000); Robertson et al. (2000) and Hoerling et al. (2001) have succeeded in reproducing a fraction of low-frequency variability of the NAO during the last decades based on the SST records. Rodwell et al. (1999) attribute this oceanic influence to a positive-feedback mechanism between a tripolar pattern of SST in the North Atlantic Ocean and the NAO. This tripolar pattern has one node in the higher latitudes of the North Atlantic, a node of opposite sign off the east coast of USA, stretching across the subtropical Atlantic and a node of the same sign as in the higher latitudes in the tropical North Atlantic. The tripole appears to be driven by the NAO on a variety of time scales and, in turn, there is evidence that it may force the NAO to some extent (Rodwell et al., 1999). The sign of the association is that positive SST anomalies in the low-latitude North Atlantic are associated with a negative NAO index (according to the usual NAO sign convention).

Although, most of the research involving the analyses of SST and sea-level pressure (SLP) datasets of the Atlantic Ocean have focused on the North Atlantic region, giving little attention to the South Atlantic, there are works suggesting that the South Atlantic Ocean, both tropical and subtropical, could also play an important role (Robertson et al., 2000).

There is, therefore, an obvious need for further research to elucidate the relationships between the Atlantic Ocean and the climate, and the consequences for forecasting on timescales longer than seasonal ones. In this study, the characteristics of the spatial and temporal variations of European land-surface temperatures (LSTs) and their associations with the Atlantic SST anomalies, especially from interannual to interdecadal time scales, are examined through their spectral properties. For this purpose, singular spectral analysis (SSA) is applied to the spatial principal components (PCs) obtained from monthly anomalies of SST in the Atlantic Ocean and LST of Europe.

The present paper is devoted to identifying modes eventually present in both the Atlantic SST and European LST, at interannual and interdecadal timescales. We seek to determine whether there are preferred modes of variability, associated with certain periodicities at these scales, and whether these modes are stationary modes in time. It is important to note that the main concern in this paper is not to predict European temperature in the future but rather to identify and quantify the parts of the SST field that could be related with the European LST. Additionally, this study contributes to the current debate on ocean–atmosphere coupling in the Atlantic region by first studying the modes of variability of the entire Atlantic and European area temperatures and secondly, trying to detect connections between oceanic and land modes of variability on a wide range of temporal and spatial scales.

The work is organized as follows: Sections 2 and 3 describe the data and the methodology used, respectively. Section 4 presents the main results regarding the SST and LST modes of variability and the SSA analysis of these modes. In Section 5, a detailed description of the common oscillation modes of the LST and SST presented in Section 4 is carried out. In the same way, in Section 6, the nonlinear trends presented in Section 4 are fully described, including possible physical explanations. Finally, in Section 7, a discussion of the results and some concluding remarks are presented.

2. Data

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data
  5. 3. Methodology
  6. 4. Analysis
  7. 5. Common oscillatory modes
  8. 6. Nonlinear trend analysis
  9. 7. Discussion and concluding remarks
  10. Acknowledgements
  11. References

For the European land-surface temperatures, we used gridded monthly air-temperature data from January 1850 to April 2007, covering western, central and southern Europe, southern Scandinavia and North Africa (Longitude: 10°W–20°E; Latitude: 35°N–60°N). The data (provided by the CRU, University of East Anglia, UK) were defined on 5° latitude by 5° longitude grid-box basis and are expressed as anomalies from the corresponding monthly averages of the period 1961–1990 (CRUTEM3, Brohan et al., 2006).

The monthly mean of sea-surface temperatures over the Atlantic Ocean, between 40°S and 80°N, derived from the Hadley Centre Ice and Sea-surface Temperature dataset [HadISST1.1 (Rayner et al., 2003)] from January 1870 to December 2006, were used in this study. Missing data were excluded from the analysis. Monthly anomalies over the period 1870–2006 (total period) were constructed.

The Met Office Hadley Centre's mean SLP dataset, in an updated form using NCEP/NCAR reanalysis field, has been also used (HadSLP2r, Allan and Ansell, 2006). This is a combination of monthly globally complete fields of land- and marine-pressure observations on a 5° latitude–longitude grid from January 1850 to May 2007.

Additionally, for some specific analyses, we have used the following time-series data:

  • 1.
    The monthly index representative of the NAO, from January 1826 to October 2007, developed by Jones et al. (1997). The index was formulated using pressure data from Gibraltar (36.1N, 5.4 W) and Iceland, the latter computed mainly using data from Reykjavik (64.1N, 22.9 W).
  • 2.
    The Bivariate ENSO time series or the commonly named ‘BEST’ ENSO index (Smith and Sardeshmukh, 2000). This time series is based on combining an atmospheric component of the ENSO phenomenon (the Southern oscillation index or ‘SOI’) and an oceanic component (Nino 3.4 SST) from January 1871 to October 2007.
  • 3.
    The time series corresponding to the AMO index. This series comes from the National Oceanic and Atmospheric Administration, Physical Science Division (NOAA PSD), which uses the Kaplan SST dataset computing the area weighted average over the North Atlantic (0°N–70°N). AMO index covers the period 1856–2007. A smoothed version of the index with a 121-month smoother is used.

To study the temporal variability of the European LST in time scales greater than one year, a low-pass filter was applied to these different datasets, which lent to pass only frequencies lower that 0.08 cycles/month (period > 1 year). Although the application of a low-pass filter is not necessary in the case of the SST, because the temporal variability for scales lower than one year is not prominent for this variable, in order the analysis to be consistent, we submitted all datasets to the same treatment. In the filtering process, we lost 4 years, 2 at the beginning of the series and 2 at the end. For this reason the final common period analysed was 1872–2004.

3. Methodology

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data
  5. 3. Methodology
  6. 4. Analysis
  7. 5. Common oscillatory modes
  8. 6. Nonlinear trend analysis
  9. 7. Discussion and concluding remarks
  10. Acknowledgements
  11. References

Principal Component Analysis (PCA) was first used to reduce the data dimension of LST and SST fields. PCA has been used widely to reduce a large number of interrelated variables to a few independent PCs that capture much of the variance of the original dataset (Wilks, 1995). It produces a few major spatial-variability patterns [or empirical orthogonal functions (EOFs)], and the corresponding time series that represent the time evolution of the spatial-variation patterns. Rotation (varimax) of PCA is then applied to the first few major patterns to capture better the physically meaningful and simplified spatial patterns (Barnston and Livezey, 1987).

Then Singular Spectral Analysis (SSA) is applied to each retained rotated LST-PC and SST-PC to identify those that have quasi-oscillatory components. SSA is a powerful tool for time-series analysis that can identify intermittent oscillation spells, in short, noise time series (Vautard et al., 1992). The use of SSA technique allows to simultaneously identify oscillations with widely different time scales without commonly used prefiltering of data (Moron et al., 1998; Zhang et al., 1998). SSA is a data-adaptive method (i.e. the bandwidth and shape of the filters is not provided by the user but by the data) whose advantage is to take into account time-correlations over a wide range of time scales (Plaut and Vautard, 1994).

SSA has been successfully applied to many geophysical and climatological time series to study and predict periodic activities (Ghil and Mo, 1991; Ghil and Vautard, 1991; Plaut and Vautard, 1994; Gámiz-Fortis et al., 2002; Paluš and Novotná, 2006). SSA solves eigenvalue problems stemming from the lag-autocovariance matrix of a single time series with a pre-defined window length (Vautard et al., 1992; Elsner and Tsonis, 1996). The order election of the lagged-covariance matrix M represents a trade-off between significant information and statistical confidence. The choice of the dimension M is not obvious, but SSA is typically successful at analysing periods in the range (M/5, M). It is worth noting that SSA does not resolve periods longer than the window length M. Hence, if we want to reconstruct a strange attractor, whose spectrum includes periods of arbitrary length, the large M the better, avoiding to exceeding M = N/3 (N being the length of the data) (otherwise statistical errors could dominate the last values of the auto-covariance function). A common recommendation is to choose M = N/4 (Vautard and Ghil, 1989). The PCs produced by SSA are called temporal PCs (T-PCs) and the empirical orthogonal functions (EOFs) are the time EOFs (T-EOFs) that describe the temporal oscillatory behaviour of the T-PCs. Eigenvalues are sorted in descending order, indicating the variance of each corresponding T-PC. When two consecutive eigenvalues are nearly equal, and the corresponding T-PCs are in quadrature, these two T-PCs form a pair and potentially represent an oscillation. The time evolution of each identified oscillation component is extracted based on the reconstruction technique developed by Plaut and Vautard (1994). Each reconstructed time series (RCs) represents an isolated oscillation and the original series is exactly the sum of all the RCs. Additionally, the Maximum Entropy Method (MEM) has been used to evaluate the spectral contents of the T-PC time series, and the Monte Carlo (MC) technique was used for the significance study (see Gámiz-Fortis et al. (2002), for further details). Using this methodology, we were able not only to identify the oscillatory modes common to the Atlantic SST and European LST, but also to find the different time lags between these common isolated oscillations. There are other methodologies which are able to extract common oscillatory modes for two different fields, for instance, the multichannel SSA. In this work, we have rather preferred to use the mono-channel SSA to avoid the analysis be influenced by the different physical characteristics of the variables and datasets used in this study.

Finally, we generated multiple-linear-regression models that used some SST-RCs time series, with periodicities common to the European LST-PCs, as explanatory variables for the RC-LSTs. This modelling does not mean that cause-and-effect relationships can be established between the SST and LST fields, but rather it is a form of quantifying the strength of relationships found based on the recurrence of the common oscillatory modes along the time. For each common oscillatory mode (RC), the regression model can be written as follows:

  • equation image(1)

where RC-LSTi is the reconstructed component for the LST-PCi and RC-SSTj is the reconstructed component for the SST-PCj with the same associated periodicity. The βj parameters are the regression coefficients and quantify the importance of the explanatory factors RC-SSTj. The unknown random effects are represented by e, which is a vector of independent, normally distributed noise.

The method of ordinary least squares was used to estimate the parameters βj (Draper and Smith, 1998) by minimizing the sum of squared errors (SSE), i.e. the unexplained variance part. Here, we also used the coefficient of multiple determination R2, which measures the fraction of variance in the response variable that can be explained by variations in the explanatory factors (see Junge and Stephenson (2003) for more information). However, a high value of R2 does not imply that a particular model is appropriate. In fitting the model, the assumption is made that the noise e is a realization of an independent stochastic process with normal distribution. The validity of this assumption should be checked.

