The predictability of winter climate over the North Atlantic–European (NAE) region during ENSO events is investigated. Rather than employing traditional composite analyses, the authors focus on the impacts of six individual events: three El Niño events and three La Niña events. The investigation is based on the analysis of ensemble simulations with an atmospheric GCM forced with prescribed sea surface temperatures (SST) for the period December 1985–May 2001, and on observations. Model experiments are used to separate the respective roles of SST anomalies in the Indo-Pacific basin and in the Atlantic basin.
A significant (potentially predictable) climate signal is found in the NAE region for all six ENSO events. However, there are notable differences in the impacts of individual El Niño and La Niña events. These differences arise not simply from atmospheric internal variability but also because the atmosphere is sensitive to specific features of the SST anomaly fields that characterize the individual events. The different impacts arise partly from differences in Indo-Pacific SST and partly from differences in Atlantic SST. SST anomalies in both ocean basins can influence tropical convection and excite a Rossby wave response over the North Atlantic. The evidence presented here for the importance of Atlantic Ocean conditions argues that, in the development of systems for seasonal forecasting, attention should not be focused too narrowly on the tropical Pacific Ocean.
Seasonal climate forecasting (SCF) is a revolutionary technology aiming to predict the likely state of the climate several months ahead. The scientific basis for SCF lies in the lower boundary of the atmosphere (i.e., ocean and land), which acts as a climate pacemaker by modulating the variability of the atmosphere on seasonal and longer time scales. SCF relies on the extraction of these potentially predictable low-frequency signals, either by use of empirical statistical methods (e.g., Barnston 1994) or by employing sophisticated dynamical models (e.g., Palmer and Anderson 1994). Although still in its infancy, SCF has already demonstrated its potential to bring enormous benefits to society by helping to better manage climate-related risks (Goddard et al. 2001).
Globally speaking, most of the SCF skill derives from the El Niño–Southern Oscillation (ENSO) phenomenon. ENSO is the dominant mode of climate variability on seasonal-to-interannual time scales, and arises from strong ocean–atmosphere interactions in the tropical Pacific Ocean. These interactions lead to a large-scale shift of atmospheric convective activity in association with anomalous warming [the El Niño (EN) phase] or cooling [the La Niña (LN) phase] of the central and eastern equatorial Pacific. Although centered on the Pacific, ENSO has major impacts around the globe. Remote climate impacts, or “teleconnections,” arise through the propagation of wavelike disturbances that are excited in the tropical Pacific (e.g., Trenberth et al. 1998). A better understanding of ENSO teleconnections appears of critical importance to advance our capability to make reliable SCF, especially for extratropical regions where the direct influence of the ocean on the atmosphere is comparatively weak (e.g., Kushnir 1994).
Our current canonical view of ENSO teleconnections was first documented using composite and correlation analyses of observational data (Ropelewski and Halpert 1987; Kiladis and Diaz 1989). For some regions, such as the tropical Atlantic and the Pacific–North American (PNA) sector, the remote impact of ENSO and the associated seasonal predictability are now well established (Livezey and Mo 1987; Kiladis and Diaz 1989). Physical mechanisms explaining the remote effect of tropical Pacific sea surface temperature (SST) have also been proposed. Lau and Nath (1996) argued that the tropical Atlantic is influenced by the Pacific through a shift of the Walker circulation referred as the “atmospheric bridge.” Horel and Wallace (1981) suggested that the extratropical regions can be influenced by SST from the tropical Pacific via a wave train emanating from the Tropics and propagating northward and eastward into higher latitudes. In contrast, the impact of ENSO on the North Atlantic–European (NAE) sector is less clear and, in some respects, controversial. Most studies suggest a weak but significant ENSO response over the NAE sector, but there remain considerable uncertainties regarding the regional details and quantitative aspects of the response.
Composite views of the observed impact of ENSO on the NAE 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 EN phase, the trajectory of Atlantic depressions tends to shift southward, leading to anomalous warmer/wetter conditions over south-central Europe and colder/drier conditions over northeastern Europe. Conversely, during the LN phase, the storm track tends to move northward and transport more heat and moisture toward northern Europe. These results are consistent with the correlation analyses of Kiladis and Diaz (1989) and van Oldenborgh et al. (2000), which showed a statistically significant link between EN (LN) events and enhanced (decreased) rainfall in the southern Europe, especially in the spring.
There is evidence of an asymmetry between the phases of ENSO, with a more robust remote response during the cold LN phase than the warm EN phase. Fraedrich (1990) and Wilby (1993) studied the observed synoptic variability of climate in Europe and noted that the cyclonic-type circulation associated with EN is highly variable while the anticyclonic-type circulation associated with LN is more consistent between events. A similar result was obtained by Pozo-Vasquez et al. (2001) who identified a response ressembling to the positive phase of the North Atlantic Oscillation (NAO+) during LN but no significant signal during EN. Cassou and Terray (2001b) provide a different perspective. They employed a cluster analysis technique to identify the preferred regimes of atmospheric flow. One of the regimes they identified had a clear association with LN conditions, but there was no corresponding EN regime. The spatial pattern of the LN regime features a center of enhanced anticyclonic flow west of Ireland, and differs from the NAO+ pattern found by Pozo-Vasquez et al. (2001).
Like the observational studies, results from model experiments have generally suggested the existence of a weak, but significant, response to ENSO over the NAE region. In the Prediction of Climate Variations on Seasonal to Interannual Time-scales project (PROVOST), several different atmosphere models were forced with SST data for the period 1973–86 (Graham et al. 2000). Seasonal predictability, assessed from the multimodel ensemble, was found to be weak over Europe but increased during extreme ENSO events, thereby suggesting the existence of an ENSO–NAE link (Graham et al. 2000). With regard to the spatial structure of the ENSO response, no robust picture emerges from past model studies. For example, Palmer and Anderson (1994) find an NAO-like response, whereas May and Bengtsson (1998) find a EN response that has no clear projection on the NAO. Cassou and Terray (2001a) find an NAO-like response but argue that this is an unrealistic feature of their model. Rigorous comparison between these model studies is made difficult by the fact that different authors employ different models and analyze different sets of ENSO events.
In summary, while there is general agreement that ENSO does influence the NAE region, there is considerable uncertainty regarding the details of this influence, both in terms of the spatial structure of anomalies and the magnitude of the signal. One of the reasons for this uncertainty is that most of the research, whether based on observations or model studies, has focused on composites of several ENSO events. The need to separate the EN and LN events in the composite analysis has been established, but the need to consider the impact of individual EN or LN events has not been seriously investigated. In reality, however, every such event is unique and different events will have different impacts. This is not simply because of internal atmospheric variability but rather because the precise ocean conditions differ between events. The importance of these differences has already been demonstrated for the PNA sector (Hoerling and Kumar 1997; Hoerling et al. 1997, 2001; Kumar and Hoerling 1997). In this study we aim to investigate their importance for understanding the impacts of ENSO on the NAE region. We focus on the following specific questions:
What is the pattern and predictability of the remote response in the NAE region to individual El Niño and La Niña events?
