Early 20th century Arctic warming in retrospect

Authors

  • Kevin R. Wood,

    Corresponding author
    1. Joint Institute for the Study of the Atmosphere and Ocean, University of Washington, 3737 Brooklyn Ave. NE, Seattle, WA 98105, USA
    • Joint Institute for the Study of the Atmosphere and Ocean, University of Washington, 3737 Brooklyn Ave. NE, Seattle, WA 98105, USA.
    Search for more papers by this author
  • James E. Overland

    1. NOAA Pacific Marine Environmental Laboratory, 7600 Sand Point Way NE, Seattle, WA 98115, USA
    Search for more papers by this author

Abstract

The major early 20th century climatic fluctuation (∼1920–1940) has been the subject of scientific enquiry from the time it was detected in the 1920s. The papers of scientists who studied the event first-hand have faded into obscurity but their insights are relevant today. We review this event through a rediscovery of early research and new assessments of the instrumental record. Much of the inter-annual to decadal scale variability in surface air temperature (SAT) anomaly patterns and related ecosystem effects in the Arctic and elsewhere can be attributed to the superposition of leading modes of variability in the atmospheric circulation. Meridional circulation patterns were an important factor in the high latitudes of the North Atlantic during the early climatic fluctuation. Sea surface temperature (SST) anomalies that appeared during this period were congruent with low-frequency variability in the climate system but were themselves most likely the result of anomalous forcing by the atmosphere. The high-resolution data necessary to verify this hypothesis are lacking, but the consistency of multiple lines of evidence provides strong support. Our findings indicate that early climatic fluctuation is best interpreted as a large but random climate excursion imposed on top of the steadily rising global mean temperature associated with anthropogenic forcing. Copyright © 2009 Royal Meteorological Society

1. Introduction

The recent widespread warming of the Earth's climate is the second of two marked climatic fluctuations to attract the attention of scientists and the public since the turn of the 20th century. Pronounced high-latitude surface air temperature (SAT) anomalies accompanied by climate-related environmental change are hallmarks of both episodes (Figure 1). Evidence of wintertime warming in Northern Europe was first described early in the century by Ekholm (1901) and Wallén (1916). By the beginning of the 1930s, it was recognised that a climatic fluctuation had occurred over a broad area that had substantial social and economic impacts (e.g. Kincer, 1933). This realisation brought about a new perspective on climate variability and motivated decades of research into the causes and effects of the early climatic fluctuation, which Bengtsson et al. (2004) called one of the most puzzling climate anomalies of the 20th century.

Figure 1.

Geographical pattern of temperature change in winter for the 20-year period 1920–1939, based on station data beginning in 1880 (Willett, 1950). Units are °F/10

The early climatic fluctuation is particularly intriguing now because it shares some of the features of the present warming that has been felt so strongly in the Arctic, even though anthropogenic greenhouse gas forcing has become more of a factor in recent decades (IPCC, 2007). These appearances lend credence to the notion that present warming might be, in part, due to a naturally occurring multi-decadal oscillation in the climate system (e.g. Kerr, 2005; Knight et al., 2006; Keenlyside et al., 2008). Discussion of potential causes of the early climatic fluctuation, especially with respect to the relative and intertwined roles of internal or unforced variations in the climate system and other categories of natural and anthropogenic forcing, has been underway for more than 70 years and is still ongoing (Veryard, 1963; Hegerl et al., 2007). Nor has the long search for significant climatic cycles or oscillations been particularly successful (e.g. Brunt, 1937; Roe, 2009).

The evidence of a large climatic fluctuation around the North Atlantic beginning in the 1920s attracted considerable scientific interest at the time. We have sought out and reviewed the work of scientists who studied this event first-hand and used insights mined from the old literature to address three basic questions about the early climatic fluctuation. First, what is the best characterisation of its temporal and spatial development? As Ångström (1939) pointed out, a good understanding of the spatial domain was necessary to understand the causes of the early climatic fluctuation, yet recent literature is not entirely consistent on this point (e.g. Johannessen et al., 2004; Serreze and Francis, 2006). Second, how does variability in the atmospheric circulation, which was almost immediately identified as a leading factor in the early climatic fluctuation, relate to regional SAT and sea surface temperature (SST) anomalies, particularly at high latitudes? And third, if variability in atmospheric circulation was such a major factor for the early climate fluctuation, why did the largest SAT anomalies appear in regions of the Arctic and Northern Atlantic where the Arctic Oscillation/Northern Annular Mode (AO/NAM) or North Atlantic Oscillation (NAO) are poorly correlated with SAT anomalies? Answers to these questions help place the early climatic fluctuation in context within the spectrum of climate change and point the way toward an eventual discovery of its fundamental causes.