4. Analysis

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data
  5. 3. Methodology
  6. 4. Analysis
  7. 5. Common oscillatory modes
  8. 6. Nonlinear trend analysis
  9. 7. Discussion and concluding remarks
  10. Acknowledgements
  11. References

4.1. European LST variation

Spatial EOFs were calculated for the entire filtered European temperature dataset (January 1852–April 2005). Three significant spatial EOFs, accounting for 65.64% of the total variance, were found and varimax rotation was applied. Figure 1 shows the corresponding spatial patterns by drawing the isolines of the loading factors. The leading mode (Figure 1(a)) explains 30.19% of the variance. Figure 2(a) shows the corresponding PC series (LST1 hereinafter). The correlation coefficient between the LST1 and the monthly NAO index (Figure 2(d)) is 0.47 (statistically significant). Figure 1(a) shows that this mode of variability represents mainly the variability of western and central Europe, southern Scandinavia and the British Isles. Other authors have reported similar patterns as the spatial signature of the NAO on the temperature over the studied area (Hurrell, 1995; Osborn et al., 1999; Pozo-Vázquez et al., 2001a). The time series of the LST1 shows a statistically significant upward trend from 1880 to the end of the series. A more detailed discussion about the trend will be presented later in the paper. Figure 1(b) shows the spatial pattern associated with the second LST–EOF, which explains 20.38% of the total variance, and Figure 2(b) shows the corresponding PC series (LST2 hereinafter). This mode is representative mainly of the variability over south-western Europe, especially over the Iberian Peninsula (Figure 1(b)). Correlation with the NAO index is − 0.11 (statistically nonsignificant). The most pronounced trends correspond to the period 1872–1890 (downward trend) and 1915 to the end of the series (upward trend) (Figure 2(b)). Figure 1(c) shows the pattern associated with the third LST-EOF (15.35% of the total variance). This pattern is representative mainly of the variability of the south-eastern part of Europe. The correlation between the corresponding PC series (LST3 hereinafter) and the NAO index is negligible.

thumbnail image

Figure 1. Loading factors (per ten) for the (a) first, (b) second and (c) third S-EOF resulting from the PCA of the European monthly temperatures during the period 1852–2005

Download figure to PowerPoint

thumbnail image

Figure 2. PC series (thin line) corresponding to the (a) first, (b) second and (c) third S-EOF resulting from the analysis of the European monthly temperature in the period 1852–2005. The corresponding SSA-filters (nonlinear trends plus oscillatory modes, red line) and associated nonlinear trends (thick line) are also shown. Units are °C (anomalies). (d) Monthly NAO index (blue line) along with its SSA-filter (dotted line). For the sake of a fair comparison with the Atlantic SST analysis results, only the period from 1870 to 2004 is shown in the plots. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

SSA is then applied to each of these LST-PCs to further identify and extract the oscillation components. A window length of M = 460 months (38 years) and the Vautard and Ghil (1989) algorithm were used in SSA. The oscillation pairs identified by SSA on LST-PCs are summarized in Table I. Because SSA cannot resolve periods longer than the window length, we identified the zero frequency as the nonlinear trend, but it is important to note that this nonlinear trend could be composed of quasi-oscillatory modes with periods longer than 38 years.

Table I. Oscillation pairs identified by SSA on European LST-PCs
LST-PCOscillation periods in years (rank order for the associated eigenvalues; explained variance)Test: against RN process and RN + signal (RN + components)
  1. The second column contains the peak period (in years), the pair order and the fraction of total variance (%) explained by the pair. An estimation of significance level using Monte Carlo-SSA (MC-SSA) is also shown (last column).

  2. RN, red noise.

 Nonlinear trend (1–4; 22.7%)< 80% (RN + significant oscillations)
 7.5 years (2–3; 12.3%)> 95% (RN + 5–6 + 7–8 + 13–14 + 15–16)
 2.3 years (5–6; 6.0%)> 95% (RN)
LST15.2 years (7–8; 4.5%)> 95% (RN)
 13.7 years (11–12; 3.9%)> 80% (RN + 2–3 + 5–6 + 7–8 + 13–14 + 15–16)
 3.6 years (13–14; 3.8%)> 95% (RN)
 2.1 years (15–16; 2.5%)> 95% (RN)
 Nonlinear trend (1–2; 29.6%)> 80% (RN + 4–5 + 9–10)
 2.4 years (4–5; 5.8%)> 95% (RN)
LST221 years (6–7; 5.3%)< 80% (RN + 4–5 + 9–10)
 3.1 years (9–10; 3.8%)> 95% (RN)
 1.5 years (12–13; 3.4%)> 95% (RN)
 Nonlinear trend (1–2; 7.0%)< 80% (RN + significant oscillations)
 2.3 years (3–4; 6.0%)> 95% (RN)
LST33.2 years (8–9; 4.3%)> 95% (RN)
 2.1 years (11–12; 3.9%)> 95% (RN)
 7.5 years (1–2; 8.8%)> 95% (RN)
 2.3 years (3–4; 8.3%)> 95% (RN)
NAO5.9 years (5–6; 7.5%)> 95% (RN)
 2.6 years (7–8; 7.0%)> 95% (RN)
 13.7 years (9–12; 6.3%)< 80% (RN + significant oscillations)
 1.8 years (10–11; 6.4%)> 95% (RN)

Besides the nonsignificant zero-frequency mode (trend), the significant modes in the LST1 were located at the frequencies (in cycles per month) 0.011, 0.036, 0.016, 0.006, 0.023 and 0.044, corresponding to periods around 7.5, 2.3, 5.2, 13.7, 3.6 and 2.1 years, respectively. For LST2, significant modes appeared at frequencies 0.035, 0.027 and 0.055 cycles/month, corresponding to periods around 2.4, 3.1 and 1.5 years, respectively. Also an oscillation with periods of 21 years was shown by the eigenvalue spectrum, but it was not significant using the MC technique. Similar results to these latest are found for the LST3. By comparison, the significant modes in the monthly NAO index were located at the frequencies 0.011, 0.037, 0.014, 0.031, 0.055 and 0.017 cycles/month, corresponding to the periods around 7.5, 2.3, 5.9, 2.7, 1.5 and 4.9 years, respectively.

Using this methodology, we conclude that the LST1 series can be represented by a nonlinear trend that contains multi-decadal variability, a set of oscillations with associated periods around 7.5, 2.3, 5.2, 13.7, 3.6 and 2.1 years, and a red-noise process. A reconstruction of the LST1 series, using the nonlinear trend and these six quasi-oscillatory modes, called SSA-filtered LST1 series, was computed using these components and it is shown in Figure 2(a) along with the raw time series. From Table I, the variance explained by this filter is around 55%. Table II presents an evaluation of the performance of this reconstruction, and also for the LST2, LST3 and the NAO index. The raw series has a variance value of 0.71, while the SSA-filtered LST1 has a variance of 0.24. Over the period 1852–2005, the correlation between the original and the SSA-filtered series is 0.74 (significant at 95% confidence level), and the percentage of phase accordance is 77% (see first column in Table II). Note that the variance explained by the SSA filter is 43% for the LST2 and 38% for the NAO (Tables I and II). However, while most of the oscillatory modes found for the NAO are common to the LST1, most of the explained variance for the LST2 case is due to the nonlinear trend (around 30%). For the LST3 series, explained variance by the SSA filter is only 21%.

Table II. Statistical results for the SSA modelling
 LST1 vs SSA-filtered LST1LST2 vs SSA-filtered LST2LST3 vs SSA-filtered LST3NAO vs SSA-filtered NAO
  • *

    Statistically significant at the 95% confidence level.

 Period 1852–2005Period 1852–2005Period 1852–2005Period 1827–2005
MSE0.380.270.460.25
MAE0.490.410.540.39
Correlation coefficient0.74*0.66*0.49*0.66*
% phase accordance77766873
% variance explained by the SSA filter55432138

Common oscillations in north-western Europe and in the NAO index (around 7.5, 2.3–2.6 and 5.2–5.9 year-periods) are very likely to be associated with the same source, with the SLP field leading the land-surface temperature by 0–1 months. In this sense, the atmospheric SLP field is considered a better specifier of European temperatures on the interannual time scale than the oceanic field (Junge and Stephenson, 2003).

4.2. Atlantic SST variation

In this section, we present the results of applying a rotated PCA to the Atlantic monthly SST. Results reveal five significant modes of variability. Figures 3 and 4 show the spatial loading patterns relative to this field and the associated PCs, respectively. The combined variance associated with these five patterns accounts for circa 52% of the total variance. The first mode (Figure 3(a)), which explains 17% of the variance, represents the ocean-temperature variability in the tropical South Atlantic. The highest loading factors are found between the equator and 30°S latitude. This equatorial mode is called the Atlantic Niño and shows strongest amplitude during May–July (Deser et al., 2010). The associated PC series (SST1 hereinafter, Figure 4(a)) presents some short periods with trends and considerable variability. The second SST-EOF (Figure 3(b)) explains similar variance (16%) and shows a quadripolar pattern in the North Atlantic section, having positive loading factors in the high latitudes around 60°N, extending from the subpolar gyre to the Labrador Sea, and around 20°N in the tropical area; and two negative loading centres around 35°N and over the North Sea. This pattern includes what is commonly called North Atlantic horseshoe pattern, described e.g. in Czaja and Frankignoul (2002). The associated temporal series (SST2, Figure 4(b)) presents a correlation coefficient of − 0.41 with the monthly NAO index, and shows some periods dominated by negative values (1901–1925 and 1970–1990), while an intermediate period (1930–1970) is dominated by positive values. The third SST-EOF (9% of explained variance) is shown in Figure 3(c), and presents a quadripolar spatial pattern in the North Atlantic area, with centres of action southward (and with opposite signs) to those described for the second mode, possibly partly due to a propagation effect. Particularly, two negative anomaly centres are found to the south of Greenland and around 15°N latitude, while the positive anomaly centres are located about 30°N latitude and over the North Sea. The corresponding temporal series (SST3, Figure 4(c)) presents a considerable upward trend from 1920 to the beginning of the 1950s. The fourth SST-EOF (9% of explained variance) is shown in Figure 3(d), and can be associated with the south-western Atlantic area (near Brazil) variability, in opposition to the Gulf of Guinea and the North Atlantic, near Greenland. The slope of the associated temporal series (SST4, Figure 4(d)) presents a steep upward trend from the middle of the 1960s to the beginning of the 1980s. From 1970 onwards, positive anomalies can be found most of the time. Finally, the fifth mode (Figure 3(e), 6.5% of explained variance) shows an extensive area of high loading factors in the central North Atlantic region. The main feature of the associated temporal series (SST5, Figure 4(e)) is a notably interdecadal variability between 1930 and 1960 and at the end of the record.

thumbnail image

Figure 3. Loadings factors (per ten) for the (a) first, (b) second, (c) third, (d) fourth and (e) fifth S-EOF resulting of the Atlantic monthly sea-surface temperature PCA for the 1872–2004 period. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

thumbnail image

Figure 4. PC series (thin line) corresponding to the (a) first, (b) second, (c) third, (d) fourth and (e) fifth S-EOF resulting from the analysis of the Atlantic monthly sea-surface temperature in the period 1872–2004. The corresponding SSA-filters (dotted line) and associated nonlinear trends (thick line) are also shown. Units are °C (anomalies). This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

Using a similar methodology as the previous section (M = 460 months and Vautard and Ghil algorithm), SSA is applied to each of these five SST-PCs to further identify and extract the oscillation components. The oscillation pairs identified by SSA on SST-PCs are summarized in Table III. In all the cases, the first eigenvalues are associated with the nonlinear trend components. These trends are the most important components because they explain the highest fraction of the total variance for each SST-PC, but its statistical significance is doubtful due to the record lengths. Oscillations with similar periodicities to those found in the LST1 appear for the 13.7 years peak (SST-PCs 1, 2, 3 and 4), 7.5 years (SST-PCs 3, 4 and 5), around 5.4 years (SST-PCs 1, 2, 3 and 5), 3.6 years (that appears for all the SST-PCs), and around 2.3–2.7 years (SST-PCs 1 and 4).