What is the specific role of the Atlantic Ocean in the ENSO–NAE teleconnections? How does forcing from the Atlantic and Pacific interact to Shape European climate variability?
We concentrate on boreal winter when the ENSO impact is likely to be strongest. We give specific consideration to the role of the Atlantic Ocean because there is evidence from a range of studies (Ratcliffe and Murray 1970; Deser and Blackmon 1993; Palmer and Sun 1985; Cassou and Terray 2001b; Czaja and Frankignoul 1999; Dong et al. 2000) that Atlantic SST is, in addition to ENSO, a significant influence on seasonal climate in the NAE region. We follow the methodology used by Dong et al. (2000) but extend their study by considering a much larger set of winters. The questions we have posed cannot be addressed solely through analysis of observations because, for individual ENSO events, there is no way to separate the influence of the ocean from internal atmospheric variability. Consequently, our approach is based on model experiments and we use ensemble simulations to achieve the necessary separation.
The structure of the paper is as follows. In section 2, we describe the experimental design and SST forcing. In section 3, we analyze the simulated and observed atmospheric response during EN and LN episodes. In section 4, we investigate the individual and combined role of the Pacific and Atlantic in shaping NAE climate. Some conclusions are drawn in section 5 and the implications of our results for SCF are investigated.
2. Experimental design
In this section, we describe the methodology and experimental design used to investigate the ENSO impact over Europe.
Two experiments are conducted by integrating the Third Hadley Centre Atmospheric General Circulation Model (HadAM3) over the period 1 December 1985– 30 April 2001 with different SST boundary conditions. The model has a resolution of 2.5 × 3.75 in the horizontal and 19 levels in the vertical. The reader is referred to Pope et al. (2000) for a detailed description of the formulation and behavior of HadAM3. Our first experiment is forced by global observed (GLOB) SST and sea ice extents (SIE) from Reynolds et al. (2002). For our second experiment, IPAC the SST and SIE are identical to GLOB in the Indian and Pacific Ocean basins (IPAC) but differ in the Atlantic basin (between 30°S and 75°N) where climatological SST and SIE values, averaged over the period 1986–2000, are imposed. We refer to the difference between the GLOB and IPAC experiments as “ATL” as it gives a measure of the influence of Atlantic SST on the atmosphere (including its interactions with the Pacific).
Each experiment consists of an ensemble of 10 atmospheric simulations integrated from slightly different initial states. The initial conditions were taken from 10 consecutive time slices of a spinup integration. The earliest data analyzed in this study is taken from one year after the integrations were started there by leaving ample time for the ensemble members to separate.
b. Data analysis
We use the model and observational data to investigate the properties of the atmospheric response during ENSO episodes. For comparison with past studies we first examine the composite responses for EN, LN, and ENSO (EN − LN) in both the observations and the model data. We then proceed to examine the atmospheric anomalies associated with individual EN and LN events. Our analyses are based on winter seasonal anomalies [average over December–February (DJF)] of two variables. The first variable is the geopotential height (GPH) at 500 mb, which provides a picture of the large-scale circulation. Observations of GPH are derived from the reanalysis of the National Centers for Environmental Prediction (NCEP). The second variable is the precipitation, which provides information about the tropical diabatic heating (related, in turn, to the forcing of Rossby waves that can propagate into the extratropics) and displacements of the storm track. Observations of precipitation are derived from the Global Precipitation Climatology Project (GPCP) (Huffman et al. 1997).
In the case of the model results, two questions are of particular interest. First, is the ocean infuence during a particular ENSO event strong by comparison with atmospheric internal variability? A positive answer implies potential predictability in the sense that knowledge of the SST field provides a useful constraint on the possible range of atmospheric states. We address this question by using information from the ensemble simulations to assess the statistical significance of the SST-forced signal (details are given below). Second, is there agreement between the model simulations and the observations? The approach we take here is pattern based, and focuses primarily on comparison of the observed and simulated patterns of GPH anomalies. An alternative approach (Brankovic et al. 1994; Graham et al. 2000) is to compare meteorological variables at specific locations, but a danger with this second approach is that it can give an unduly pessimistic impression if, as is common, model errors give rise to small displacements or distortions of otherwise well predicted patterns. This danger is the reason for our choice of a pattern-based approach. To compare simulated and observed anomaly patterns we use both qualitative inspection and quantitative calculations that are described below.
The procedure used to calculate seasonal anomalies and to assess their significance is as follows. In the case of observational data (subscript OBS), the winter seasonal anomalies (denoted by prime) of a generic variable (denoted by Y) for a specific year (superscript t) is defined by
where the overbar denotes the time-mean operator, averaging over the 1986–2000 climatological period containing T = 15 winters:
In case of model data, defining anomalies first requires us to determine what is the relevant signal for an ensemble of simulations. With multiple realizations of climate, the temporal variability can be separated into an externally induced response, associated with the SST forcing, and an internally induced component, associated with chaotic atmospheric fluctuations. Consequently, model outputs can be decomposed as follows:
where the subscript RUN corresponds to the experiment (GLOB or IPAC), k is the ensemble member index corresponding to a specific initial condition, Y tRUN,EXT is the SST-forced signal, and Y t, kRUN,INT is the internally induced noise. Understanding the relative contribution of the signal and noise gives a measure of the potential predictability of the atmosphere. An unbiased estimate (denoted by hat) of the time-varying signal is provided by the ensemble mean (denoted by EM) over K = 10 simulations:
It follows that the ensemble mean anomaly for a particular year provides an unbiased estimate of the SST-forced signal for that year. Ensemble mean anomalies are calculated as follows:
Note that, for the sake of comparison, model anomalies for GLOB and IPAC are defined relative to the same model climatology, that is, computed from the GLOB experiment. (There are some small differences between the GLOB and IPAC climatologies but these differences are not large enough to have a significant impact on our results.)
The significance of an anomaly is gauged by comparing its value to a suitable measure of variance (see the appendix). In the case of the observations the total interannual variability, σ̂OBS, is used, and thus the significance measure is really just a pointer to those regions where the anomaly is large relative to other years. In the case of the model simulations, however, a more meaningful measure of significance is possible. A signal-to-noise ratio is computed by calculating the ratio of the ensemble mean anomaly (which provides an estimate of the signal) to an estimate of the internally generated atmospheric variability (which is the noise). Details of how the latter quantity is estimated are provided in the appendix.