Our investigation shows that the course of SAT change in the Atlantic and Pacific sectors of the Arctic (Figure 2) depends on the time-varying superposition of major atmospheric circulation patterns. In the Atlantic sector meridional circulation patterns associated with high sea-level pressure (SLP) over Northern Europe were salient during the early climatic fluctuation. Parallel SAT changes in regions where dynamically driven temperature contrasts are the norm (e.g. the Greenland–Europe temperature see-saw) are congruent with fluctuations in SST anomaly patterns: specifically, SST anomalies in the neighbourhood of the Gulf Stream front and along the Pacific coastline of North America. These findings support the proposition that the early climatic fluctuation was principally due to intrinsic variability in the large-scale atmosphere/ocean/land system. Although the physics underlying multi-decadal variations in SST anomaly patterns is still debated, the evidence here favours atmospheric forcing as a primary driver. Thus, the early climatic fluctuation is best interpreted as a large but random excursion caused by the temporal superposition of climate patterns imposed on top of the steadily rising global mean temperature associated with anthropogenic forcing.

Figure 2.

Winter (DJFM) SAT anomalies from land-based stations north of 60°N latitude in the Atlantic sector (90°W–45°E) and Pacific sector (135°E–90°W) of the Arctic based on CRUTEM3v data (Brohan et al., 2006). The black curve is produced by a low-pass Butterworth filter constructed to remove frequencies higher than 0.1 cycles per year. Dashed lines represent periods of discontinuous station coverage

This article is organised as follows. In the next section, the scope of early literature on the early climatic fluctuation is outlined in general terms, and instrumental data sources are identified. Section 3 describes in some detail results reported by scientists working in the first half of the 20th century. The relationships between variations in atmospheric circulation and regional temperature anomaly patterns are described in Section 4, and the relevance of meridional atmospheric circulation in the Atlantic sector of the Arctic is demonstrated using newly derived circulation indexes similar in construction to the canonical station-based NAO index. Section 5 addresses low-frequency climate variations observed in the Atlantic and Pacific sectors, and in Section 6 our discussion and conclusions are presented.

2. Data and methods

Our investigation was guided by the findings of the scientists who studied the early climatic fluctuation first-hand. There were more than a thousand scientific papers related to climate change and the early climatic fluctuation written during the first half of the 20th century. Some of this work appeared in journals that ceased publication long ago. We have compiled our own database using the reference lists found in various symposia proceedings and other publications that appeared during the period as a starting point. ‘Climatic changes in the Arctic in relation to plants and animals’ (CIEM/ICES, 1948), ‘Recent Climatic Fluctuations’ (Lysgaard, 1949) and ‘Selective annotated bibliography on climatic changes’ (Brooks, 1950) are examples. Excellent reviews by Veryard (1963), Wallén (1984) and Ellsaesser et al. (1986) are also valuable guides both to the literature and its content. A subset of about one hundred of these research papers was most relevant to the subject of this study.

We have applied the insight gained from the early literature to our investigation and synthesis of the instrumental record. The longer records span about 250 years but are limited to land surface observations (SAT, SLP and the main atmospheric circulation indexes) for stations around the North Atlantic. Gridded SAT and SST data cover the period from 1850 and are also increasingly sparse further back in time. Datasets used in this study include CRUTEM3v (Brohan et al., 2006) for land SAT, and SST data from HadISST (Rayner et al., 2003) and HadSST2 (Rayner et al., 2006). Long-period station data were also obtained from the World Monthly Surface Station Climatology (1981-updated), from the Climate Research Unit (CRU) (Jones et al., 1997) and from the Annual to Decadal Variability in Climate in Europe (ADVICE) project (Jones et al., 1999). Additional historical data were digitized from original publications such as Repertorium für Meteorologie (available online from the NOAA Central Library).