Table III. As in Table I but for the Atlantic SST-PCs
SST-PCOscillation periods in years (rank order for the associated eigenvalues; explained variance)Test: against RN process and red noise + signal (RN + components)
  1. RN, red noise.

 Nonlinear trend (1–2; 33%)< 80% (RN + significant oscillations)
 13.7 year (3–4; 9.4%)> 90% (RN + 6–7 + 9–10 + 11–12 + 13–14 + 16–17)
 4.9 years (6–7; 6.4%)> 95% (RN)
SST12.3 years (9–10; 4.3%)> 95% (RN)
 3.6 years (11–12; 4%)> 95% (RN)
 2.7 years (13–14; 3.2%)> 95% (RN)
 5.5 years (16–17; 2.6%)> 95% (RN)
 64 years (1–2; 20.6%)< 80% (RN + significant oscillations)
 9.3 years (3–4; 12.7%)> 95% (RN)
SST216.6 years (5–6; 8.9%)< 80% (RN + significant oscillations)
 12.1 years (7–8; 7.2%)> 90% (RN + 3–4 + 11–12 + 13–14)
 3.6 years (11–12; 4.6%)> 95% (RN)
 5.2 years (13–14; 4.3%)> 95% (RN)
 Nonlinear trend (1–2; 32.8%)< 80% (RN + significant oscillations)
 13.7 years (4–5; 9.2%)> 95% (RN + 6–7 + 9–10 + 11–12 + 13–14)
SST34.5 years (6–7; 6.7%)> 95% (RN)
 3.6 years (9–10; 4.2%)> 95% (RN)
 5.2 years (11–12; 3.9%)> 95% (RN)
 7.5 years (13–14; 3.5%)> 95% (RN)
 Nonlinear trend (1–3; 50%)< 80% (RN + significant oscillations)
 21 years (4–5; 8.4%)< 80% (RN + significant oscillations)
SST412.1 years (6–7; 6.4%)> 90% (RN + 8–9 + 12–13 + 16–17)
 3.6 years (8–9; 3.9%)> 95% (RN)
 7.5 years (11–14; 2%)> 80% (RN + 8–9 + 12–13 + 16–17)
 4.4 years (12–13; 2.5%)> 95% (RN)
 2.7 years (16–17; 2.2%)> 95% (RN)
 Nonlinear trend (1–2; 19.4%)< 80% (RN + significant oscillations)
 21 years (3–4, 8.9%)< 80% (RN + significant oscillations)
SST59.3 years (5–6; 6.9%)> 90% (RN + 11–12 + 19–20)
 7.5 years (8–9; 5.8%)> 90% (RN + 11–12 + 19–20)
 5.9 years (11–12; 5%)> 95% (RN)
 3.6 years (19–20; 2.8%)> 95% (RN)

4.3. Connection between European LST and Atlantic SST

As this paper seeks to determine which modes in the Atlantic SST contain relevant information for European temperatures, in this section, we study the relationship of the quasi-oscillatory modes found in the European LST and the Atlantic SST.

Cross-correlation maps depicting the correlation coefficients between grid points of SSTs and the time series of LST-PCs are reproduced in Figure 5. The first two LST variation patterns, LST1 and LST2, are found to have a strong statistically significant correlation with Atlantic SSTs. Particularly, LST1 is positively correlated with the SST anomalies over the North Sea and the south of Newfoundland, and negatively correlated with tropical North Atlantic and southern Greenland. This pattern resembles the former pattern associated with the SST3 mode (Figure 3(c)). The LST2 is strongly correlated with the pattern represented fundamentally by SST2 (Figure 3(b)), although the most important correlations are found in the region close to the Iberian Peninsula and the Mediterranean Sea. For the LST3 case correlation values are lower, with the highest values in the eastern Mediterranean Sea.

thumbnail image

Figure 5. Contemporaneous correlation coefficient maps between grid points of Atlantic SST and the time series of, (a) LST1, (b) LST2 and (c) LST3. Contour lines indicate statistically significant areas at 95% level. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

For a meaningful evaluation of the relationship between the SST and LST fields on interannual to decadal time scales, the trend components of the datasets need to be removed. The fact of eliminating the trend when we work with variables of different physical natures is necessary because, otherwise, the oscillatory modes remain largely disguised by this trend. At this point, we removed the linear trend at each grid point of the standardized SST field. When the correlation is calculated between detrended SST and the reconstructed oscillation components obtained from the previous LST-PCA, the resulting patterns are shown in Figure 6 for LST1. We see that (1) the association remains strong when the components of 13.7 and 5.2 years of LST1 are examined, showing similar Atlantic regions with significant correlations; (2) the component of around 7.5 years of LST1 also shows some significant areas correlated with Atlantic SSTs, but they are smaller than when we examine the association with the previous peaks and seem to be shifted to the central North Atlantic. Additionally, some contribution to this mode seems to come from the South Atlantic Ocean; (3) low correlation or nonsignificant correlation (not shown) with the Atlantic SST is found for those components with periods around 3.6 and 2.3 years, respectively. Similarly, for the oscillatory modes coming from LST2 and LST3 (not shown), the interannual component, with a period of around 3 years, shows a very low association with the Atlantic SST, while the 2.3-year oscillation does not show significant correlation. Due to the weak correlation found between the LST2 quasi-oscillatory modes and the SST, it might be thought that the previous strong correlation found between the LST2 and the SST is basically provided by the nonlinear trend of SST2. This point will be treated later in this work. Based on these results, only the common oscillatory modes of LST1 and their association with the SST are analysed in the following sections.

thumbnail image

Figure 6. Contemporaneous correlation coefficient maps computed based on the Atlantic detrended SSTs and, (a) the 7.5 years component, (b) 5.2 year component and (c) 13.7 year mode resulting from the SSA analysis of the LST1. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

5. Common oscillatory modes

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data
  5. 3. Methodology
  6. 4. Analysis
  7. 5. Common oscillatory modes
  8. 6. Nonlinear trend analysis
  9. 7. Discussion and concluding remarks
  10. Acknowledgements
  11. References

5.1. Quasi-decadal oscillation of LST1

The 7.5-year oscillation explains an important part of the variance of the LST1 (Table I). We denote this mode as quasi-decadal mode, hereafter (QD). Oscillations with a period of around 7.5 years are found for SST3, 4 and 5 (Table III). A significant oscillatory mode with roughly a 7.5-year period is also found in the monthly NAO index series (Table I). Figure 7(a) shows the reconstructed components associated with this oscillation for the LST1, SST3 (which is the SST-PC where this oscillatory mode is more stable) and the NAO. The most striking feature of this mode is its important increase in the amplitude for the associated RCs along the last 40 years of the record, from 1960 to 2004. An additional correlation analysis between this oscillation and the QD mode in the NAO shows that NAO leads the LST by 1–2 months for the years prior to 1923. From 1923 onward, however, the LST leads the NAO, just when the 7.5-year oscillation in the LST1 shows the highest amplitude. Figure 7(b) shows the cross-correlation, for different lags, between the 7.5-year mode in the LST1 and the common mode in the SST3, 4 and 5. This lagged correlation analysis indicates that the coupling is strongest when LST leads SST-PCs by a few months. This result is suggesting that the land is reacting more rapidly than the ocean to a common atmospheric cause, and, taking into account that SST variations are primarily determined by atmospheric flow patterns, we might conclude that these lags indicate an air-to-sea forcing. Although, further analysis would be required to investigate how or if these SST anomalies feedback on the atmospheric circulation, leading to a final weakening or intensification of European land-surface temperatures, the simple appearance in both fields (SST and LST) of a common periodicity is able to establish the association between a particular state of the SST and the LST at this time scale, and the most important thing, these states recur along the time. Taking this into account, the appearance of different regions and variables with similar oscillatory modes peaking at different lags could be used to quantify the association between them.

thumbnail image

Figure 7. (a) Reconstructed components associated with the QD mode (7.5-year periodicity) for the LST1 (solid line), SST3 (dashed line) and NAO index (dotted line). (b) Lead-lag cross-correlation of QD oscillation from LST1 with QD oscillation from SST3 (dashed line), SST4 (dotted line) and SST5 (solid line). Negative lags indicate that SST leads LST, while positive lags indicate SST lags LST

Download figure to PowerPoint

As all the selected RCs are oscillatory modes with the QD period, substantial opposite-sign lag correlations can be expected when shifting the lag by approximately half a period (Figure 7(b), negative lags). For this case, the maximum correlation coefficient is − 0.89, when the QD mode of SST3 leads the QD mode of LST1 by 37 months, while it is − 0.49 when the SST5 leads by 29 months. Similar lags to the SST3 are found for the SST4, but in this case, the correlation is lower (∼− 0.52). To test the stability of these estimates, the correlation coefficients are re-estimated for the two time periods 1872–1959 and 1960–2004. The reason for this particular choice lies in the high amplitude of the LST1 QD mode detected from 1960 onward. Similar correlation coefficients are found at similar lags for both periods.

In Table IV, a summary of the analysis results of the relationship between the LST1 and SST modes is presented. Particularly, for the different common LST1 and SST modes, the lag of the maximum correlation between the apparition of SST anomalies and the later manifestation of LST anomalies, this maximum correlation and the estimated regression parameter (βj) of the regression model of the LST1 modes that uses SST modes as explanatory variables, are presented. Additionally, the determination coefficient of the multiple regression analysis along with the standard deviation of the residual series is also presented. As all the standardized oscillatory modes have a variance of unity, the magnitudes of the estimated parameters βj give an indication of the relative importance of the different SST modes in association with the LST modes variability. For the QD oscillation of LST1, the relative contribution to the model for the SST3 is 0.90, while for the SST4 and SST5 is only 0.06 and 0.1, respectively. The fact that some SST-PCs play a minor role in the regression model, even though these present high correlation coefficients with the LST1, is due to the high mutual correlation shown between the same quasi-oscillatory modes from different SST-PCs. This effect leads to that only some of the factors remain important in the combined regression model.

Table IV. Summary of the regression analysis results
 QD-SST1QD-SST2QD-SST3QD-SST4QD-SST5R2σe
  • The four LST1 modes of variability are regressed based on different SST modes of variability at different time lags. Triplets show (lag, r, βj), indicating the lag (in months) with maximum correlation coefficient when the SST is leading the LST, the value of this maximum correlation (r) and the regression coefficient (βj). The last two columns show the determination coefficient of a multiple regression analysis that uses all the SST modes and the standard deviation of the resulting residual time series.

  • *

    Statistically significant at the 95% confidence level.