Most of the model analyses we present are based on ensemble mean anomalies. Additional information is available, however, in the individual ensemble members. In section 3c, we compare the ensemble members, and the observations, using a simple pattern correlation statistic. The statistic is computed as the area-weighted correlation of anomalies at all grid points in the North Atlantic region (20°–85°N, 110°W–35°E). In order to compute this statistic, the observations are interpolated to the model grid using bilinear interpolation. While the simplicity of a pattern correlation may mask variations in the amplitude of the forced response, or confuse the comparison when there are subtle variations in the phase of a pattern, it provides a zeroth-order way of quantifying the contribution of individual ensemble members. A more extended use of the information available from individual members will be the subject of further research.
c. Selection of ENSO events
A preliminary step in our study is to categorize ENSO years of interest. There is no single universally accepted ENSO index, but all the common indices are highly correlated. We employ the Niño-3.4 SST index (e.g., Trenberth 1997), being the average SST in a box located in the tropical eastern Pacific (5°S–5°N, 170°–120°W). During our period of investigation, 1986–2001, strong fluctuations in this index indicate that several anomalous warming (EN) and cooling (LN) events occurred in the tropical Pacific Ocean, with extreme values between −2° and +3°C and a standard deviation of about 1°C.
Because past work suggests that the influence of ENSO on the NAE region is quite weak, we employ moderately stringent criteria to select EN and LN events for study. ENSO events vary both in their amplitude and in their evolution. The evolution is important because the response in the NAE region lags, by several months, the changes in the tropical Pacific (e.g., Trenberth et al. 1998). We anticipate, therefore, that the largest impact in the NAE region will be for those events that are both strong and mature. With this expectation in mind, we select ENSO events that simultaneously satisfy two criteria: (i) |Niño-3.4| ≥ 1°C on average during the winter season (DJF) and (ii) |Niño-3.4| ≥ 0.65°C on average during the preceding autumn season [September–October–November (SON)]. The precise threshold values are arbitrary but have been chosen on the basis that the typical interannual variability of SST in the Niño-3.4 region is about 1°C. The DJF criteria selects strong events while the SON criteria selects those events that are comparatively mature. With the above criteria, three warm winters (EN1: 1987/88, EN2: 1991/ 92, EN3: 1997/98) and three cold winters (LN1: 1988/ 89, LN2: 1998/99, LN3: 1999/2000) qualify over the 1986–2001 period. Our list of ENSO events is consistent with those listed in other studies (Pozo-Vasquez et al. 2001; Hoerling and Kumar 1997) but generally constitutes a subset because of our more stringent double-season criteria.
d. SST forcing
The winter distributions of sea surface temperature anomalies (SSTA) associated with the above-selected ENSO events are displayed in Fig. 1. Note that we show global SSTA because we are not solely interested in the conditions found in the tropical Pacific Ocean. The contours show SSTA relative to the 1986–2000 climatology. In order to gauge the magnitude of the anomalies relative to interannual variability, we shade the regions where SSTA exceed one standard deviation of the local interannual SST variability. Lastly, the position of the 27°C isotherm is also shown in the figure. This contour provides an approximate guide to the regions of strongest convective activity where the atmosphere may be particularly sensitive to small changes in SST (Spencer and Slingo 2003).
Indices defined over specific regions can be a useful way to quantify key differences between fields. Table 1 shows a set of indices that measure some important aspects of the SST forcing for the six ENSO events. A basic measure of the ENSO intensity is provided by the Niño-3.4 index (NINO: 5°S–5°N, 170°–120°W), while the tropical western Pacific (TWP: 15°S–15°N, 120°– 155°E) index gives additional information about the SST forcing over the west Pacific warm pool. The tropical North Atlantic (TNA: 0°–20°N, 70°–10°W) index provides a measure of the state of the tropical Atlantic Ocean in the region influencing the convection in the intertropical convergence zone (ITCZ) and over South America. The last two indices in Table 1 provide measures of the state of the whole tropical belt, and will be refered to later. The integrated warm pool (IWP) index is the integrated value of SSTA within the Tropicswide warm pool, defined as the region where SST exceeds 27°C. The tropical belt (TRB: 25°S–25°N, 180°) index for GPH anomalies gives an idea of the zonal mean atmospheric response in the Tropics.
As seen in Figs. 1a,b,c, all the LN events show qualitative similarities in the Indo-Pacific region. In the equatorial Pacific, negative SSTA stretch from the coast of South America west to about 160°E. Farther west positive SSTA are found, and extend into midlatitudes on the western side of the North Pacific. In the Indian Ocean, SSTA are near zero for all three events.
Notwithstanding these qualitative similarities, there are important quantitative differences between the events. As seen in Table 1, the SSTA in the Niño-3.4 region is much bigger during LN1 than during LN2 and LN3. Furthermore, despite the similarity implied by the Niño-3.4 index, the SSTA pattern for LN3 appears considerably more diffuse than that for LN2 (as might be expected because LN3 is a decaying phase). The maximum equatorial SSTA is located farther east in LN3 than in LN2. In addition to the differences in Pacific SSTA, Fig. 1 shows significant differences in Atlantic SSTA between LN events. LN1 and LN3 exhibit very weak SSTA over most of the Atlantic, while LN2 exhibits strong warm anomalies in the vicinity of the Iberian coast (≃1.5°C) and the United States. (≃1°C). Differences also exist in the tropical Atlantic, where warmer-than-average conditions prevail during LN2 and LN3 while cooler-than-average conditions are encountered during LN1 (see the TNA index in Table 1).
Like the LN events, the EN events (Figs. 1d,e,f) all exhibit a similar pattern of positive SSTA in the tropical Pacific, but there are differences in the magnitude and center of action. EN3, known as the El Niño of the century, is associated with very warm water in the eastern equatorial Pacific and along the coast of South America. EN2 and EN1 present a more modest warming in the tropical Pacific and the maximum SSTA is located farther west. There also exist differences in the tropical Atlantic, which is warmer than average during EN1 and EN3 but cooler during EN2 (see the TNA index in Table 1). EN2 also differs from EN1 and EN3 by the presence of strong positive SSTA east of Newfoundland (about 1.5°C).
In summary, while the SSTA patterns associated with individual EN and LN events show qualitative similarities in the tropical Pacific, there is strong interevent variability in the global pattern of SSTA. Significant differences are seen in both the tropical Pacific and in the tropical Atlantic. We note that in three events (EN1, EN3, and LN1) the SSTA in the tropical eastern Pacific and tropical Atlantic are in phase. However, the remaining three events (EN2, LN2, and LN3) exhibit an opposition of phase between these regions.
3. The response to ENSO over the NAE region
a. Composite view
ENSO teleconnections are first examined through composite lenses. Composite analysis focuses attention on the common features of different events, albeit at the cost of obscuring interevent differences. Here, EN, LN, and ENSO (EN − LN) composites are displayed in Figs. 2 and 3 for observations and model. The ENSO composite of observed GPH (Fig. 2c) exhibits a tripolar pattern with positive anomalies over the tropical belt (maximum of about 20 m) and Greenland Sea (maximum of 90 m) associated with negative anomalies (maximum of 80 m) extending from the United States to southern Europe across the Atlantic subtropical gyre. The dipole structure of the GPH response over North America resembles the extension of the PNA pattern into the region (Horel and Wallace 1981). The structure of the anomaly over Europe, featuring a dipolelike pattern over western Europe, is broadly consistent with the composite study of Fraedrich and Müller (1992).
