Recent extreme events such as the devastating 2003 European summer heat wave raise important questions about the possible causes of any underlying trends, or low-frequency variations, in regional climates. Here, we present new evidence that basin-scale changes in the Atlantic Ocean, probably related to the thermohaline circulation, have been an important driver of multidecadal variations in the summertime climate of both North America and western Europe. Our findings advance understanding of past climate changes and also have implications for decadal climate predictions.
Instrumental records show that during the 19th and 20th centuries, there were marked variations on multidecadal time scales in the summertime climate of both North America (1–4) and western Europe (5). In the continental United States, there were significant variations in rainfall and drought frequency (1–4), and it has been suggested (1, 4) that changes in the Atlantic Ocean, associated with a pattern of variation known as the Atlantic Multidecadal Oscillation (AMO) (6, 7), were responsible. If confirmed, such a link would be important for climate predictions because the AMO is thought to be driven by the ocean's thermohaline circulation (6) and may be predictable (8, 9). However, thus far the evidence for an Atlantic link is mainly circumstantial, being derived from observations and showing correlation rather than causality. Clarifying whether AMO-related changes in the Atlantic Ocean were indeed responsible for the observed variations in North American summer climate and whether, in addition, there were impacts on other regions is therefore an important challenge.
Figure 1 shows the time series and pattern of North Atlantic sea surface temperatures (SSTs) that characterize the AMO during the period 1871 to 2003 (10). There are AMO warm phases in the late 19th century and from 1931 to 1960; cool phases occur from 1905 to 1925 and from 1965 to 1990. The spatial pattern shows anomalies of the same sign over the whole North Atlantic, with the largest anomalies (σ ∼0.3°C) found just east of Newfoundland.
To identify the climate variations associated with the AMO, we considered a simple composite difference of observational sealevel pressure (SLP), precipitation, and surface air temperature (SAT) data between the warm phase from 1931 to 1960 and the subsequent 30 years, 1961 to 1990, which were dominated by a cool phase of the AMO (Fig. 2, A to C). In the North Atlantic region, there are two prominent low-pressure anomalies, one centered over the southern United States (∼60 Pa) and the other centered just west of the United Kingdom (peaking at ∼150 Pa). The low-pressure anomaly over the southern United States is associated with precipitation reductions of up to 20% (0.1 to 0.3 mm/day), consistent with (1) and (4). Over western Europe, there is enhanced precipitation (0.1 to 0.3 mm/day, or 5% to 15% of the mean summer value). This multidecadal change in European precipitation has been previously documented (5) but has not been linked with the AMO. The SAT fields (Fig. 2C) show warm anomalies (0.25°C to 0.75°C) over the United States and also over central Europe. The precipitation fields (Fig. 2B) also show large positive anomalies in the Sahel region of North Africa, consistent with earlier work (11), and in the Caribbean.
A simple significance test (12) suggests that the major observed anomalies shown in Fig. 2 are unlikely to have arisen from internal fluctuations of the atmosphere. To investigate whether they arose in response to changes in the ocean, we first examined results from an ensemble of six simulations with an atmospheric general circulation model. These “C20” simulations were forced with historical global SST data for the period 1871 to 1999, and variations in the ensemble mean provide information about the ocean forcing of climate (13). Figure 2, D and E, shows the ensemble mean SLP and precipitation anomalies corresponding to the composite differences computed from observations. The SLP shows low-pressure anomalies centered over the southern United States and in the region of the United Kingdom that are in good agreement with the observations. The anomalies over the United States are very similar in magnitude to the observations (60 to 75 Pa), whereas the anomalies west of the United Kingdom are weaker in the model ensemble mean (∼50 Pa) than in the observations (∼150 Pa). The most likely reason for this discrepancy is a larger component of internal variability in the observations, a hypothesis supported by analysis of the ensemble spread (fig. S3).
The precipitation field (Fig. 2E) shows reduced precipitation over the United States and northern Mexico, and the magnitude of the anomalies (0.1 to 0.3 mm/day) is in agreement with the observations. Over western Europe, the model indicates enhanced precipitation consistent with the observations, but on the basis of our sample of six ensemble members, the anomalies are not statistically significant and are therefore not seen in the figure. The model shows large increases in precipitation in a tropical band stretching from the eastern Pacific, through the Caribbean and tropical Atlantic, to North Africa. The anomalies in the Caribbean and northern South America agree with the land observations in these regions, but the anomalies in the Sahel region are markedly weaker than is observed. This discrepancy might be a consequence of errors in the representation of land-surface feedbacks (14, 15).