QD-LST1(37, − 0.88*, − 0.90*)(37, − 0.52*, − 0.06)(29, − 0.49*, − 0.1)0.770.48
 ID-SST1ID-SST2ID-SST3ID-SST4ID-SST5  
ID-LST1(59, − 0.86*, − 0.52*)(73, 0.78*, 0.12*)(63, − 0.69*, − 0.32*)(54, 0.81*, 0.07)0.870.40
 IA1-SST1IA1-SST2IA1-SST3IA1-SST4IA1-SST5  
IA1-LST1(5, − 0.34*, − 0.07)(17, 0.45*, 0.23*)(31, − 0.62*, − 0.50*)(20, − 0.22*, 0.08)0.470.70
 IA2-SST1IA2-SST2IA2-SST3IA2-SST4IA2-SST5  
IA2-LST1(8, 0.18*, 0.08)(7, 0.18*, 0.09)(8, − 0.24*, − 0.02)(13, 0.08, 0.01)0.030.98

Following the above result, we can consider the SST3 to be the most important one related to the 7.5-year oscillation of LST1, suggesting that the North Atlantic SST field contains important information associated to the temperatures over north-western Europe at this time scale (R2 = 0.77). Additionally, the assumption of a Gaussian distribution in the error term was checked by analysing the empirical quantiles of its distribution against the theoretical quantiles of the standard normal distribution (not shown). This analysis shows some deviation of the noise term from the normal distribution, suggesting that the SST do not provide a good model of north-western European temperatures at this time scale, and other variables like SLP must be considered for a good fit. In fact, the contribution of the NAO to this regression model (taking into account that the NAO index leads the QD mode of LST1 by 1 month) improves the skill of the new model, being now 0.87, and the correlation coefficient between the LST1 QD mode and the modelled time series is 0.91.

The space-time behaviour of SST and SLP fields associated with the 7.5-year oscillation of LST1 was studied by a regression analysis of the LST1 reconstructed oscillation on SST and SLP, shown through eight equally spaced phases, with negative lags indicating that SST/SLP lead high positive anomaly temperatures in the LST1-associated region, while positive lags signify that LST leads. This plot presented in the Figure 8 depicts the evolution of SST and SLP anomalies over a life cycle (eight phases) for the reconstructed LST1 7.5-year oscillation. Concerning the spatial patterns, the 7.5-year wave presents atmospheric and oceanic patterns that are very close to the associated patterns linked to the familiar NAO dipole (Hurrell and van Loon, 1997). In this context the 7.5-year oscillation presents coupled variability. Together with this SLP pattern appears the familiar tripole of SST anomalies (e.g. Tourre et al., 1999). The main features of the SLP pattern are an increase in the westerly wind at about 50°N and an anomalous anticyclonic circulation centred at 40°N–35°W. When the NAO is most intense and intensified westerly is presented (Figure 8, lag 0), coinciding with a positive SST lobe off Cape Hatteras and negative lobe off Newfoundland, which are associated with the positive phase of the NAO. This is consistent with the fact that to the north of the anticyclone, where the westerly wind is stronger, the ocean surface is cooler than normal, while to the west of the anticyclone, where the southerly winds are stronger, the ocean is substantially warmed (Zorita et al., 1992). These intensified western SST patterns are not static but rather propagate along the North Atlantic current region. Below 20°N, weak SST anomalies of the same sign as the one off Newfoundland extend over most of the basin, except in the south-western part of the ocean. How this atmospheric–oceanic structure emerges is a currently debated question (da Costa and Colin de Verdiere, 2002; Paluš and Novotná, 2004, 2008).

thumbnail image

Figure 8. Linear-regression coefficients of the detrended monthly mean SSTs and SLPs anomalies onto the standardized time series of the LST1 QD oscillation. Negative lags indicate that SST/SLP leads. Positive lags indicate LST leads. Units are anomalies in °C (colours) and contours show anomalies in hPa. Contour interval is 0.05 hPa. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

Regarding the association of this mode with the Atlantic SST, in agreement with the previous Figure 7(b), maximum SST anomalies are found (at the lag ∼− 37 and 12 months) in the SST3-associated region. Particularly, Figure 8 shows that the middle and upper latitudinal lobes of SST3 present the highest anomaly values, while the tropical part presents very low anomaly values. Additionally, in agreement with the previous result (Figure 7(b)), some smaller contribution can be seen in the South Atlantic (SST4).

5.2. Interdecadal oscillation of LST1

We use the term interdecadal oscillation (ID, hereinafter) to refer to the quasi-oscillatory mode with period around 13.7 years found in the LST1. Modes with similar periods (12–13.7 years) are present in all the SST-PCs but for the SST5, where the closest peak found has a period of around 9 years (Table III). Figure 9(a) shows the reconstructed components associated with this mode for LST1, SST1, SST2 and SST4. ID mode for SST3 (not shown) is basically in a phase opposite to that of SST2. Note (Figure 9(a)) that the frequency of this mode is very stable prior to 1970, but this stability breaks down after 1970 for the LST1 case. Also, it is noteworthy that this LST1 oscillation presents the highest amplitude values in the period around 1920–1960, where the QD oscillation of the LST1 and NAO have their lower amplitude values. Just considering the period 1872–1970, the maximum correlation coefficients between the LST1 and SST-PCs ID modes are plotted in Figure 9(b) and a summary is given in Table IV. High correlations are found for all the SST-PC ID modes at lags between 6 and 20 and between − 54 and − 73 months, depending on the associated region. Positive lags are suggesting again the presence of an atmosphere-to-sea forcing, while the others (between − 54 and − 73 months) could be a reflection of the ID periodicity that exist in all the time series. A multiple-linear-regression model considering these lags (Table IV) shows that the Atlantic regions, where the association with the LST1 ID mode is maxima, come from SST1 (0.52) and SST3 (0.32). A smaller contribution from SST2 (0.12) is also found. With regard to the skill of the model, R2 is 0.87 and the standard deviation of the error is 0.40. The analysis of the former results suggests that the SST ID mode is the mode that provides the highest association with the LST1 ID time series.

thumbnail image

Figure 9. (a) Reconstructed components associated with the ID mode (13.7-year periodicity) for the LST1 (solid line), SST1 (short dashed line), SST2 (dotted line) and SST4 (large dashed line). (b) Lead-lag cross-correlation of ID oscillation from LST1 with ID oscillation from SST1 (short dashed line), SST2 (dotted line), SST3 (solid line) and SST4 (large dashed line) computed from the period 1872–1970. Negative lags indicate SST leads LST, while positive lags indicate SST lags LST

Download figure to PowerPoint

Figure 10 shows the joint spatial patterns of SST and SLP at the considered frequency. The evolution of the spatial patterns through their cycles is shown in eight panels obtained in a similar way to that of the QD oscillation, but now considering lags every 20 months. Maximum SST anomalies are found for the regions associated to the SST1 and SST3, with positive anomalies of SST to the south of Greenland and in the tropical North Atlantic, and negative anomalies in the Gulf Stream (coinciding with negative anomalies in the tropical South Atlantic), which are associated to the appearance of positive anomalies of land-surface temperature in north-western Europe around 60 months later (Figure 10, lag − 60).

thumbnail image

Figure 10. The same as Figure 8 but for the ID oscillatory mode (13.7-year periodicity) of LST1. Time lag between each map is 20 months. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

The resulting regression maps show the propagation of the surface thermal anomalies in the North Atlantic. Warm SST in the storm-formation region is associated with cold SST to the south of Greenland and in the tropical North Atlantic (Figure 10, lag 0). The warm anomaly in the storm-formation region moves eastwards and then appears to split into two branches over the Mid-Atlantic Ridge, in agreement with Sutton and Allen (1997) and Dima and Lohmann (2004) (Figure 10, lags 20, 40 and 60, successively). At lag − 80 (the half of the cycle), we find a 180° out of phase pattern with that seen at lag zero, turning the cycle into its opposite phase.

Also, it is important to note that the SLP patterns associated with this ID mode reflect this periodicity that exhibits opposite phases at an 80-month lag. These results appear to indicate that the oceanic advection and atmospheric teleconnections may combine to determine this cycle in the north-western European LST.

5.3. First interannual oscillation of LST1

The term first interannual oscillation is used here to indicate the quasi-oscillatory mode with a period of around 5.2 years found in the LST1 (IA1, hereafter). Oscillations with periods of around 5.2 years are clearly detected using SSA for the SST-PCs 1, 2, 3 and 5 (Table III). For clarity, Figure 11(a) shows only the reconstructed components associated with this mode in the SST3, where this oscillation presents higher amplitudes. A comparison of the similar mode derived from the NAO index is also shown. Note that the IA1 mode is, in general, less stable than the QD and ID ones for the LST1, with periods 1872–1910 and 1935–1965 showing the maximum amplitudes. For the case of the NAO, this oscillatory mode shows very low amplitude. Figure 11(b) shows the lagged cross-correlation between the IA1 mode in the LST1 and the common modes in the SST-PCs and NAO. Maximum correlation coefficients are − 0.34 when the SST1 leads the LST1 by 5 months, 0.45 when the SST2 leads by 17 months, − 0.62 when the SST3 leads by 31 months and − 0.22 when the SST5 leads by 20 months. When we fit a linear-regression model, considering these lags, the contribution to the model is only around 0.07–0.08 for SST1 and SST5 (statistically nonsignificant), while it is 0.23 for SST2 and 0.50 for SST3 (Table IV). Considering only the reconstructed oscillations with period around 5.2 years of SST2 and SST3 as associated factors, the skill of the multiple-linear-regression model for the IA1 mode of LST1 is 0.47 (Table IV).

thumbnail image

Figure 11. (a) Reconstructed components associated with the IA1 mode (5.2-year periodicity) for the LST1 (solid line), SST3 (short dashed line), and NAO index (dotted line). (b) Lead-lag cross-correlation of IA1 oscillation from LST1 with IA1 oscillations from SST1 (solid line), SST2 (large and short dashed line), SST3 (short dashed line), SST5 (large dashed line) and NAO (dotted line) computed for the period 1872–2004. Negative lags indicate that SST leads LST, while positive lags indicate that SST lags LST

Download figure to PowerPoint

An analysis of the cross-correlation of the LST1 IA1 mode with the 5.9 years mode of the NAO index shows that the NAO leads the LST1 by 3 months (correlation coefficient is 0.35). However, the contribution of this mode in the NAO to the multiple-linear-regression model slightly improves the skill (R2 = 0.51).

The space-time behaviour of SST and SLP fields associated with the 5.2 years oscillation of LST1, obtained in a similar way to that of the QD and ID oscillations (not shown), indicates that for the SST case, at lag 0, the spatial pattern associated to the IA1 mode is very similar to that associated with the ID mode (Figure 10). Anomalous values show the same order of magnitude in both cases. Some differences can be found in the south-western Atlantic (SST4), where the SST does not show any association with the LST1 IA1 mode (Table IV). Additionally, the eastward propagation of SST anomalies from the Gulf Stream appears to be more intense than for the ID oscillation. For this oscillatory component, the SLP patterns show a cyclic behaviour that reflects this oscillatory mode of the NAO.

5.4. Second interannual oscillation of LST1

For the quasi-periodic oscillation with period around 3.6 years, we use the term second interannual oscillation of LST1 (IA2, hereafter). Oscillatory modes with similar periods are found in all the SST-PCs. However, cross-correlation plots (not shown) of these reconstructed components with the IA2 mode of LST1 show very low values (<0.20), and this is reflected in the low skill of the adjusted multiple-linear-regression model (R2 = 0.03) using the IA2 modes of SST-PCs. A summary of these results can be found in the last row of Table IV. This means that the association (linear) of the Atlantic SST with this LST1 component is very low. However, the monthly NAO index does not show a significant power spectrum for this period. As it is well known, the ENSO phenomenon has a distinctive oscillation of around 4 years. In an additional analysis, the SSA was applied to the monthly ENSO index. Results show a very clear oscillation with a period of around 3.6 years and accounting for 12% of the total variance of the series. However, although the correlation between this 3.6 years oscillatory mode of the ENSO and the 3.6 years oscillatory mode of the Atlantic SST-PCs is very high (r∼0.90 for SST2 when the ENSO leads the Atlantic SST by around 6 months) the lagged cross-correlation coefficients with the LST1 IA2 reconstructed components are low (∼0.17). This result suggests that the ENSO could be associated with the appearance of the 3.6-year oscillation in the Atlantic SST but a more complex mechanism must be taken into account in the propagation of this oscillation to the north-western European temperatures. Another important factor to take into account is that this quasi-oscillatory mode explains a small part of the variance of the Atlantic SST field.