Circulation indexes include the NAO (Hurrell, 1995), the Southern Oscillation Index (SOI) (Trenberth, 1984), and the North Pacific Index (NPI) (Trenberth and Hurrell, 1994). We have also used principal component (PC) indexes including the AO/NAM index, a variant of the Pacific–North American (PNA*) index and the unnamed PC 3 and PC 4 of the Northern Hemisphere SLP field (Quadrelli and Wallace, 2004; Overland and Wang, 2005), all based on the NCEP-NCAR reanalysis (Kistler et al., 2001). PC indexes were used to verify the performance of two special station-based indexes we have constructed using standardised SLP differences. These are designated meridional indexes 3 and 4 (MI 3 and MI 4) as they are primitive equivalents to PC 3 and PC 4. MI 3 and MI 4 are defined, respectively, by the standardised SLP difference between Iceland and Arkhangel'sk (Russia), and Iceland and the homogenised Edinburgh/Aberdeen record plus Bodø (Norway). Station-based indexes are particularly useful as they often cover much longer periods than is possible with PC indexes. Comparisons between the leading four PC indexes and their station-based equivalents are found in the Supporting Information. We used both the NCEP-NCAR reanalysis and the newly released 20th Century Reanalysis (Whitaker et al., 2004; Compo et al., 2006; Compo et al., in prep.) to investigate spatial patterns in SLP and SAT fields associated with these circulation indexes. Twentieth Century Reanalysis data were provided by the NOAA Earth System Research Laboratory, Physical Sciences Division (http://www.cdc.noaa.gov).

One of the inherent difficulties in working with long instrumental time series is accounting for inhomogeneities and errors in the data and for missing data. We rely heavily on datasets that have already been extensively checked for issues of this type, such as CRUTEM3v and HadSST2. However, these data sets are not guaranteed to be free of discontinuities or other problems (e.g. Thompson et al., 2008). Results derived from the new 20th Century Reanalysis have been qualitatively compared with the NCEP/NCAR Reanalysis and are presented side-by-side when indicated. When we used data from sources less well quality-controlled, we have checked for problem data and where possible we have run similar analyses with different combinations of data from nearby stations. We have also grouped data into regions with relatively stable station distributions to avoid statistical problems related to a rapidly changing observational network. Correlations reported here were computed at the 95% confidence interval. We did not encounter any remarkable discrepancies in the early research findings compared to the direct use of instrumental observations, although this is not surprising given that most of the data available today are from the same original sources. We did find, however, that early analyses often reflected an emerging understanding of atmospheric dynamics that is surprisingly consistent with current thinking (e.g. Exner, 1913). This general consistency combined with extensive cross-checking provides confidence that our conclusions based on the instrumental record are not seriously affected by data quality issues.

3. Perspectives from the early research

Scientists working in the first decades of the 20th century detected the onset of the early climatic fluctuation almost immediately and tracked its spatial and temporal development as its environmental and economic impacts became increasingly clear. Wintertime warming in Northern Europe was described around the turn of the 20th century by Ekholm (1901) and Wallén (1916), an earlier date than commonly associated with the early climatic fluctuation today. By the 1930s, positive SAT trends in the eastern United States, Northern Europe and elsewhere had been documented (e.g. Kincer, 1933; Scherhag, 1936; Callendar, 1938).

Analyses of the air temperature records that were available worldwide indicated that the early climatic fluctuation was characterised by a time-dependent patchwork of warming and cooling regions (e.g. Lysgaard, 1949; Willett, 1950; Mitchell, 1963, and as shown in Figure 1). The existence of regional differences in SAT anomaly patterns at the higher northern latitudes was also frequently noted by early writers (e.g. Veryard, 1963, and references cited therein). For this reason, spatial anomaly fields of SAT can be considerably altered by even modest differences in the time interval used in their construction (Mitchell, 1963).

The largest SAT anomalies were observed around the high latitudes of the North Atlantic, from western Greenland though northern Fennoscandia and into Russia (Jensen, 1939; Ahlmann, 1948). These anomalies appeared rapidly during the first few years of the 1920s between Iceland and Svalbard, and by the 1930s spread over the broader region. In contrast, the largest SAT anomalies observed in Alaska and western Canada occurred in the early 1940s and were associated with a major El Niño event (e.g. Trenberth and Hurrell, 1994). During these same 3 years, a record-breaking series of severe winters occurred in Europe (Liljequist, 1943). Warmer conditions continued in western Greenland and in the North Atlantic region for several more decades and have never returned to their 19th century state (Figure 2).