The ENSO–NAE link appears to be significant in the ENSO composite as the magnitude of the anomalies exceeds the typical range of interannual variability over the whole North Atlantic. The ENSO composite of observed precipitation (Fig. 2f) clearly shows the classical tropical EN signature, with anomalously wet conditions over the eastern Pacific, and anomalously dry conditions over the western tropical Atlantic and South America (−4 mm day−1). In the midlatitude North Atlantic the ENSO response exhibits a small but significant increase of precipitation (+1.5 mm day−1) in a band extending from the southeastern United States to southern Europe, where there are strong meridional gradients in GPH (Fig. 2c). This result is consistent with Fraedrich (1990), who argues that the increase in precipitation results from a strengthening of the North Atlantic storm track associated with enhanced westerly winds.
The EN (Fig. 2a) and LN (Fig. 2b) composites generally exhibit a weaker and less significant response and than the ENSO composite. The EN GPH response appears roughly opposite to the LN GPH response but there are differences in the magnitude, position, and significance of extrema. In the tropical Atlantic and over Africa, the response to EN is more significant than the response to LN, whereas in midlatitudes the response to LN is more significant. The magnitude of the midlatitude center of action is also greater in LN (∼60 m) than EN (∼20 m). The EN (Fig. 2d) and LN (Fig. 2e) patterns of precipitation show considerable antisymmetry in the Tropics, which suggests linear behavior in this region. In midlatitudes, there are significant anomalies in LN that extend across the Atlantic basin while in EN the response is rather less coherent.
Figure 3 shows the results of a similar composite analysis for the model simulations. Comparison between the observed (Figs. 2c,f) and simulated (Figs. 3c,f) ENSO composite suggests the model has significant skill, at least in the composite sense. Both composites reproduce the tripolar structure in GPH over the North Atlantic, and a dipole pattern in the precipitation response (dry Tropics, wet midlatitudes). The magnitude of GPH anomalies in the model simulations is smaller than that in the observations, as is expected because the former is an ensemble average. The simulated EN and LN GPH composites capture the result from observations that the EN response is more significant over the tropical Atlantic whereas the LN response is more significant in the midlatitude North Atlantic. In the regions where the signal is significant there is generally a good correspondence between the model results and the observations. The simulated EN and LN PREC composites (Figs. 3d,e) also show a good correspondence to the observed composites (Figs. 2d,e).
The results of the composite analysis have identified some asymmetries between EN and LN, and have indicated that the model simulations have significant skill in capturing key features of the observations. Composite analysis does, however, obscure the differences between individual events, which are the particular focus of this study. Hence, we now move on to analyze these differences.
b. The response to individual EN and LN events
In this section we examine the atmospheric response to individual EN and LN events. We study the differences between events both in the observations and in the model simulations. In addition, we examine the skill of the model in reproducing the atmospheric anomalies that were observed.
Figure 4 shows the observed and simulated GPH anomalies during the three LN events. As in the composite (Fig. 2c), the observed GPH anomalies (Figs. 4d,e,f) all show a positive (anticyclonic) center of action over the midlatitude North Atlantic, surrounded by negative anomalies (or weaker positive anomalies) to the north, south, and east. The exact location and shape of the positive center differs, however, between events. In LN1, a band of positive anomalies stretches from the southeastern United States and penetrates well into Europe. In LN2 the band is broader and there is little penetration into Europe. LN3 resembles LN2 over Europe but, rather than stretching back to the southeastern United States, the positive anomalies appear more localized over the Atlantic Ocean. These observed differences in the GPH anomaly patterns may arise simply from the internal variability of the atmosphere, or they may reflect differences in the SST forcing.
Model simulations make possible identification of those regions where the SST forcing is important. In Figs. 4a,b,c the shaded regions indicate a statistically significant response to the SST forcing. The fact that, in these significant regions, there are notable differences in the GPH anomaly patterns implies that these differences do arise, at least in part, from differences in the SST forcing and not simply from atmospheric internal variability.
One of the regions where there are significant differences between the three simulated LN events is in the Tropics. Because the gradients of GPH are always weak in the Tropics, these differences do not appear prominently in Fig. 4, although close inspection does indicate a weaker response over the tropical Atlantic for LN2 than for the other two events. Inspection of the pattern of global GPH anomalies for LN2 shows that the strongest response in the Tropics is found at ∼160°W, where Fig. 1b indicates that the largest negative SSTA are found. By contrast, for LN1 and LN3 relatively strong negative GPH anomalies are found throughout the whole tropical strip (results not shown). These negative anomalies are a response to the reduced deep convection over the central equatorial Pacific, the impacts of which are communicated efficiently throughout the global Tropics by equatorially trapped Kelvin and Rossby waves. The differences in the Tropics-wide response to the three events are quantified in Table 1. The column labeled TRB provides a measure of the simulated zonal mean GPH response in the Tropics, and shows that the weakest tropical response is found for LN2, and the strongest for LN1. The explanation of the relative weakness of the tropical response in LN2 must lie in the pattern of SSTA shown in Fig. 1b. Comparison with Fig. 1a (for LN1) and Fig. 1c (for LN3) shows that LN2 was unusual in having relatively large positive SSTA in the western tropical Pacific, where deep convection is usually most active (see the TWP index in Table 1). A plausible hypothesis is that enhanced convection in this region competes for influence on the Tropicwide response with the reduced convection farther east. It appears that, in contrast to the other two LN events, in LN2 there is a high degree of cancellation between these two influences in the Tropics, and this cancellation leads to the weak Tropics-wide response. A simple quantitative measure of the forcing of the Tropicwide response is provided by the IWP index in Table 1. As previously explained, the IWP value is the integral of SSTA within the Tropics-wide warm pool, defined as the region where SST exceeds 27°C. The spatial integral captures the potential for cancellation between the effects of anomalous convection in different parts of the Tropics. The high correlation of 0.87 between the IWP and TRB indices indicates that the IWP index is a useful predictor of the Tropics-wide response.
Returning to Fig. 4, comparison of the observed and simulated anomalies provides a means to assess the model skill in capturing the response to individual LN events. Some of the key differences between the observed anomaly patterns are indeed captured by the model. In LN1, for example, the model captures the band of positive anomalies stretching from the southeastern United States into Europe, and in LN3 the model captures the more localized center of action over the North Atlantic Ocean. The agreement between the model simulations and observations for LN1 and LN3 is generally very good, implying a high degree of potential predictability. For LN2 there is also good agreement over the western part of the domain, but over Europe the agreement is poor. The fact that the model simulates significant positive anomalies over central and eastern Europe, whereas negative anomalies were observed, may point to errors in the HadAM3 model (Pope et al. 2000; but see also the last paragraph in this section).