5.5. Model fits of LST1

Figure 12 shows the modelled time series (Table IV) for three of the common quasi-oscillatory modes (QD, ID and IA1) of the LST1, based on the associated SST regions. Because of the low skill, the IA2 mode model is not shown. The mode which presents highest association with the Atlantic SST is the QD one, with a correlation coefficient of 0.88 between the observed time series and the modelled one, followed by the ID mode (r = 0.82). Periods at the beginning and at the end appear to have a worse association, showing not only different amplitudes but also a time lag. In agreement with the previous results (lower skill of the multiple-linear-regression model), the IA1 mode presents a lower correlation coefficient with the estimates (r = 0.68), especially during the periods when this oscillatory component shows instabilities (1910–1935 and 1970 onward). Finally, Figure 12(d) shows the sum of the three reconstructed oscillations (QD + ID + IA1) for the LST1 adding, additionally, the estimates corresponding to the regression models were obtained based on the Atlantic SSTs. For comparison purposes, the SSA-filtered detrended LST1 time series is also shown. The correlation between these two series is 0.85. We find that from 1970 onwards, the contribution of the common oscillations comes basically from the QD oscillation. Differences between this last series and the (QD + ID + IA1) one are due principally to the two additional components considered significant in the SSA-filter, the modes with periods of around 3.6 and 2–2.3 years. A summary of the model fits can be found in Table V. Although SST is able to explain a high percentage of variance for some individual oscillatory modes (QD, ID and IA1), when we consider the total contribution of these three modes to the raw detrended LST1, the contribution is only 12% of total variance, reaching the percentage of phase accordance 64%. Association of these SST modes with the north-western European temperatures would be, likely, more important when we consider seasonal data.

thumbnail image

Figure 12. Observed values (solid lines) and modelled (dashed line), using multiple-linear-regression model based on the Atlantic SST-PCs oscillations, for different quasi-oscillatory modes found in the LST1. Particularly, (a) the model and observation of the QD mode of the LST1, (b) the ID and (c) the IA1 mode. (d) Sum of the three reconstructed components (QD + ID + IA1), both fitted (dashed line) and observed (solid line). The SSA-filtered detrended LST1 time series (dotted line) is also shown

Download figure to PowerPoint

Table V. Statistical results for the multiple linear regression modelling summarized in Table IV for the period 1872–2004
 QD-LST1 vs modelledID-LST1 vs modelledIA1-LST1 vs modelledIA2- LST1 vs modelledSSA-filtered detrended LST1 vs modelledRaw detrended LST1 vs modelled
  • *

    Statistically significant at the 95% confidence level.

MSE0.010.0070.0050.020.070.56
MAE0.080.060.060.100.210.60
Correlation coefficient0.88*0.82*0.68*0.18*0.65*0.34*
% phase accordance878169567764
% variance explained by the model77674644212

6. Nonlinear trend analysis

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data
  5. 3. Methodology
  6. 4. Analysis
  7. 5. Common oscillatory modes
  8. 6. Nonlinear trend analysis
  9. 7. Discussion and concluding remarks
  10. Acknowledgements
  11. References

An exhaustive analysis of the zero frequency terms found in the previous sections (for both LST and SST PCs) shows that they come from the contribution of two terms: a warming linear trend, may be associated with the anthropogenic effect of increased greenhouse-gas concentrations and a quasi-oscillatory mode with a period of around 60–100 years, depending on the PC considered (MD, hereinafter). Figure 13 presents the trend component, along with a linear fit, for the different LST and SST modes of variability. Note that the SST2 mode (Figure 13(e)) does not present the linear trend component. However, this linear trend is more pronounced for the LST field, especially for the LST1 and LST2, while for the Atlantic SST, this term is significant mainly for SST1 and SST4. For SST5, linear trend and MD oscillatory component show very low amplitude. However, the most stable oscillatory mode with period around 64 years is found for the SST2 and LST2. In the Atlantic Ocean, the variation at multidecadal time scales has been termed as AMO (e.g. Kerr, 2000), corresponding to a quasi-periodic cycle of the North Atlantic SST. We have used the AMO index, calculated by averaging the annual mean SST observations over the regions 0°N–60°N, 75°W–7.5°W, to analyse the relationship between the AMO mode of variability and the SST MD modes we have found based on the SSA analysis. Figure 13(e) shows the time series associated with this index. From Figure 13, we can see that MD time series all seem rather similar in the second half of the period (1941–2004), with evident differences in the first half (1872–1940). A direct comparison from the AMO index and the multidecadal oscillatory modes of the LST1, LST2 and LST3 can be seen in Figure 13(i). A more detailed study using these two sub-periods has been carried out to provide significant differences between sub-periods. Figures 14 and 15 show the spatial patterns of SST and SLP variations associated with the multidecadal oscillations of (1) LST1, (2) LST2, (3) LST3 and (4) AMO index, for the sub-periods 1872–1940 and 1941–2004, respectively. Shown are the regression coefficients ( °C per SD in colours and hPa per SD in contours) obtained by regressing the SST and SLP data on a normalized version of the reconstructed MD components and AMO index. Prior to the analysis, the linear trend is removed and a 25-year running mean filter is applied for both SST and SLP fields. Results are summarized in Tables VI and VII for each sub-period. Highly significant positive correlations are found between the AMO index and the multidecadal oscillatory components of North Atlantic SST-PCs, with values of 0.94, 0.84 and 0.57 for MD-SST2, MD-SST3 and MD-SST5, respectively, for the first half of the period (see last column in Table VI); and 0.88, 0.76 and 0.92, respectively, for the second half of the period (see last column in Table VII). These values agree with the spatial pattern of North Atlantic sea-surface temperatures that characterize the AMO during the period 1871–2003 studied by Sutton and Hodson (2005). Particularly, the patterns (Figures 14(d) and 15(d)) show SST anomalies of the same sign over the whole North Atlantic, with the largest anomalies found just east of Newfoundland. The associated spatial pattern of SLP shows a North–South dipole with one centre of action to the south-east of Greenland and the second centre over the Azores. Low correlation at lag 0 is found between AMO index and MD-SST1 (r = 0.47 and − 0.59 for the two periods, respectively) associated with the South Atlantic, being the correlation coefficient only 0.20 for MD-SST4 during the 1872–1940 period. In the second half of the period, the correlation between the MD-SST4 and AMO index increases to − 0.72, associated with the northern centre of the SST4 dipole.

thumbnail image

Figure 13. (a–c) Nonlinear trends (solid lines), obtained based on the SSA, of the three LST-PCs series and (d–h) the five SST-PCs. A trend estimation, computed using the least-squared method (dotted lines), and the multidecadal oscillations of periods in the range 60–100 years (dashed lines) are also shown. For the SST2 PC series case, (e), the Atlantic multidecadal oscillation (AMO) index (solid line) is also shown. Trend component for this case was nonsignificant and was not plotted. (i) Comparison of AMO index (solid line) and MD oscillatory modes of LST1 (short dashed line), LST2 (dotted line) and LST3 (large dashed line). Units are temperature anomalies in °C

Download figure to PowerPoint

thumbnail image

Figure 14. Spatial patterns of SST and SLP variations associated with the multidecadal oscillations of (a) LST1, (b) LST2, (c) LST3 and (d) AMO index for the sub-period 1872–1940. Shown are the regression coefficients ( °C per SD in colours and hPa per SD in contours) obtained by regressing the SST and SLP data on a normalized (unit variance) version of the reconstructed components and AMO index. Prior to the analysis, the linear trend is removed and a 25-year running mean filter is applied for both SST and SLP fields. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

thumbnail image

Figure 15. Similar to Figure 14 but for the sub-period 1941–2004. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

Table VI. Correlation coefficients between the 65–100 years quasi-oscillatory mode (MD mode) of the LST-PCs, the AMO index and the common mode in the SST-PCs (sub-period 1872–1940)
 MD-LST1MD-LST2MD-LST3AMO_ index
  • *

    Statistically significant at the 95% confidence level.

AMO_index− 0.28*0.68*− 0.32*1.00*
MD-SST1− 0.010.46*0.48*0.47*
MD-SST2− 0.46*0.78*− 0.30*0.94*
MD-SST3− 0.080.63*0.060.84*
MD-SST4− 0.77*0.56*− 0.91*0.20*
MD-SST50.49*− 0.080.41*0.57*
Table VII. Correlation coefficients between the 65–100 years quasi-oscillatory mode (MD mode) of the LST-PCs, the AMO index and the common mode in the SST-PCs (sub-period 1941–2004)
 MD-LST1MD-LST2MD-LST3AMO_ index
  • *

    Statistically significant at the 95% confidence level.

AMO_index0.83*0.65*0.66*1.00*
MD-SST1− 0.37*0.10*0.11*− 0.59*
MD-SST20.79*0.33*0.36*0.88*
MD-SST30.63*0.090.11*0.76*
MD-SST4− 0.48*− 0.14*− 0.12*− 0.72*
MD-SST50.93*0.77*0.81*0.92*

For the period 1872–1940, correlation values between AMO index and MD mode of the LST-PCs (at lag 0) are − 0.28 and − 0.32 for LST1 and LST3, respectively, and 0.68 for LST2 (Table VI). Similar SST/SLP spatial patterns are found for MD-LST1 and MD-LST3 (Figure 14(a) and (c)). For these two MD-LST-PCs, maximum correlations are found with the MD-SST4, indicating that positive anomalies of temperature in north-western Europe (LST1) and in south-eastern Europe (LST3) could be forced primarily by SST anomalies sited to the south-western Atlantic.

Note in Figure 14 that the associated SLP patterns, for the period 1872–1940, for the LST1 and LST3 MD modes, show a dipole pattern, with a sign opposite that observed for the LST2 MD mode and AMO. This suggests that relatively warm air is transported by strong westerly winds to north-western Europe (LST1) and south-eastern Europe (LST3). However, for the MD-LST2, maximum correlation is found with the MD-SST2 (r = 0.78 for the period 1872–1940), and the North Atlantic SST spatial pattern associated is very similar to that related to AMO (Figure 14 (b) and (d)). Particularly, note that the tropical region and the area close to the north-western Africa reveal to be important. In addition, for the LST2 MD mode, the associated SLP shows a pattern similar to that of the AMO, suggesting that relatively warm air from the region close to the north-western African coast is transported around a low-pressure centre sited over the Azores to the Iberian Peninsula without affecting the rest of Europe.