The results from the C20 simulations provide strong evidence that the major North Atlantic features identified in the observations arise in response to changes in the oceans. Results from similar experiments with other atmospheric models are consistent with this conclusion (16). However, because in each case the model was forced with global SST fields, these experiments do not clearly demonstrate the role of the Atlantic Ocean. To clarify the role of the Atlantic, we forced the model with an idealized AMO SST anomaly pattern, based on the North Atlantic part of Fig. 1B (see fig. S1). For these experiments, SST anomalies did not vary in time, but integrations of 10 or 20 years' duration were carried out to separate the ocean's influence from atmospheric internal variability (12).
Figure 2, F to H, shows the response to the AMO SST pattern. Over the Caribbean, central America, and the United States, the SLP and precipitation fields show excellent agreement with both the C20 simulations and the observations. The low-pressure anomalies in western Europe are also reproduced. Over the United Kingdom, there are positive precipitation anomalies, which were not seen in the C20 simulations but are in the observations, whereas precipitation anomalies over the Sahel are again weaker than observed (17). The SAT fields (Fig. 2H) show a prominent warm anomaly over the southern United States and Mexico, in agreement with the observations, and also show positive anomalies in western Europe. Overall, Fig. 2 provides compelling evidence that the AMO has indeed been responsible for marked changes in the regional atmospheric circulation and for associated anomalies in precipitation and surface temperature over the United States, southern Mexico, and, probably, western Europe.
The fact that the AMO SST experiments and the C20 simulations give such similar results suggests that, for the particular decadal change considered (1931 to 1960 compared with 1961 to 1990), the Atlantic Ocean was the dominant oceanic influence on summertime climate in the regions considered. To investigate the importance of the AMO over a longer period of time, we examined the correlation between the AMO SST index (Fig. 1A) and indices of SLP for the U.S. and U.K. regions (12). The SLP indices were computed both from the observations and from the C20 simulations. All the SLP indices are significantly anticorrelated with the AMO index (fig. S2). These results confirm that the AMO is an important influence on both regions, but they also suggest that other influences are important, especially for the U.S. region (for which the correlation with observed SLP is lowest). The much higher correlation between the AMO index and simulated U.S. SLP could suggest that these other influences [e.g., associated with changes in the Pacific Ocean (3)] are underestimated in the HadAM3 simulations.
To understand better the AMO influence, we performed two additional model experiments, using as forcing patterns the tropical part (0°N to 30°N) and the extratropical part (30°N to 70°N) of the AMO SST pattern (see fig. S1). The results (Fig. 3) indicate that AMO-related climate anomalies over the United States and Mexico were forced by tropical Atlantic SST anomalies, whereas those in the region of western Europe were primarily (but not exclusively) forced by SST anomalies in the extratropics. The low-pressure center over the southern United States may be a response to the anomalous latent heating of the atmosphere implied by the enhanced precipitation in the tropical Atlantic (18). The low-pressure center near the United Kingdom may be a downstream response to the largest positive SST anomalies east of Newfoundland (19, 20). In Fig. 3, A and B, the impacts of the AMO are not restricted to the Atlantic basin but extend throughout the tropics. This finding supports suggestions (21, 22) that the Atlantic Ocean may be an important driver of multidecadal climate variability on a global, as well as regional, scale.
Overall, our results provide strong evidence that during the 20th century the AMO had an important role in modulating boreal summer climate on multidecadal time scales. We have focused here on time mean anomalies, but some of the most important impacts are likely to be associated with changes in the frequency of extreme events. There is evidence that the frequency of U.S. droughts (4) and the frequency of European heat waves (23) are both sensitive to Atlantic SSTs.
The results also have implications for the interpretation of instrumental and proxy (24) climate records, which relies on understanding and quantifying how regional climate is related to large-scale drivers such as the AMO. Progress in understanding these drivers is especially important as studies that address the attribution of observed climate changes to anthropogenic causes give increasing attention to regional scales (25). Our results suggest, for example, that the change in phase of the AMO in the 1960s may have caused a cooling of U.S. and European summer climate; a further change in the AMO may have contributed to recent warming in these regions.
In addition, our results have implications for predicting the climate of the next few decades. In the absence of anthropogenic effects and assuming a period of 65 to 80 years (1, 26), we should now be entering a warm phase of the AMO, as suggested by Fig. 1A. Our results would then suggest a forecast of decreased (relative to 1961 to 1990) summer precipitation (increasing drought frequency) and warmer temperatures in the United States together, possibly, with increased summer precipitation and temperatures in western Europe. In reality, anthropogenic effects on climate now appear to be important (27), so current and near future trends in North Atlantic regional climate may be shaped by a competition between the AMO and these anthropogenic effects. Moreover, these two influences may not add linearly. Models suggest that anthropogenic warming will lead to a slowdown of the Atlantic thermohaline circulation (27). If true of the real world, this could favor an earlier than expected shift toward the negative phase of the AMO, with a corresponding shift in the AMO influences on U.S. and European summer climate.
Supporting Online Material
Materials and Methods
Figs. S1 to S3