Positive SST anomalies were found along the northern boundary of the Gulf Stream system and in the northern Atlantic beginning in the 1920s (Bjerknes, 1959) and later along the Pacific coast of North America (e.g. Ketchen, 1956; Rodewald, 1957). Recording thermographs installed on ships operating on scheduled routes from US East Coast ports to Bermuda and other points to the South revealed large meanders and eddies along the Gulf Stream front (Church, 1932, 1937; Hachey, 1939). Russian oceanographers found the temperature of Atlantic water flowing into the northern Barents and East Greenland seas and into the Arctic Ocean had increased over the 40 years since Nansen collected the first oceanographic data in this region, and that a progressive decline in the thickness of the overlying layer of cold and relatively fresh Arctic water had occurred (Schokalsky, 1936; see also Rodewald, 1965; Hansen and Meincke, 1984). Dunbar (1954) presciently observed that if this cold layer disappeared as a consequence of another increase in Atlantic circulation in the future, the Arctic would become ‘considerably milder, and moister.’

Environmental impacts associated with the early climatic fluctuation were also observed in many areas but were most apparent around the northern North Atlantic. Climate impacts were similar in kind to recent changes documented in the Arctic (ACIA, 2005). These include: declines in glaciers and reduced sea ice extent in the Atlantic sector, an increase in the rate of the trans-polar sea ice drift, and pervasive shifts in marine and terrestrial biogeography, from echinoderms to birds (e.g. Jensen, 1939; Koch, 1945; Ahlmann, 1948; Zubov, 1948). Several recent studies have shown that much of the observed 20th century climate impact on glaciers and tundra in the Atlantic regions occurred during these years (Dowdeswell et al., 1997; Fraedrich et al., 2001; Zeeberg and Forman, 2001; Nordli et al., 2005).

It was widely recognised that a variation in the atmospheric general circulation after the turn of the 20th century was an important factor in the development of the early climatic fluctuation (e.g. Scherhag, 1937; Zubov, 1948; Bjerknes, 1963). This conclusion was founded on empirical studies of the general circulation (Walker, 1909, 1910; Exner, 1913; Defant, 1924; Wagner, 1929). The largest positive SAT anomalies around Svalbard and the northern North Atlantic were associated with an increase in southerly winds and more frequent storms in the region compared to previous decades. Veryard (1963) observed that the whole temporal and geographic distribution of SAT and SST anomalies and other climate-related environmental impacts during the early climatic fluctuation could be accounted for in terms of variations in the atmospheric general circulation.

A more fundamental cause for early climatic fluctuation was never established, although a number of familiar theories were advanced. An apt summary of the status of debate on the causes of climatic change by the mid-20th century was provided by Manley (1961), who described the outcome of a conference on this topic in Nature:

‘… supporters of each of the popular theories—solar variation, atmospheric turbidity, carbon dioxide, ozone, variations in the Earth's orbital elements—find their several gods alternately set up and cast down.’

4. Atmospheric variability and regional temperature anomaly patterns

The temporal evolution of regional SAT anomaly patterns in the Northern Hemisphere in winter is related, in part, to the time-varying superposition of the AO/NAO and PNA* circulation patterns (Quadrelli and Wallace, 2004). For the period 1953–2002, 44% of the Arctic mean SAT variance is explained by the AO and PNA* combined. This value is consistent with earlier studies (e.g. Wallace et al., 1995; Dickson et al., 2000). In the Atlantic sector of the Arctic, similarly defined linear combinations between the AO/NAO and the unnamed PC/MI 3 and 4 are regionally important (Figure 3, top). Over the period of record (1851–2006), the two meridional indexes explain 40–60% of the wintertime SAT variance in the region between Iceland and Svalbard where the NAO is practically uncorrelated.

Figure 3.