According to Figure 13(i), for the second half of the period (1941–2004) more similar results are found for all MD-LST-PCs and AMO, being the correlation values between them 0.83, 0.65 and 0.66 for the MD-LST1, MD-LST2 and MD-LST3, respectively (Table VII). All the LST-PC MD modes show highest correlation coefficients with the MD mode of SST5 (r = 0.93, 0.77 and 0.81 for the MD-LST1, MD-LST2 and MD-LST3, respectively), associated to the region located in the central North Atlantic Ocean. Highest SST regression coefficients are found for MD-LST2 (Figure 15(b)), reaching 0.55 °C/SD in the region placed to the east of Newfoundland. For SLP, spatial pattern associated to the MD mode of LST2 is important to note the positive SLP centre sited over Europe, which could indicate some influence from the Mediterranean Sea. Similar results are found for the MD-LST3 (Figure 15(c)), but now both the SST as SLP anomaly centres are weaker. Also note that for LST3, the nonlinear trend explains a small part of the variance (only 7%, Table I), and the variance associated to the MD mode is very small.

In summary, the former results suggest that, at multidecadal time scales and during the first half of the period, the mechanism that regulates the north-western and south-eastern European temperatures (MD-LST1 and MD-LST3 modes) is different from that affecting the Iberian Peninsula temperatures (MD-LST2). For the second half of the period, similar SST patterns for all the MD-LST-PCs are found, but again the temperatures of the Iberian Peninsula seem to be affected by a different mechanism that could include some influence from the Mediterranean Sea.

7. Discussion and concluding remarks

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data
  5. 3. Methodology
  6. 4. Analysis
  7. 5. Common oscillatory modes
  8. 6. Nonlinear trend analysis
  9. 7. Discussion and concluding remarks
  10. Acknowledgements
  11. References

In this work, we have sought to quantify the association between the SSTs and the European LSTs variability. To this end, the common space-time variability of the LSTs and Atlantic SSTs has been analysed by using, first, PCA, and then SSA. We use monthly data of LST and SST, which are distributed over 5° latitude by 5° longitude grids (for LST) and 2° latitude by 2° longitude grids (for SST). The time period of study covers 1872–2004, and a low-pass filter that retains only the variability greater than one year was applied to the LST and SST data prior to the analysis.

With respect to the spatial variability of European LSTs, three significant spatial EOFs were found, associated with different European regions. The first mode (LST1) covers mainly the variability of western and central Europe, southern Scandinavia and the British Isles. The second one (LST2) is fundamentally representative of the variability of the south-west of Europe, especially the Iberian Peninsula; while the third one (LST3) represents the south-eastern part of Europe. Similar variability patterns were also obtained in other studies (Pozo-Vázquez et al., 2001a; Castro-Díez et al., 2002). From this result, we can note the well-differenced behaviour that presents the Iberian Peninsula compared to the rest of Europe, mainly justified by its localization with respect to the south centre of NAO dipole (Pozo-Vázquez et al., 2001a; Castro-Díez et al., 2002).

For the Atlantic SSTs, the PCA result shows several zones with a well-differentiated behaviour: the tropical South Atlantic (SST1), the quadripolar patterns in the North Atlantic Ocean (SST2 and SST3), the south-western Atlantic Ocean (SST4) and the central North Atlantic (SST5).

The application of SSA to the spatial PCs, corresponding to the European LST and Atlantic SST, shows the existence of significant common quasi-periodic oscillations to the LST1 and SSTs, whereas for the LST2 and LST3, there is only a weak association with the oscillatory modes of the SSTs. For the LST1, these common oscillations have periods of around 13.7 (ID), 7.5 (QD), 5.2 (IA1) and 3.6 (IA2) years. Additionally, we find that some of these oscillations are also present in the NAO index (QD and IA1) and in the ENSO index (IA2). However, these common quasi-oscillatory modes do not represent a high fraction of the explained variance for the LST1 (only 24.5%). These LST1 oscillatory modes were isolated and their spatial–temporal structures associated with the SST and SLP fields were characterized using lead–lag regression analysis. Using the former methodology, we were able not only to highlight the Atlantic Ocean regions that show a common temporal variability to the European LST, but also to quantify this association between the SSTs and the north-western European temperatures.

For the ID mode of LST1 (period ∼13.7 years), the principal contributions of the Atlantic Ocean come from the regions sited in the tropical South Atlantic (SST1) and from the centres of the quadripolar pattern of the SST (SST3), specially the SST anomalies of the south of Greenland, and opposite-signed anomalies in northern European waterways as well as the area emanating eastward from the east coast of the United States. This is in agreement with the results of Tourre et al. (2010), who find a double tripole configuration SST pattern over the entire Atlantic basin for a quasi-decadal oscillation (with period between 8 and 14 years), that leads the tropical inter-hemispheric out-of-phase relationship. Also, the North Hemispheric SST/SLP spatial pattern of this ID mode is in agreement with the results provided by Arguez et al. (2009), who found a North Atlantic SST mode that exhibits considerable spectral energy in the quasi-decadal (∼12 years) range, that influences the air temperature anomalies over Europe. Warmer land-surface temperatures on the northern of Europe are associated with resulting westerly wind anomalies across the sub-polar North Atlantic that increases maritime influence in northern of Europe (in agreement with the work of Arguez et al. (2009), (Figure 3(c)). The warm anomaly in the storm-formation region moves eastwards and then appears to split into two branches over the Mid-Atlantic Ridge, in agreement with Sutton and Allen (1997) and Dima and Lohmann (2004). As in this work, Arguez et al. (2009) also find that the thermal impacts of the low phase of ID mode are the opposite of those associated with the high phase (colder conditions over Europe due to the reverse pressure anomaly gradients and resulting changes in thermal advection that including continental and maritime effects). Based on the oscillatory character of this mode, we find that warm SST to the south of Greenland and in the tropical North Atlantic, along with cool SST in the Gulf Stream region and in the tropical South Atlantic, are associated with positive anomalies of LST in north-western Europe around 60 months later (or 12 months earlier with opposite sign considering these states recur along the time).

Concerning the spatial SLP pattern, the QD mode of LST1 (period ∼7.5 years) exhibits the familiar NAO dipole (Hurrell and van Loon, 1997). Along with this SLP pattern appears the well-known tripole of SST anomalies (Tourre et al., 1999). Rodwell et al. (1999) showed that such patterns do emerge from simple regression of the SST on the NAO index. Results from other studies have also suggested a strong relationship between the NAO index and temperatures over northern Europe at timescales between 6 and 10 years (Hurrell and van Loon, 1997). The evolution, during the 7.5-year cycle, is essentially towards a transition between the positive and negative phases of the NAO by simultaneous weakening and strengthening of each principal action centres. However, the transition phases show other patterns, different from the North–South dipole. This result is in agreement with the studies of da Costa and Colin de Verdiere (2002), who found a 7.7-year coupled oscillation for the SLP and SST of the North Atlantic region, and the more recent study carried out by Antunes et al. (2010), where the North Atlantic mean SLP field variability is analysed, although the periodicity found in this last case is ∼9 years. To this QD mode, the main contribution of the Atlantic Ocean comes from the regions sited to the south of Greenland, over the North Sea, and in the mid-latitudes around 40°N (three upper latitudinal lobes of the SST3). Similarly to the results of Moron et al. (1998) and da Costa and Colin de Verdiere (2002), this mode involves a phase opposition between the North Atlantic's subtropical and subpolar gyres, with a westerly intensification coinciding with a positive lobe off Cape Hatteras and negative lob off Newfoundland. In addition, we find that warm SST to the south of Greenland and cool SST to the south of Newfoundland and in the North Sea are related, with a delay of around 36 months, with warm LST in north-western Europe (or a few months earlier with opposite sign).

For the IA1 LST1 mode (period ∼5.2 years), the main contribution comes from the regions associated with the SST quadripole in the Atlantic Ocean. For this case, positive SST anomalies to the south of Greenland and in the tropical North Atlantic, together with negative SST anomalies in the mid-latitudes around 40°N and in the North Sea are related to warm north-western European LST around 31 months later. Because of most of the ENSO-related variability is contained in the interannual range between 3 and 6 years, further research is required to determine if this quasi-oscillatory mode could be associated with the ENSO global phenomenon. Note that the detected Atlantic regions in our study are in agreement with the results of Tourre et al. (1999), who found a quasi-oscillatory mode of period around 4.4 years for the Atlantic SST and SLP fields. They cautiously argued that this oscillatory mode depends more upon Atlantic local conditions, while a 3.5-year period is largely associated with the global ENSO mode.

The IA2 mode (period ∼3.6 years) explains only a small fraction of the variance both for the LST1 (∼4%) as well as for the SST (∼5%). We find a strong correlation between the IA2 mode of the ENSO and the IA2 mode of the quadripole pattern of the Atlantic SST; however, the linear influence of the Atlantic SST on this LST1 component is very low. At this point, it is important to note that the impact of ENSO on the European sector is not very clear. While most studies suggest a weak but significant ENSO response over the European sector, there remain considerable uncertainties regarding the regional details and quantitative aspects of the response, both in terms of the spatial structure of anomalies and the magnitude of the signal. Composite views of the observed impact of ENSO on the European region were obtained by van Loon and Madden (1981) and Fraedrich (1990). These authors identified a North–South dipole in the surface pressure field over Europe during El Niño winters. A positive pressure anomaly is located over Scandinavia, and a negative anomaly extends from France to the Black Sea. Fraedrich and Müller (1992) confirmed these results and argued further that, during the El Niño phase, the trajectory of Atlantic depressions tends to shift southward, leading to anomalous warmer conditions over south-central Europe and colder conditions over northeastern Europe. Conversely, during the La Niña phase, the storm track tends to move northward and transport more heat and moisture towards northern Europe. There is evidence of an asymmetry between the phases of ENSO, with a more robust remote response during the cold La Niña phase than the warm El Niño phase. Fraedrich (1990) and Wilby (1993) studied the observed synoptic variability of climate in Europe and noted that the cyclonic-type circulation associated with El Niño is highly variable, while the anticyclonic-type circulation associated with La Niña is more consistent between events. A similar result was obtained by Pozo-Vázquez et al. (2001b), who identified a response resembling to the positive phase of the NAO during La Niña but no significant signal during El Niño. In this last work, a temperature analysis showed statistically significant negative anomalies during La Niña events over the Iberian Peninsula and positive anomalies over the British Isles and southern Scandinavia. All these studies have established the need to separate the El Niño and La Niña events in the analysis; however, there are some additional factors that must be taken into account. First, there are different types of El Niño and La Niña events, with different characteristics that can lead to different responses of the extratropical atmospheric circulation. Second, climatological planetary atmospheric waves, natural noise and the complexity of the numerous feedbacks (and maybe nonlinear relationships) can hide the signal of ENSO in the extratropics (Trenberth, 1997). It appears that the possible influence of ENSO in the European regions is more likely to be found during extreme events of ENSO and during the winter.

Estimations of the ID, QD and IA1 modes of LST1, using multiple-linear-regression models that take into account the previous time lags found for the SST regions acting as potential predictors for these modes, show that although the SST explains a high percentage of variance of the individual oscillations (75, 67, and 50% for the QD, ID and IA1 mode, respectively), the total contribution of these three modes explains only a small fraction of the variance (12%) of the monthly north-western European temperatures.