Figure 5 shows the observed and simulated GPH anomalies during the three EN events. There are striking differences between the observed anomaly patterns (Figs. 5d,e,f), particularly in the midlatitudes. Whereas for EN1 and EN3 the major feature over the North Atlantic Ocean is a center of negative anomalies, EN2 is dominated by very large positive anomalies centered in the United Kingdom. The differences between these patterns explain why the composite analysis (Fig. 3a) indicated no significant response to EN over the mid-and high-latitude North Atlantic. There are also notable differences in the model response to the three EN events. The results indicate a significant SST-forced response for all three events but, whereas the anomaly patterns for EN1 and EN3 are similar, the pattern for EN2 is quite different. EN2 shows only a marginally significant response over the tropical Atlantic, and a pattern of anomalies that is suggestive of a wave train propagating from the Gulf of Mexico northeastward toward Europe. Comparison of the model results with the observations suggests that, in the NAE region, the model had significant skill for EN1 and EN3, but the prominent positive anomalies over the United Kingdom in EN2 were not captured. The model actually predicted anomalies of the opposite sign in this region, which—as previously—may point to model error.
Table 1 (column TRB) shows that the Tropics-wide response is weaker in EN2 than in EN1 and EN3. In this sense, EN2 is reminiscent of LN2, and inspection of Fig. 1 suggests that the explanation may be the same. In the west Pacific on the equator, negative SSTA greater than 0.5°C are present in EN2, whereas in EN1 and EN3 there are no significant negative anomalies in this region. Furthermore the IWP index (Table 1) has a much smaller value for EN2 than for the other two events. It seems likely, therefore, that in EN2 like LN2 the relative weakness of the tropical response is explained by competition between different regions of anomalous convection.
Lastly, we note that the poor agreement between model and observations that we found over Europe for LN2 contrasts with the results of Dong et al. (2000), who found good agreement in this region for the same event. We have investigated the causes of this discrepancy and found that it stems from the use of a different model climatology. Rather than following our procedure (described in section 2), Dong et al. (2000) used a climatology that was derived by forcing the atmosphere model with climatological SST data obtained by averaging over the period 1961–90. We believe that the procedure we adopted in this paper is superior because it should be more directly comparable to the observed climatology (which can only be computed by directly averaging the atmospheric fields over a number of different years). Nevertheless, the discrepancy between our result for LN2 and that of Dong et al. (2000) highlights the fact that model skill inferred by comparison between simulated and observed anomalies may be sensitive to the climatology employed.
c. Examination of individual ensemble members
Thus far we have considered a qualitative comparison between the ensemble mean of the model simulations and the observations. In this section we consider a more quantitative analysis and investigate what additional insights can be gained by analysis of the individual ensemble members. Recall that the ensemble mean is an estimate of the SST-forced signal, with the noise of internal climate variability minimized in the averaging procedure. The observations (and any one ensemble member), however, can be decomposed conceptually into the forced signal and a potentially large component of random internal variability. If the model ensemble mean does not resemble the observations, it may be because the model is in error, or it may be that the forced signal in the observations is obscured by internal noise. The extent of agreement (or disagreement) among the individual ensemble members is a measure of this noise and thus gives information about which of these scenarios is more likely. Is the ensemble mean simply composed of the residual of a finite number of essentially random atmospheric states, or is the response strong enough to significantly shift the climatological probability density function and result in the signal being clearly seen in each ensemble member?
To address this question, we compare the winter GPH anomalies for each ensemble member and the observations with the model ensemble mean for the three EN and three LN cases using a simple pattern correlation statistic in the North Atlantic region. Correlations are shown in Fig. 6. In the case of EN1, each ensemble member is positively correlated with the ensemble mean indicating that there is a relatively high signal-to-noise ratio for the forced response, although the range of correlations (from 0.1 to 0.8) indicates that there is certainly a small possibility that the signal might be obscured by internal variability. In fact, the observations for this event are well correlated with the ensemble mean. For EN3, the signal-to-noise ratio of the forced response is even higher, with all ensemble members, and the observations, being highly correlated with the ensemble mean. This result indicates a high degree of potential predictability. For the EN2 case, the ensemble members indicate a similar level of correlation with the ensemble mean as was seen in the EN1 case. However, for EN2, the observations are negatively correlated with the ensemble mean. This may of course be a chance occurrence (i.e., the SST-forced signal in the observations was overwhelmed by a large random fluctuation of the atmospheric circulation) but the contrast between the observations and the individual ensemble members suggests it is more likely that this case points to errors in the model's response to the SST forcing.
For the LN cases the situation is more complicated. For LN1, 9 out of 10 of the ensemble members are positively correlated with the ensemble mean but there is 1 member that has a negative correlation. If we imagine that this ensemble member was the real observed climate, then we might have concluded (as in EN2) that there was a significant error in the model response. In reality, the observed GPH anomalies were positively correlated with the ensemble mean suggesting predictability. The situation is more exaggerated in the LN2 and LN3 cases; one ensemble member is negatively correlated with the ensemble mean and, in addition, other members have low correlations. While for all the LN cases the observed anomalies are positively correlated with the model ensemble mean, the notable spread of correlations within the ensemble indicates that the forced signal is more prone to contamination by random internal variability. This result suggests lower predictability for the LN events in our sample than for the EN events EN1 and EN3.
We investigated the sensitivity of the results shown in Fig. 6 to details of the analysis procedure. We varied the domain over which the correlation calculations were performed, and found that excluding North America (domain 20°–80°N, 60°W–35°E) did not alter our conclusions. Cross validating the correlations (by removing the appropriate ensemble member from the calculation of the ensemble mean) only changes the magnitude of the correlation by, on average, 0.1 and thus does not affect the main conclusions drawn from this figure.
Lastly, to provide further quantitative evidence that the model has skill in predicting interevent differences, we compared the skill of the GLOB simulations with that achieved by specifying the model EN or LN composite response. For one event (EN2) neither the GLOB simulations nor the composite shows positive correlation skill. Of the other five events, four show higher skill in GLOB than in the composite. The average skill of the GLOB simulations is 0.53, whereas the average skill of the composite is 0.46.
4. The role of Pacific and Atlantic SST forcing
Our analyses thus far have shown that SSTA associated with ENSO can significantly shift NAE climatic conditions above the level of natural variability, and that the response to individual EN and LN events is not always the same. In this section our aim is to understand some of the reasons why the atmospheric response in the NAE region varies between events. In particular, we investigate the role of SSTA in Indian and Pacific Oceans versus those in the Atlantic Ocean. To identify the respective roles of SSTA in the different ocean basins we make use of the IPAC experiment in which SSTA were restricted to the Indian and Pacific Ocean basins and climatological SST was used in the Atlantic Ocean. Differences between the results from IPAC and the model climatology provide information about the influence of SSTA in the Pacific and Indian Oceans, while differences between the results from GLOB and the results from IPAC provide information about the role of Atlantic SSTA. We refer to the latter differences as ATL. The results for IPAC and ATL for the three LN and EN events are presented in Fig. 7 and Fig. 8, respectively.