Winter (DJFM) vector wind and temperature anomaly patterns produced by regression onto a linear combination of MI 3 and MI 4 indexes using the 20th Century Reanalysis (A) and the NCEP-NCAR Reanalysis (B). The meteorology underpinning meridional patterns over the North Atlantic is illustrated using December 1938 as an example (C). SLP contours are plotted at 5 hPa intervals over the SAT anomaly field (1908–1958 reference period), based on the 20th Century Reanalysis. Rossby (1939) observed that the westward advance of the Siberian High (D) caused a complete change in weather type in northwestern Europe. In Svalbard, this was the warmest December on record (1911–2007), with rain showers reported mid-month (US Weather Bureau, 1941)

Warming in the Atlantic sector of the Arctic during the early climatic fluctuation was associated with increased southerly winds and a greater number of storms penetrating into the region than in previous years (Hesselberg and Johannessen, 1958). These changes were, in turn, associated with the appearance of anomalous high pressure over Europe (e.g. Rossby, 1939; Eythorsson, 1949). Pettersson (1949) found enhanced meridional circulation during the period 1920–1939 compared to previous decades, and observed that the Icelandic Low had deepened (implying a positive NAO), while at the same time the Siberian High tended to spread westward toward Europe (Figure 3, bottom). This description fits SAT/SLP patterns and vector winds associated with MI 3 and MI 4 indexes.

The sequence of positive values among the main circulation indexes before and during the early climatic fluctuation is uncommon in the context of the longest instrumental records available (Figure 4). A nearly unbroken positive run in the winter NAO index occurred from 1903 to 1915. The concurrent appearance of large SAT anomalies of opposite sign at stations in North Africa and Northern Scandinavia, along with negative SST anomalies in the Atlantic trade wind belt (15°–30°N), are evidence of direct climate impacts associated with the AO/NAO during this period (e.g. Thompson and Wallace, 2001). From 1920, the indexes again tend toward the positive phase. The positive run of MI 3 and MI 4 indexes extends to approximately 1950. During the early climatic fluctuation, the SOI is also correlated with SAT and SST anomaly patterns normally associated with the NAO. This relationship appears to persist into the 1950s but does not occur to the same degree before or after the early climatic fluctuation (see also Supporting Information). This suggests influence from the tropical Pacific was relatively prominent. Regressions of 20th Century Reanalysis SLP and SAT fields onto these circulation indexes for the period 1908–1957 are shown in Figure 5. The similarity between the SLP and SAT anomaly patterns associated with the NPI, SOI and NAO indexes is striking and implies an unusual consistency in the winter planetary scale atmospheric circulation over these years.

Figure 4.

Winter (DJFM) time series for the leading circulation indexes plus meridional indexes MI 3 and MI 4. Station-based versions are shown because of their greater length of record compared to PC indexes. Positive values in the NAO, MI 3 and MI 4 indexes are salient during the period around the early climatic fluctuation

Figure 5.

SLP and SAT anomaly patterns associated with five circulation indexes for the period 1908–1957 based on the 20th Century Reanalysis. Results are consistent with the same analysis performed using data from Trenberth and Paolino (1980) and CRUTEM3v (Brohan et al., 2006)

The sequence of variations in the general circulation—insofar as indicated by station-based indexes including MI 3 and MI 4—provides an explanation for much of the spatial and temporal pattern of climate impacts during the early climatic fluctuation on inter-annual to decadal time scales. The long period of + NAO circulation after the turn of the century combined with the other strong index-SST correlations from approximately 1920 to 1950 suggests anomalous atmospheric forcing of SST in the oceans may underlie low-frequency variations in SAT that are not otherwise explained. This is the subject of the next section.

5. Low-frequency climate variation in the Atlantic and Pacific sectors

Climate variations in the Atlantic and Pacific sectors of the Arctic are both associated with fluctuations in SST anomaly patterns (Figure 6). Variations in the latter region, and particularly the ∼1976–1977 climate shift, are linked to SST anomalies along the west coast of North America; part of the Pacific Decadal Oscillation (PDO) pattern (Mantua et al., 1997). Winter SAT variations in the Atlantic sector are similarly congruent with SST anomalies along the northern boundary of the Gulf Stream. We refer here to the general region between 35°–45°N and 40°–75°W (i.e. the Northwest Corner) where the Gulf Stream cold wall is found and where Gulf Stream meanders and eddies occur. Spatial patterns for the period 1908–1957 are shown in Figure 7.

Figure 6.