Additionally, although SSA also detects, for all the PCs, nonlinear trends which explain high variance percentages, these consistently appear as nonsignificant, probably due to the short length of the record. An exhaustive analysis of these terms reveals that the nonlinear trends can be seen as the sum of two contributions: a linear trend, presumably due to the anthropogenic effect, and a multidecadal oscillation with period around 64–100 years, depending on the region represented by the PC, which appears to be associated with the AMO. Other authors also have found a multi-decadal signal of uniform polarity over the North Atlantic SST (Delworth et al., 1993; Deser and Blackmon, 1993; Kushnir, 1994; Kerr, 2000; Enfield et al., 2001; Sutton and Hodson, 2005; Dima and Lohmann, 2007; Arguez et al., 2009; Ting et al., 2009; Guan and Nigam, 2009). Particularly, in the study of Arguez et al. (2009), the AMO signal in the North Atlantic SST is associated with a timescale of about 50 years. In that study, an analysis of the influence of the AMO in the European air temperatures, for the period 1906–2005, using composites analysis of annual mean air temperatures anomalies during the high and low phase of the AMO is carried out, finding, in agreement with other authors (Sutton and Hodson, 2005), that the AMO impact on the air temperatures over Europe is most intense in summer.

Additionally, we find that the multidecadal oscillations associated to the European temperatures show a different behaviour for the sub-periods 1872–1940 and 1941–2004. For the first sub-period, the associated pressure anomalies over the North Atlantic region are relatively strong and oriented in such a way as to promote the thermal advection of SST characteristics. In this case, the mechanism that regulates the north-western and south-eastern European temperatures is different from that affecting the Iberian Peninsula temperatures. Particularly, influence from the SST close to the north-western African coast appears to be relevant to the Peninsula. For the second half of the period, similar SST patterns for all the LST-PCs are found. For this second sub-period, the pressure anomalies are weaker and the resulting impact on the European temperatures does not appear to be caused by geostrophic advection from the Atlantic SST. This result is partially in agreement with the study of Arguez et al. (2009), although is important to note that they use the period 1906–2005. In addition, the temperatures of the Iberian Peninsula seem to be affected by a different mechanism that could include some influence from the Mediterranean Sea. Once again, the differenced behaviour that presents the Iberian Peninsula regarding the rest of Europe can be noticed. Further research needs to be carried out to clarify the nature of the AMO and its relationship with the different behaviour found for the MD oscillatory modes of European temperatures before and after 1940.

As already commented, many other studies have also revealed, in the Atlantic SST and SLP fields, quasi-oscillatory modes with periods of around 12 years (Sutton and Allen, 1997; Moron et al., 1998; Tourre et al., 1999; Dima and Lohmann, 2004; Arguez et al., 2009), 7.5 years (Moron et al., 1998; Dima and Lohmann, 2004), 3–5 years (Moron et al., 1998; Tourre et al., 1999; Dima and Lohmann, 2004) and, more recently, multidecadal oscillations with periods in the range 60–100 years (Lohmann et al., 2004; Sutton and Hodson, 2005; Grosfeld et al., 2008). However, the mechanisms behind these oscillatory modes are not fully understood. Especially, the contribution of different latitudes and ocean basins is still highly controversial, mixing processes such as advection, convection and teleconnections (Deser et al., 2010). While we do not provide information concerning the underlying physics of the common oscillatory modes found for the European LST and Atlantic SST, our results suggest an important influence of the SLP as a main component in the SST/LST relationship; however, physical explanations for the connection between the SST and LST are not yet clear. Cassou and Terray (2001); Rodwell et al. (1999) and Czaja and Frankignoul (2002) suggest a mechanism of the NAO behaviour connected to the North Atlantic SST. During times with negative anomalies in the low-latitude North Atlantic SST, the NAO index is positive. By contrast, positive low-latitude North Atlantic SST winter anomalies go with negative NAO index values. These results support the findings of our analysis about the spatial behaviour of SST and SLP fields associated to the QD and ID oscillatory modes of the LST1 (Figures 8 and 10 at lag 0), where the NAO dipolar pattern appears very clearly, although in our case, the SST of the upper latitudes also play an important role. Additionally, it is important to note that SLP patterns other than the NAO could also have a dominant role for determining European temperatures (Junge and Stephenson, 2003).

To sum up, although the results of this paper suggest that the Atlantic SST alone does not provide a valuable simple forecast model for monthly European temperatures at interannual to interdecadal time scales, we have identified the regions in the Atlantic Ocean that present a linear relationship with the European temperatures. This is a key result for some reasons. Firstly, our study contributes to gain insight into the part of the European temperature anomalies that may be related to ocean processes, revealing the scale-dependent LST–SST lead–lag relationships and quantifying the importance of these relationships between the SST and the LST. And secondly, although the development of specific prediction schemes is beyond the scope of this study, the relevant SST regions in terms of LST variability found in this paper have permitted to explain about 12% of the north-western European LST, reaching the percentage of phase accordance 64%. Although other modelling studies have also concluded that the signal of sea surface temperature forcing on the atmospheric circulation is small (also at decadal scales), and that, therefore, the prospects for long-term prediction of air temperatures based on the state of SSTs will not necessary be bright (Trigo et al., 2002; Paeth et al., 2003), the modelling scheme developed in this paper allow us to test quantitatively the importance of different factors by including/removing them from a linear model, providing information about the phase or persistence of SST. Also results are encouraging as only information contained in the Atlantic SST has been used to explain the European temperature variability. The performance of statistical forecasting models that use other variables such as the SLP as potential predictor variables could be improved by the addition of variables that represents these SST regions. In addition, for longer time scales (multidecadal), the role played by the SST becomes more important.

Summarizing, this work has investigated links between SST variations in the Atlantic Ocean and LSTs in Europe. The usefulness of this analysis rests in the finding of quasi-oscillatory modes that contain a substantial part of the variability of the European LST and Atlantic SST, providing connections that suggest coupled mechanisms of variability. At European latitudes, where there is absence of a strong coupling atmosphere-ocean such as El Niño Southern Oscillation, the identification of a clear connection between atmosphere and ocean is a complex task. Further research should be conducted in some aspect such as the clarification of the mechanisms that produce these connections or the advantages of using the information contained in the SST to produce forecasts of European temperatures.

Acknowledgements

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data
  5. 3. Methodology
  6. 4. Analysis
  7. 5. Common oscillatory modes
  8. 6. Nonlinear trend analysis
  9. 7. Discussion and concluding remarks
  10. Acknowledgements
  11. References