An alternative way to assess the influence of Atlantic SSTA would be to perform an additional experiment in which SSTA were restricted to the Atlantic basin and climatology was imposed in the Pacific and Indian Oceans. We cannot assume, however, that the response to global SSTA will be a linear sum of the response to Indian–Pacific SSTA and the response to Atlantic SSTA, because nonlinear interactions are very likely. The way in which we compute the Atlantic influence takes into account not only the direct effect of Atlantic SST but also any nonlinear interactions with the effects of Indian/Pacific SST.
Before embarking on discussion of the results, it is worth briefly reviewing the mechanism via which remote SSTA, particularly in the Tropics, may influence the atmosphere of the NAE region. When SSTA arise in tropical regions where the mean SST is high (above about 27°C) the atmospheric response typically involves changes in deep convection above, or nearby, the anomalous SST (e.g., Spencer and Slingo 2003). Associated with the changes in convection is anomalous diabatic heating, an anomalous divergent circulation, and a “Rossby wave source” (e.g., Hoskins and Ambrizzi 1993; Ambrizzi and Hoskins 1997). As the name suggests, the Rossby wave source acts as a forcing for Rossby waves that can propagate into the extratropics, and thereby give rise to remote climate impacts. An important point to appreciate in the present context is that the precise form of the extratropical response can be very sensitive to details of the SSTA. This sensitivity arises from a number of factors. First, the convection responds to absolute rather than anomalous SST, thus the location and sign of anomalies may be as important as their magnitude. Second, the excitation of a Rossby wave disturbance is sensitive not only to the changes in convection but also to the background flow with which the anomalous divergent circulation interacts. Third, the propagation of waves critically depends on the shape and extension of the forcing. Waves generated from a zonally elongated source tend to propagate meridionally, while waves generated from a more zonally restricted source tend to be refracted more strongly back toward the equator (Ambrizzi and Hoskins 1997). Finally, a Rossby wave disturbance that propagates into the extratropics may trigger instabilities that totally reshape the final response. These four factors make it difficult to anticipate reliably, without conducting model experiments, what the extratropical response to a particular pattern of SSTA will be.
a. LN response
Figure 7 shows the results from IPAC and ATL (GLOB–IPAC) for the three LN events. A first impression indicates that it is SSTA in the Indian and Pacific Oceans that dominate the response in the NAE region. In addition, the pattern of the IPAC response is quite similar for the three events. There is a horseshoe-shaped pattern of positive height anomalies over eastern North America that may be the tail end of a Rossby wave train emanating from the tropical Pacific (e.g., Hoerling and Kumar 1997). These positive anomalies reach eastward across the midlatitude Atlantic Ocean and appear to link with a further center of positive anomalies over southern Europe. The similarities between the IPAC results for the three LN events suggest that the differences between the Indian and Pacific SSTA (Fig. 1) do not have a major impact on the NAE region. This said, there are some differences. The largest GPH anomalies over North America are seen for LN1, which may be a simple consequence of the fact that the Pacific SSTA are largest for this event (see the NIÑO column in Table 1). The strongest response over the high-latitude North Atlantic is seen in LN3.
The ATL results in Fig. 7 indicate a significant role for Atlantic SSTA in LN3 and LN1 but only a marginally significant role in LN2. To help understand these results we present in Fig. 9 the ATL precipitation anomalies for the three LN and EN events. For LN3, Fig. 9 indicates that the Atlantic SSTA caused a significant increase in precipitation over South America and the western tropical Atlantic. This increase in precipitation might have been anticipated from the presence of significant positive SSTA in the tropical Atlantic at this time (Fig. 1c; Table 1). Associated with the enhanced precipitation will be a Rossby wave source, and it is likely that the GPH anomalies seen in Fig. 7f are the extratropical response to this tropical forcing. The pattern of GPH anomalies suggests a wavelike disturbance with limited zonal extent that is refracted back toward the equator at around 60°N. Behavior of this type is precisely what is predicted by Rossby wave theory (Ambrizzi and Hoskins 1997) for a situation—such as here— where the zonal extent of the forcing is confined.
For LN1, Fig. 9 indicates a rather noisy precipitation field. In the tropical Atlantic the main signal is a modest reduction in precipitation over northern South America, with a modest increase over eastern South America (the Nordeste region). This dipole signature is expected from the weak SSTA dipole that can be seen in Fig. 1a. Although the precipitation anomalies are weak, they may play a role in forcing the circulation response over the extratropical Atlantic. Figure 7d suggests a possible Rossby wave train propagating from the southern Carribean region. Alternatively, it could be that SSTA at higher latitudes played a role. The atmosphere is, generally speaking, less sensitive to SSTA in the extratropics than to SSTA in the Tropics, but Fig. 1a shows that during LN1 there was a significant positive SSTA south of Greenland, with significant negative anomalies (possibly associated with sea ice anomalies) in the Greenland–Iceland–Norwegian Seas region to the northeast. The enhanced local baroclinicity associated with these SSTA may have contributed to the forcing of the negative height anomalies centred over Iceland in Fig. 7d.
For LN2, Fig. 9 indicates an increase in precipitation over eastern South America and also over the equatorial Atlantic Ocean. These signals are consistent with the positive SSTA in the equatorial Atlantic during this event (Fig. 1b; Table 1). We might expect that, as in LN3, the tropical precipitation anomalies should be associated with an atmospheric response over the extratropical North Atlantic. However, the extratropical signal in Fig. 7e is weak and only marginally significant. The reason why Atlantic SSTA appear to have a less significant influence on the extratropics in LN2 than in LN3 is not clear. As was noted earlier, however, the excitation of an extratropical response can be very sensitive to details of the tropical forcing. Although the precipitation anomalies over the tropical Atlantic region are similar in LN2 and LN3, there are differences. In addition, there are differences in the mean flow that arise from the response to the different SSTA in the Pacific and Indian Oceans. Together these two factors may explain why Atlantic conditions appear to have had a stronger influence in LN3 than in LN2. An alternative (admittedly speculative) possibility is that the large SSTA in the extratropical North Atlantic during LN2 acted to counter the influence from the Tropics. Further investigation of these issues, and especially the relative importance of the different factors involved, will require additional experiments and is an important area for future work.
b. EN response
Figure 8 shows the results from IPAC and ATL (GLOB–IPAC) for the three EN events. As for the LN events, SSTA in the Indian and Pacific Oceans appear to be the dominant influence, but the pattern of the IPAC response is less consistent between the EN events. For example, in EN2 and EN3 there are significant anomalies over North America that resemble the tail end of the PNA (Hoerling et al. 1997), but in EN1 there are no such anomalies. Over the North Atlantic, the GPH anomaly patterns for the three events are all quite different. The wide range of results from the IPAC experiment for the three EN events implies that the differences in the SSTA in the Indian and Pacific Oceans (Fig. 1) have an important influence on the atmosphere.