(Left) Winter (DJFM) land-only SAT anomaly time series in the Atlantic sector north of 60°N (as in Figure 2) compared to the SST anomaly in the 5° × 5° grid box centred at 35°N–70°W (data from CRUTEM3v and HadSST2). (Right) Pacific sector SAT anomaly compared to SST anomaly at 50°N–130°W. Correlation coefficients (r) are 0.55 and 0.60, respectively (1900–1999)

Figure 7.

Regression of the global SST field (DJFM) on (A) the Atlantic sector mean winter SAT, (B) Pacific sector SAT, (C) the NAO and (D) the NPI for the period 1908–1957 (using CRUTEM3v and HadISST). Black squares indicate the grid boxes corresponding to the SST time series in Figure 6

Large-scale influences on the atmosphere due to enhanced energy flux from warm SSTs along the Gulf Stream front were hypothesised by Minobe et al., (2008). These authors suggested deep convection through the marine atmospheric boundary layer into the upper troposphere could produce remote effects by forcing the standing circulation and thus the Atlantic storm track. High-resolution data necessary to rigorously test these relationships are lacking but it follows that anomalous SSTs observed in this region during the early climatic fluctuation may have had similar impacts. The increase in cyclone frequency on the northern margin of the Atlantic storm track documented in the literature, enhanced temperature advection into the Arctic associated with positive MI 3 and MI 4 circulation patterns, and the retreat of sea ice and other impacts on the regional cryosphere are all consistent with a systematic influence of this description. The relative stability of index-SAT correlations through major climate shifts in both the Atlantic and Pacific sectors further supports this interpretation (Figure 8). An ocean–atmosphere interaction of this nature would explain the parallel warming of Europe, Iceland and western Greenland that took place during these years, even though the normal dynamically induced North Atlantic temperature see-saw continued in effect (Loewe, 1937; Lysgaard, 1949; Loewe, 1966).

Figure 8.

Scatter plot of SAT anomaly in Iceland and Alaska regions versus the most relevant circulation index available. Regression lines correspond to periods before and after fluctuations in respective SST anomaly patterns. The consistency of the slope as the y-intercept fluctuates is an indicator of systematic forcing acting through the atmospheric circulation. This is especially clear in the Iceland example, where the shift in the 50-year mean SAT is highly significant while the MI 4 index correlation is nearly constant. The dashed line covers the period between 1963 and 1992 when Atlantic SSTs were somewhat lower

It is an unsettled question whether the existence of multi-decadal SST anomaly patterns in the North Atlantic and Pacific Oceans is primarily due to variations in thermohaline circulation (THC), other imperfectly understood factors intrinsic to the ocean, or are a response to stochastic forcing by the atmosphere (e.g. Frankignoul et al., 1997; Knight et al., 2006; Wunsch, 2007; Roe, 2009; Kwon et al., in prep.). It is probable that different combinations of factors dominate in each case. Without solid physical understanding, it is impossible to distinguish among various models for characterising this variability, including an autoregressive (AR-1) process (Overland et al., 2006). However, fluctuations in western boundary currents in general and SST along the Pacific coast of North America are ultimately linked to variations in wind forcing. The appearance of SST anomalies along the Gulf Stream front in the 1920s follows a long positive trend in the NAO culminating in the 13-year run of + NAO circulation over the North Atlantic discussed previously. In the Pacific, positive SST anomalies along the North American coast peak in the 1940s in association with a strong El Niño. This implies that both low-frequency SST patterns could have been a dynamic response to wind forcing (albeit with substantial inertia and the potential for positive feedbacks) rather than an initial cause of the early climatic fluctuation.

6. Discussion and conclusions

We found that scientists working in the first decades of the 20th century readily detected the early climatic fluctuation and developed an understanding of its physical dimensions, particularly with regard to its spatial extent, regional timing and association with shifts in the atmospheric circulation. Many of the environmental impacts seen today were also reported then. And like today, the social and economic effects of the event caused concern among scientists and the general public.

The largest climate and environmental impacts of the early climatic fluctuation were centred in the high latitudes of the North Atlantic. SAT and SST anomalies that appeared in the Pacific in the 1940s contribute to the Arctic mean temperature but were forced quasi-independently by coupled variations associated with El Niño-Southern Oscillation (ENSO), the PDO, and the Aleutian Low (e.g. NPI). The complexity of SAT anomaly patterns observed during the early climatic fluctuation—and a portion of the multi-decadal signal in the Arctic mean temperature record—is thus related to the time-varying superposition of major circulation patterns. Low-frequency SAT variations in the Arctic are correlated in about equal measure with variations in SST along the Gulf Stream front and along the Pacific coast of North America.