The Spanish Ministry of Science and Innovation, with additional support from the European Community Funds (FEDER), projects CGL2007-61151/CLI, CGL2010-21188/CLI and the Regional Government of Andalusia project P06-RNM-01622, financed this study. S. R. Gámiz-Fortis is supported by the University of Granada under a postdoctoral contract.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data
  5. 3. Methodology
  6. 4. Analysis
  7. 5. Common oscillatory modes
  8. 6. Nonlinear trend analysis
  9. 7. Discussion and concluding remarks
  10. Acknowledgements
  11. References
  • Allan RJ, Ansell TJ. 2006. A new globally complete monthly historical mean sea level pressure data set (HadSLP2): 1850–2004. Journal of Climate 19: 58165842.
  • Antunes S, Pires O, Rocha A. 2010. Spatio-temporal patterns of pressure over the North Atlantic. International Journal of Climatology DOI: 10.1002/JOC.2052. (in press).
  • Arguez A, O'Brien JJ, Simith SR. 2009. Air temperature impacts over Eastern North America and Europe associated with low-frequency North Atlantic SST variability. International Journal of Climatology 29: 110.
  • Barnston AG, Livezey RE. 1987. Classification, seasonality and persistence of low-frequency atmospheric circulation patterns. Monthly Weather Review 115: 10831126.
  • Bjerknes J. 1969. Atmospheric teleconnections from the equatorial Pacific. Monthly Weather Review 97: 163172.
  • Blender R, Luksch U, Fraedrich K, Raible CC. 2003. Predictability study of the observed and simulated European climate using linear regression. Quarterly Journal of the Royal Meteorological Society 129: 22992313.
  • Brohan P, Kennedy JJ, Harris I, Tett SFB, Jones PD. 2006. Uncertainty estimates in regional and global observed temperature changes: a new dataset from 1850. Journal of Geophysical Research 111: D12106, DOI:10.1029/2005JD006548.
  • Carson DJ. 1998. Seasonal forecasting. Quarterly Journal of the Royal Meteorological Society 124: 126.
  • Cassou C, Terray L. 2001. Oceanic forcing of the wintertime low frequency atmospheric variability in the North Atlantic European sector: a study with the ARPEGE model. Journal of Climate 14: 42664291.
  • Castro-D´ıez Y, Pozo-Vázquez D, Rodrigo FS, Esteban-Parra MJ. 2002. NAO and winter temperature variability in southern Europe. Geophysical Research Letter 29(8): 14 DOI: 10.1029/2001GL014042.
  • Cayan RD. 1992. Latent and sensible heat flux anomalies over the northern oceans: driving the sea surface temperature. Journal of Physical Oceanography 22: 859881.
  • Czaja A, Frankignoul C. 2002. Observed impact of Atlantic SST anomalies on the North Atlantic Oscillation. Journal of Climate 15: 606623.
  • da Costa DE, Colin de Verdiere A. 2002. The 7.7-year North Atlantic Oscillation. Quarterly Journal of the Royal Meteorological Society 128: 797817.
  • Delworth TL, Manabe S, Stouffer RJ. 1993. Interdecadal variations of the thermohaline circulation in a coupled ocean-atmosphere model. Journal of Climate 6: 19932011.
  • Delworth TL, Mann ME. 2000. Observed and simulated multidecadal variability in the Northern Hemisphere. Climate Dynamics 16: 661676.
  • Deser C, Alexander MA, Xie SP, Phllips AS. 2010. Sea surface temperature variability: patterns and mechanism. Annual Review of Marine Science 2: 115143.
  • Deser C, Blackmon ML. 1993. Surface climate variations over the North Atlantic Ocean during winter: 1900–1989. Journal of Climate 6: 17431753.
  • Deser C, Timlin MS. 1997. Atmosphere-ocean interaction on weekly timescales in the North Atlantic and Pacific. Journal of Climate 10: 393408.
  • Deser C, Wallace JM. 1987. El Nino events and their relationship to the Southern Oscillation: 1925–1986. Journal of Geophysical Research 92: 1418914196.
  • Dima M, Lohmann G. 2004. Fundamental and derived modes of climate variability: concept and application to interannual time-scales. Tellus 56A: 229249.
  • Dima M, Lohmann G. 2007. A hemispheric mechanism for the Atlantic multidecadal oscillation. Journal of Climate 20: 27062719.
  • Draper NR, Smith H. 1998. Applied Regression Analysis. John Wiley & Sons, Inc.: New York.
  • Elsner JB, Tsonis AA. 1996. Singular Spectrum Analysis. A New Tool in Time Series Analysis. Plenum Press: New York.
  • Enfield D, Mestas-Nunez A, Trimble P. 2001. The Atlantic multidecadal oscillation and its relation to rainfall and river flows in the continental U.S. Geophysical Research Letter 28: 20772080.
  • Fraedrich K. 1990. European Grosswetter during the warm and cold extremes of the El Niño/Southern Oscillation. International Journal of Climatology 10: 2131.
  • Fraedrich K, Müller K. 1992. Climate anomalies in Europe associated with ENSO extremes. International Journal of Climatology 12: 2531.
  • Gámiz-Fortis SR, Pozo-Vázquez D, Esteban-Parra MJ, Castro-D´ıez Y. 2002. Spectral characteristics and predictability of the NAO assessed through Singular Spectral Analysis. Journal of Geophysical Research 107(D23): 4685, DOI: 10.1029/2001JD001436.
  • Gámiz-Fortis SR, Pozo-Vázquez D, Trigo RM, Castro-D´ıez Y. 2008a. Quantifying the predictability of winter river flow in Iberia. Part I: interannual predictability. Journal of Climate 21(11): 24842502.
  • Gámiz-Fortis SR, Pozo-Vázquez D, Trigo RM, Castro-D´ıez Y. 2008b. Quantifying the predictability of winter river flow in Iberia. Part II: seasonal predictability. Journal of Climate 21(11): 25032518.
  • Ghil M, Mo KC. 1991. Intraseasonal oscillation in the global atmosphere. Part I: Northern hemisphere and tropics. Journal of the Atmospheric Sciences 48: 752790.
  • Ghil M, Vautard R. 1991. Interdecadal oscillations and the warming trend in global temperature time series. Nature 350: 324327.
  • Goddard L, Mason SJ, Zebiak SE, Ropelewski CF, Basher R, Cane MA. 2001. Current approaches to seasonal-to-interannual climate predictions. International Journal of Climatology 21: 11111152.
  • Grosfeld K, Lohmann G, Rimbu N. 2008. The impact of Atlantic and Pacific Ocean sea surface temperature anomalies on the North Atlantic multidecadal variability. Tellus A 60(4): 728741.
  • Guan B, Nigam S. 2009. Analysis of Atlantic SST variability factoring inter-basin links and secular trend: clarified structure of the Atlantic Multidecadal Oscillation. Journal of Climate 22: 422840.
  • Hoerling MP, Hurrell JW, Xu T. 2001. Tropical origins for recent North Atlantic climate change. Science 292: 9092.
  • Horel JD, Wallace JM. 1981. Planetary scale atmospheric phenomena associated with the Southern Oscillation. Monthly Weather Review 109: 813829.
  • Hurrell JW. 1995. Decadal trends in the North Atlantic Oscillation: regional temperatures and precipitation. Science 269: 676679.
  • Hurrell JW, van Loon H. 1997. Decadal Variations associated with the North Atlantic Oscillation. Climatic Change 36: 301326.
  • Hurrell JW, Folland CK. 2002. A change in the summer atmospheric circulation over the North Atlantic. CLIVAR Exchanges 7: 5254.
  • Jones P, Jonsson T, Wheeler D. 1997. Extension to the North Atlantic Oscillation using early instrumental pressure observations from Gibraltar and south-west Iceland. International Journal of Climatology 17: 143350.
  • Junge MM, Stephenson DB. 2003. Mediated and direct effects of the North Atlantic Ocean on winter temperatures in northwest Europe. International Journal of Climatology 23: 245261.
  • Kerr RA. 2000. A north Atlantic climate pacemaker for the centuries. Science 288: 19841985.
  • Kushnir Y. 1994. Interdecadal variations in North Atlantic sea surface temperature and associated atmospheric conditions. Journal of Climate 7: 141157.
  • Lamb PJ, Peppler RA. 1987. North Atlantic oscillation: concept and and application. Bulletin of the American Meteorological Society 68: 12181225.
  • Latif M, Arpe K, Roeckner E. 2000. Oceanic control of decadal North Atlantic sea level variability in winter. Geophysical Research Letters 27: 727730.
  • Lohmann G, Rimbu N, Dima M. 2004. Climate signature of solar irradiance variations: analysis of long-term instrumental and historical data. International Journal of Climatology 24: 10451056, DOI: 10.1002/joc.1054.
  • Mathieu PP, Sutton RT, Dong B, Collins M. 2004. Predictability of winter climate over the North Atlantic European region during ENSO events. Journal of Climate 17: 19531974.
  • McCabe GJ, Palecki MA, Betancourt JL. 2004. Pacific and Atlantic Ocean influences on multidecadal drought frequency in the United States. Proceedings of the National Academy of Sciences of the United States of America 101: 41364141.
  • Mehta VM, Suarez MJ, Manganello JV, Delworth TL. 2000. Oceanic influence on the North Atlantic Oscillation and associated Northern Hemisphere climate variations: 1959–1993. Geophysical Research Letters 27(1): 121124.
  • Moron V, Vautard R, Ghil M. 1998. Trends, interdecadal and interannual oscillations in global sea-surface temperatures. Climate Dynamics 14: 545569.
  • Osborn TJ, Briffa KR, Tett SFB, Jones PD, Trigo RM. 1999. Evaluation of the North Atlantic oscillation as simulated by a coupled climate model. Climate Dynamics 15: 685702.
  • Paeth H, Latif M, Hense A. 2003. Global SST influence on tuwentieth century NAO variability. Climate Dynamics 21: 6375.
  • Paluš M, Novotná D. 2004. Enhanced Monte Carlo singular system analysis and detection of period 7.8 years oscillatory modes in the monthly NAO index and temperature records. Nonlinear Processes of Geophysics 11: 721729.
  • Paluš M, Novotná D. 2006. Quasi-Biennial oscillations extracted from the monthly NAO index and temperature records are phase-synchronized. Nonlinear Processes in Geophysics 13: 287296.
  • Paluš M, Novotná D. 2008. Detecting oscillations hidden in noise: common cycles in atmospheric, geomagnetic and solar data. Nonlinear Time Series Analysis in the Geosciences 112: 327353, DOI: 10.1007/978-3-540-78938-3.
  • Philander G. 1990. El Niño, la Niña, and the Southern Oscillation. Academic Press Inc.: USA; 289.
  • Plaut G, Vautard R. 1994. Spells of low-frequency oscillations and weather regimes in the Northern Hemisphere. Journal of the Atmospheric Sciences 51(2): 210236.
  • Pozo-Vázquez D, Esteban-Parra MJ, Rodrigo FS, Castro-D´ıez Y. 2001a. A study of NAO variability and its possible non-linear influences on European surface temperature. Climate Dynamics 17: 701715.
  • Pozo-Vázquez D, Esteban-Parra MJ, Rodrigo FS, Castro-Diez Y. 2001b. The association between ENSO and winter atmospheric circulation and temperature in the North Atlantic region. Journal of Climate 14: 34083420.
  • Rasmusson EM, Carpenter TH. 1982. Variations in tropical sea-surface temperature and wind fields associated with the southern oscillation/El Niño. Monthly Weather Review 110: 354384.
  • Rayner NA, Parker DE, Horton EB, Folland CK, Alexander LV, Rowell DP, Kent EC, Kaplan A. 2003. Globally complete analyses of sea surface temperature, sea ice and night marine air temperature, 1871–2000. Journal of Geophysical Research 108: 4407. DIO: 10.1029/2002JD002670.
  • Robertson AW, Mechoso CR, Kim YJ. 2000. The influence of Atlantic sea surface temperature anomalies on the North Atlantic oscillation. Journal of Climate 13: 122138.
  • Rodriguez-Fonseca B, Polo I, Serrano E, Castro M. 2006. Evaluation of the North Atlantic SST forcing on the European and Northern African winter climate. International Journal of Climatology 26: 179191.
  • Rodwell MJ, Rowell DP, Folland CK. 1999. Oceanic forcing of the wintertime North Atlantic Oscillation and European climate. Nature 398: 320323.
  • Rogers JC. 1990. Patterns of low-frequency monthly sea-level pressure variability (1899–1986) and associated wave cyclone frequencies. Journal of Climate 3: 13641379.
  • Rogers JC, van Loon H. 1979. The seesaw in winter temperatures between Greenland and Northern Europe. Part II: some oceanic and atmospheric effects in middle and high altitudes. Monthly Weather Review 107: 509519.
  • Smith CA, Sardeshmukh P. 2000. The Effect of ENSO on the intraseasonal variance of surface temperature in winter. International Journal of Climatology 20: 15431557.
  • Sutton RT, Allen MR. 1997. Decadal predictability of the Gulf Stream SST. Nature 388: 563567.
  • Sutton RT, Hodson DLR. 2003. Influence of the ocean on North Atlantic climate variability 1871–1999. Journal of Climate 16: 32963313.
  • Sutton RT, Hodson DLR. 2005. Atlantic Ocean forcing of North American and European summer climate. Science 309: 115118.
  • Terray L, Cassou C. 2002. Tropical atlantic sea surface temperature forcing of Quasi-Decadal climate variability over the North Atlantic-European region. Journal of Climate 15: 31703187.
  • Ting M, Kushnir Y, Seager R, Li C. 2009. Forced and internal twentieth-century SST trends in the North Atlantic. Journal of Climate 22: 146981.
  • Tourre YM, Paz S, White WB. 2010. Low-frequency climate variability in the Atlantic basin during the 20th century. Atmospheric Science Letters DOI: 10.1002/asl.265. (in press).
  • Tourre YM, Rajagopalan B, Kushnir Y. 1999. Dominant patterns of climate variability in the Atlantic Ocean during the last 136 years. Journal of Climate 12: 22852299.
  • Trenberth KE. 1997. Short-term climate variations: recent accomplishments and issues for future progress. Bulletin of the American Meteorological Society 78: 1081109.
  • Trigo R, Osborn TJ, Corte-Real JM. 2002. The North Atlantic Oscillation influence on Europe: climate impacts and associated physical mechanism. Climate Research 20: 917.
  • van Loon H, Madden RA. 1981. The Southern Oscillation. Part I: global associations with pressure and temperature in northern winter. Monthly Weather Review 109: 11501162.
  • van Loon H, Rogers JC. 1978. The seesaw in winter temperatures between Greenland and Northern Europe. Part I: general description. Monthly Weather Review 106: 296310.
  • Vautard R, Ghil M. 1989. Singular spectrum analysis in nonlinear dynamics, with applications to paleoclimatic time series. Physica D 35: 395424.
  • Vautard R, Yiou P, Ghil M. 1992. Singular spectrum analysis: a toolkit for short, noisy chaotic signal. Physica D 58: 95126.
  • Walker GT. 1924. Correlations in seasonal variations of weather. I. A further study of world weather. Memoirs of the Indian Meteorological Department 24: 275332.
  • Walker GT, Bliss EW. 1932. World weather V. Memoirs of the Royal Meteorological Society 4(36): 5384.
  • Wallace JM, Gutzler DS. 1981. Teleconnections in the geopotential height field during the Northern Hemisphere winter. Monthly Weather Review 109: 784812.
  • Wallace JM, Jiang QR. 1987. On the observed structure of the interannual variability of the atmosphere-ocean climate system. Atmospheric and Oceanic Variability, Cattle H (ed). Royal Meteorological Society, 1743.
  • Wilby R. 1993. Evidence of ENSO in the synoptic climate of the British Isles since 1880. Weather 48: 234239.
  • Wilks DS. 1995. Statistical Methods in the Atmospheric Sciences, Academic Press: San Diego, CA; 467.
  • Zhang X, Sheng J, Shabbar A. 1998. Modes of interannual and interdecadal variability of Pacific SST. Journal of Climate 11: 25562569.
  • Zorita E, Harin VK, Von Storch H. 1992. The atmospheric circulation and sea surface temperature in the North Atlantic area in winter: their interaction and relevance for Iberian precipitation. Journal of Climate 5: 10971108.