The lack of a significant response over North America in EN1 can be explained by the weak SSTA in the tropical Pacific during this event. Although the Niño-3.4 SSTA in EN1 is only 30% smaller than the corresponding anomaly in EN2, the precipitation anomaly (which is a highly nonlinear function of the SST) is 5 times smaller (not shown). Thus, the anomalous diabatic heating in the central Pacific during EN1 is much weaker than for the other events, and the associated response over the PNA region is also much weaker. Differences between EN3 and the other events are likely to be a consequence of much larger SSTA in the tropical Pacific during this event. In addition, however, there may be a role for SSTA in the Indian Ocean. Significant positive SSTA were found in the Indian Ocean during EN3, and there is evidence from other studies (Spencer and Slingo 2003) that anomalies in this region can have an influence on the North Atlantic.
The ATL results in Fig. 8 indicate a significant role for Atlantic SSTA in all three events, with the strongest influence in EN1. As for the LN events we use the ATL precipitation response, shown in Fig. 9, to help interpret these results. For EN1, Fig. 9 indicates a dipole pattern of precipitation anomalies in the tropical Atlantic, which implies a southward shift of the ITCZ, and an increase in precipitation over eastern South America. The precipitation anomalies arise in response to the SSTA in the tropical Atlantic. Figure 1 shows that positive SSTA were present (see also the TNA column in Table 1), with the largest anomalies located just south of the mean position of the ITCZ. As previously, we hypothesize that the extratropical GPH anomalies (seen in Fig. 8d) arise in response to the anomalous convection over the tropical Atlantic region (seen in Fig. 9d). In the case of EN1 the precipitation pattern indicates a zonally extended Rossby wave source. In this situation, Rossby wave theory (Ambrizzi and Hoskins 1997; Hoskins and Ambrizzi 1993) predicts an extratropical disturbance that also has large zonal extent and that tends to propagate meridionally with comparatively little refraction by the background flow. The pattern of anomalies in Fig. 8d is consistent with such a disturbance, and may be constrasted with the case—in LN3—of more zonally confined forcing.
Further support for the hypothesis that GPH anomalies seen in Fig. 8d arise in response to the SSTA in the tropical Atlantic comes from the study of Sutton et al. (2000) in which an atmosphere model (the same as used in the present study) was forced with an idealized tripole pattern of North Atlantic SSTA. Although this idealized pattern differs from that which arose during EN1 there are common features, in particular the presence of SSTA of one sign throughout the tropical North Atlantic region, and anomalies of opposite sign in the subtropics. The pattern of response in Sutton et al. (2000) closely resembles the pattern of anomalies seen in Fig. 8d with, in both cases, positive SSTA in the TNA region being associated with negative GPH anomalies over the midlatitude North Atlantic. Sutton et al. (2000) demonstrated the key role of anomalous tropical convection in forcing the midlatitude response, and the agreement between their results and the present results therefore lends weight to the suggestion that a similar mechanism is responsible in both cases.
For EN2, Fig. 9 shows negative precipitation anomalies in the ITCZ and over eastern south America, consistent with the negative SSTA in the tropical Atlantic (Fig. 1e; see TNA in Table 1). As the precipitation pattern is almost opposite to that found in EN1 we might expect—to the extent that the behavior is linear—an anomalous Rossby wave source with the opposite sign. In reality nonlinearities will certainly be present, but it is interesting that the pattern of GPH anomalies over the North Atlantic in EN2 is, broadly, opposite to that found in EN1: there are positive GPH anomalies in midlatitudes sandwiched by negative anomalies to the north and south. It may be, therefore, that linear theory provides the first-order explanation for the different impacts of Atlantic SSTA in EN1 and EN2.
For EN3, Fig. 9 indicates positive precipitation anomalies over eastern South America and over the equatorial Atlantic Ocean, as would be expected from the positive SSTA seen in the tropical Atlantic during this event (Fig. 1f). The arguments given above might lead us to expect a pattern of GPH anomalies over the North Atlantic that is similar to that seen in EN1, with negative anomalies in midlatitudes sandwiched by positive anomalies to the north and south. In fact, such a pattern can be seen in Fig. 8f, although over the extratropics it is not significant. As in the differences between LN2 and LN3, it seems likely that the differing influence of Atlantic conditions in EN3 and EN1 is attributable to subtle differences in the anomalous Rossby wave source.
Our results demonstrate clearly that SSTA in both the Indo-Pacific basin and the Atlantic basin exert a significant influence on the atmospheric circulation over the NAE region during ENSO events. The relative importance of Atlantic and Indo-Pacific SST varies from event to event, but typically both play a significant role. Table 1 gives a summary of the Indo-Pacific and Atlantic SST forcing but, as we have seen, the value of a single index, or even several indices, does not provide a full characterization of the important features in an SSTA field. In reality, the sign and spatial distribution of anomalies may be as important for determining the response as is the magnitude. We must keep this complexity firmly in mind as we consider the extent to which any common patterns of behavior emerge from comparison of the different events.
We have argued that in all cases the pattern of atmospheric response over the NAE region is primarily forced by changes in tropical convection, which are themselves a response to tropical SSTA. In the tropical Atlantic region, SSTA influence convection over South America and in the Atlantic ITCZ in a fairly direct way, that is, positive SSTA tend to enhance convection and negative SSTA tend to suppress convection. Beyond this basic behavior, however, there are significant differences between individual events. An illustration of the limitations of using a single index for Atlantic SST can be made by comparing the events LN2, LN3, and EN3. For each of these events the TNA SST index had a value between 0.25° and 0.35°C, yet Figs. 9b,c,f show a considerable variety of precipitation patterns. The actual convection anomalies that arise in the tropical Atlantic region are influenced not only by Atlantic SST but also by Indo-Pacific SSTA. During El Niño convection tends to be suppressed over South America and the Atlantic ITCZ, whereas during La Niña it tends to be enhanced. It follows that when Niño-3.4 and TNA SST have the same sign (e.g., LN1, EN1, EN3) there tends to be competition between the Atlantic and Pacific influences, whereas when these indices have opposite sign (e.g., LN2, LN3, EN2) their effects tend to reinforce. This behavior is another example of the comparatively linear nature of large-scale dynamics in the Tropics. We have argued that a similar type of competition—between SSTA in the western Pacific (“warm pool” region) and SSTA in the central Pacific—may determine the strength of the response over the whole Tropics.
By contrast with the linear, and comparatively predictable, behavior of the Tropics, our results suggest that the response over the NAE region is much more sensitive to details of the SSTA. Nevertheless, there is evidence of recurrent patterns. SSTA in the tropical Pacific generally excite a PNA-like pattern, the tail end of which extends over eastern North America. More noteworthy is our evidence regarding Atlantic SSTA. Results for the three EN events (and perhaps also LN2) suggest that positive SSTA and enhanced convection in the ITCZ over the tropical Atlantic are associated with a banded pattern of GPH anomalies over the North Atlantic, in which negative anomalies are sandwiched by positive anomalies to the north and south. A GPH pattern of the opposite sign is associated with negative SSTA and reduced convection. This consistency is encouraging, but we must keep in mind the small sample of events considered.