The cause of multi-decadal SST anomaly patterns and potentially related ocean–atmosphere interactions is the subject of ongoing research. As yet none of the proposed models, including quasi-periodic oscillations in THC, regime-like shifts, or a red noise (AR-1) process, can be decisively eliminated. While not devoid of interest, the pursuit of statistically significant cycles or periods in the climate—which C.E.P. Brooks likened to the search for the philosopher's stone (Gregory, 1930)—has historically been unproductive. However, after the turn of the 20th century and during the early climatic fluctuation, there was persistent anomalous wind forcing over the North Atlantic. The role of wind forcing in the creation of low-frequency SST signals in western boundary currents such as the Gulf Stream and elsewhere is supported both by a theoretical framework and by observations. This lends weight to the hypothesis that the early climatic fluctuation was a singular event resulting from intrinsic variability in the large-scale atmosphere/ocean/land system and that it was likely initiated by atmospheric forcing.

There are two interconnected explanations for why the largest climate impacts appeared in the northern North Atlantic where the AO/NAO is not strongly correlated with SAT. The proximate cause is the rise of meridional circulation patterns (MI 3 and MI 4) in the 1920s and 1930s. Observations reported in the early literature and our analysis of the instrumental record are perfectly consistent on this point. However, persistent SST anomalies that appeared during this period provide a plausible mechanism for systematic decadal forcing of the standing circulation. There is insufficient data to verify this point but observations are consistent with an influence of this description. It explains one of the most puzzling features of the early climatic fluctuation: the parallel warming of Greenland and Northern Europe that occurred even while the prominent dynamically induced North Atlantic temperature see-saw continued unabated.

The early climatic fluctuation comprises an important part of the overall record of climatic change during the 20th century. In addition to the large SAT and SST fluctuations discussed earlier, there is evidence that the magnitude of the impacts on glaciers and tundra landscapes around the North Atlantic was larger during this period than at any other time in the historical period. Given that intrinsic forcing played a major role in early climatic fluctuation then the possibility exists that the warming of the Arctic that is presently underway may slow in the future, especially if another period of predominantly negative AO/NAO and PNA* circulation patterns were to occur. This scenario is not unreasonable given that a substantial part of the 1950–2000 Arctic temperature trend corresponds to the run of positive AO/NAO and PNA* patterns in recent decades following a run of negative ones in the 1950s and 1960s (Thompson et al., 2000; Quadrelli and Wallace, 2004). Alternatively, if the combination of factors that contributed to the early climatic fluctuation were to recur, as evidenced by the rise of MI 3 and MI 4 circulation patterns, for example, then the warming of the Arctic that is presently underway may exceed current IPCC model projections, further reinforced by the rapid loss of sea ice. There is no reason a priori that the first alternative will not occur. However, in the context of rising anthropogenic forcing, the consequences of the second alternative are more worrisome.

The ultimate cause of the early climatic fluctuation was not discovered by early authors and remains an open question. All of the leading possibilities recognised today were raised by the 1950s, including internal atmospheric variability, anthropogenic greenhouse gas (CO2) forcing, solar variability, volcanism, and regional dynamic feedbacks (e.g. Manley, 1961). Greenhouse gas forcing is not now considered to have played a major role (Hegerl et al., 2007). The analyses presented here favour an interpretation of the early climatic fluctuation as an intrinsically forced phenomenon, and therefore it is best represented as a large but essentially random climate excursion imposed on top of the steadily rising global mean temperature associated with anthropogenic forcing.

Acknowledgements

We appreciate discussions with J. M. Wallace, G. Roe and N. Bond at the University of Washington. We gratefully acknowledge the support of the NOAA Arctic Research Program. This publication was funded, in part, by the Joint Institute for the Study of the Atmosphere and Ocean (JISAO) under NOAA Cooperative Agreement No. NA17RJ1232, Contribution 1432. This paper is a contribution to the Arctic System Science Program under NSF grant 0531286. PMEL Contribution 3111.

Ancillary