Finally, following a suggestion that arose in the review process, we performed an additional calculation to further quantify the role of Atlantic SSTs. We compared the skill of NAE 500-mb GPH in the GLOB and IPAC 17-yr simulations. Consistent with our claim that Atlantic SSTs play a significant role we found that GLOB is indeed more skillful than IPAC. The correlation score averaged over all years is 0.24 for GLOB compared to 0.08 for IPAC. For the event years analyzed the scores are 0.53 for GLOB compared to 0.31 for IPAC.
We have investigated the observed and simulated winter atmospheric response over the North Atlantic– European (NAE) region in six ENSO episodes that occurred during the 1986–2001 period. We have used ensemble simulations with an atmosphere GCM to examine how the NAE response varies between individual events, and to elucidate the respective roles of SSTA anomalies in the Indo-Pacific basin and in the Atlantic basin. Our key conclusions are as follows:
ENSO events have a significant influence on the climate of the NAE region.
The composite ENSO (EN − LN) response simulated by model is statistically significant over the NAE region, and shows a good correspondence with a composite of observations based on the same years. Consistent with previous studies Fraedrich (1990), the ENSO response involves a dipole pattern of pressure anomalies over the North Atlantic, with associated changes in the storm track. Relative to LN, EN favors warmer and wetter conditions in southern Europe. The existence of a significant NAE response is not simply a feature of the composite analysis; on the contrary, a significant response was found for all six of the individual ENSO events analyzed.
The impacts of El Niño and La Niña on NAE climate vary significantly between individual events.
Analysis of the NAE response to individual EN and LN events indicated common features (e.g., over eastern North America) but also significant differences in the pattern of observed and simulated anomalies. The model simulations demonstrated that these differences arise not simply from atmospheric internal variability but also because the atmosphere is sensitive to specific features of the SST anomaly field that characterize individual ENSO events. The fact that the SST influence is significant implies that the response to individual events is potentially predictable.
The variety in the atmospheric response to individual ENSO events highlights the limitations of composite analyses. As noted by Horel and Wallace (1981), composites are but “blurred images resulting from our inadvertent superposition of an ensemble of sharper patterns, corresponding to the various states of the equatorial atmosphere that have existed under the general category of warm [or cold] episodes.” In the present context, for several of the events considered in this study the atmospheric response differed considerably from the composite response.
Differences in the impacts on NAE climate of individual ENSO events arise partly from differences in Indo-Pacific SST and partly from differences in Atlantic SST.
Our model experiments have demonstrated that, during ENSO events, NAE climate is influenced not only by conditions in the Indo-Pacific Ocean but also by conditions in the Atlantic Ocean. Furthermore, the conditions in both ocean basins are important for understanding the differences between events of the impacts on NAE climate. The exact impacts that occur may be very sensitive to detailed aspects of the SST anomaly field. This sensitivity arises from the fact that tropical convection is sensitive to the absolute rather than anomalous value of SST, and that forcing of a Rossby wave response in the extratropics is sensitive to additional factors such as the background mean flow and the location and extent of anomalous convection.
Within the Atlantic basin, we have argued that SSTA in the TNA region are likely to be particularly important because they are best placed to modulate convection in the Atlantic ITCZ and South America. Our results suggest an association between SST anomalies in the TNA region and a banded pattern of geopotential height anomalies over the North Atlantic.
Our results have significant implications for efforts at seasonal climate forecasting for the NAE region. In the first place, our results are encouraging in that they indicate potentially predictable circulation anomalies in the NAE region during individual ENSO events. There is now a need to explore whether coupled (rather than atmosphere-only models) can reproduce, or even improve, the promising level of skill found in our simulations. In the second place, the fact that the climate impacts vary significantly between events means that details matter and one cannot rely on composites as a reliable guide to the impacts of any specific event. In the third place, our evidence for the importance of Atlantic conditions argues that, in the development of seasonal forecast systems, attention must not be focused too narrowly on the tropical Pacific Ocean. This point is further emphasized by the obvious recognition that Atlantic conditions are likely to be still more important during non-ENSO years.
We hope that this study highlights the need for more research to better understand the impacts of individual ENSO events on the NAE region. Much of the analysis in this paper has been qualitative and there is a need for more quantitative analyses to complement this work. There is also a need to investigate in greater detail the physical mechanisms via which SST anomalies throughout the world's oceans can influence NAE climate. In this paper we have focused especially on the role of tropical SST anomalies, but the potential role of extratropical SST anomalies should not be neglected and is worthy of further investigation. Progress in understanding the physical mechanisms is likely to require, among other approaches, more idealized model studies.
We are grateful to the NERC thematic research programme Coupled Ocean Atmosphere Processes and European Climate (COAPEC) and the Royal Society, which have funded the work of P.-P. Mathieu, Mat Collins, and Rowan Sutton, respectively. We are grateful to Brian Hoskins for his useful comments on this paper.
Statistics for Climate Variability
The purpose of this appendix [adapted from Rowell et al. (1995)] is to define the measures we use to assess the statistical significance of our results. In case of observations (subscript OBS), the interannual variance is straightforwardly calculated as
where Y tOBS is a generic variable observed at time t and the overbar corresponds to the time average over T years.
The observed signal is considered to be significant when
In case of a model experiment (subscript RUN for GLOB or IPAC of K ensemble members, the variance σ̂2EM of the ensemble mean (EM) can be expressed as the sum of an externally forced component (EXT) and a contribution from internally generated variability (INT) (Scheffé 1959):
The second term on the right-hand side arises because of the finite size of the ensemble. To assess the significance of an ensemble mean anomaly, we compare it to the magnitude of the second term, which effectively measures the sampling uncertainty. Specifically, the model response is considered to be significant when
where the internal variance is estimated from the time average of the intraensemble variance as follows:
This measure of significance is used for GPH and precipitation in the individual model outputs for GLOB (Figs. 4, 5), IPAC and ATL (Figs. 7, 8, 9). It is worth stressing that the above measure of significance is only applicable for a Gaussian variable. One could therefore argue that it is not appropriate for the precipitation field. However, we tested this possibility by repeating our analysis of the precipitation results after first applying a square-root transform to make the statistics more Gaussian and found that the results were hardly altered. It appears that, even for precipitation, seasonal mean averaging is sufficient to obtain approximately Gaussian behavior.
Current affiliation: European Space Agency, Earth Observation Science and Applications, Frascati, Italy
Corresponding author address: Dr. P.-P. Mathieu, European Space Agency, Earth Observation Science and Applications, Via Galileo Galilei, Casella Postale 64, 00044 Frascati, Italy. Email: firstname.lastname@example.org