The reduction in ice cover observed in the late 1980s and early 1990s has been attributed to the strongly positive Arctic Oscillation (AO) phase during that time. However, despite a change in the AO to more neutral conditions since then, ice extent and the fraction of old ice have continued to decrease. This mismatch between the AO index and loss of ice can be explained by the frequency of three main sea level pressure (SLP) patterns that yield overall variability in SLP, rather than the presence of a single, coherent physical pattern of SLP reduction associated with the positive mode of the AO. These three patterns were in phase during the peak AO period but their frequency has varied differently since then, with two of the patterns continuing to contribute to reduced ice cover in the western Arctic. Hence, regional atmospheric circulation remains a significant factor in recent reductions in ice cover.
1. Central Arctic Circulation and the Arctic Climate Paradox
 The co-occurrence of strongly positive Arctic Oscillation (AO) [Thompson and Wallace, 1998] index values and large changes in ice conditions in the late 1980s through early 1990s contributed to the conclusion that the AO was the most dominant atmospheric circulation regime affecting Arctic climate. Since the mid-1990s however, the AO index has been near neutral on average, yet reductions in ice extent and other changes previously associated with the AO continue to occur [e.g., Stroeve et al., 2005]. This Arctic climate paradox [Overland and Wang, 2005a] has led to suggestions that the Arctic has passed a tipping point into a new climatic state and that high-latitude atmospheric circulation may no longer be the key driver of sea ice trends [e.g., Lindsay and Zhang, 2005].
 But does the recent AO decrease necessarily mean that atmospheric circulation is no longer a dominant factor, or are there other, more regional, SLP patterns that contribute significantly to interannual variability [e.g., Polyakov et al., 2003]? To test this, we use ice motion data [Fowler et al., 2003] to calculate an index summarizing ice transport direction within an area north of Greenland and west to the Beaufort Sea (Figure 1), and compare this index to atmospheric circulation patterns.
 For this index, the fraction of vectors with west-to-east and south-to-north drift components defines the general direction of motion. Cases (days) when more than half of the domain's vectors have a west-to-east (i.e., cyclonic) and south-to-north drift direction are defined as cases of “positive” ice drift, with these patterns favoring reductions in ice cover as described below. This “west Arctic transport” (WAT) index is comparable to Proshutinsky and Johnson's  cyclonic and anticyclonic surface motion index, but also captures northward drift that is not strictly cyclonic, and is most representative of conditions in the Beaufort and Chukchi seas where ice reductions have been most severe in recent years. Lagrangian drift tracks derived from the daily motion fields were used to verify cumulative ice transport patterns.
 Mean drift calculated from days with positive and negative WAT values, and the corresponding mean sea level pressures for those months, obtained from NCEP/NCAR Reanalyses (hereafter reanalyses) [Kalnay et al., 1996], are depicted in Figure 1. The large difference in the strength and location of high and low SLP within the Arctic Basin for positive vs. negative index conditions is apparent. The positive WAT index corresponds to reduced ice cover in the Beaufort, Chukchi and Siberian seas [cf. Drobot and Maslanik, 2003], with a correlation coefficient of −0.53 (significant at p = 0.01) between the WAT index and annual ice concentration anomalies derived from passive microwave data [Cavalieri et al., 1996] for the western Arctic region indicated in Figure 1. The correlation increases to −0.61 when the WAT index leads the atmosphere by one year (significant at p = 0.01). The known relationships between atmospheric conditions, ice transport and ice extent and concentration [e.g., Proshutinsky and Johnson, 1997] suggest that this correlation is physical as well as statistical, although the relative importance of dynamical versus thermodynamical factors is uncertain.
 Comparison of the WAT and AO indices (Figures 2a and 2b) suggests that cyclonic conditions and south-to-north winds and ice transport have persisted since the late 1980s, despite the decrease in the AO index. AO and WAT values show little agreement on a monthly basis (r = 0.27), which is confirmed by examining individual monthly-mean SLP fields. The maximum correlation of the AO with the mean ice concentration anomalies for the region indicated in Figure 1 is −0.30 at 0 lag.
2. Multiple Atmospheric Circulation Patterns and Their Relationships to West Arctic Ice Transport
 Previous work suggests that the AO index integrates multiple atmospheric circulation patterns [e.g., Ambaum et al., 2001; Wallace and Thompson, 2002]. To investigate the presence and frequency of different atmospheric circulation regimes, self-organizing maps (SOMs) were generated from daily SLP for 1948–2004. This neural net-based unsupervised classification technique, which offers some advantages over principal components techniques for recognition of physical patterns [Hewitson and Crane, 2002; Reusch et al., 2005], is based on a 2-dimensional array of nodes, where each node is trained to identify spatial patterns of SLP. Each daily SLP field is then assigned to the best-matching SOM (here we used an 11 × 9 matrix for 99 potential patterns) [e.g., Cassano et al., 2006; Barry and Carleton, 2001]. Mean ice motion fields were generated for each SOM by averaging the motion vectors for all days mapped to that SOM. The result is a mean SLP field, mean ice motion field, and WAT index value derived from this motion field for each of the 99 SOMs.
 Based on the WAT index values for each SOM, three types of patterns appear most significant in terms of Arctic Basin winds and ice transport (Figure 3). The “light ice” phases of these patterns include decreased SLP in the North Atlantic (an “NAO-like” pattern resembling the positive phase of the NAO), a low pressure cell within the Arctic Basin (a “central Arctic” pattern), and a dipole pattern of high pressure over the Canadian Arctic paired with low pressure over the Siberian Arctic.
 The NAO-like pattern (Figure 3a), as was seen during the peak AO period in 1989–1993, transports ice mass from the Siberian Arctic toward Canada [e.g., Proshutinsky and Johnson, 1997; Rigor et al., 2002], with a general divergence of the ice pack within the central Arctic during cyclonic motion, particularly during summer when conditions approach free drift, with ice moving to the right of the wind direction. In contrast, the dipole pattern (Figure 3c) shifts mass between the western (Pacific) and eastern (Atlantic) sides of the Arctic Basin. Examination of the full set of 99 SOMs also illustrates that large changes in ice transport can occur due to relatively slight displacements in the positions of closed high and low pressure areas. The strength and position of these centers of action are affected in turn by factors such as cyclone frequency and storm tracks that vary considerably from year to year [Serreze and Barry, 1988] and that are influenced by local as well as larger-scale climatic conditions.
 Similar SLP patterns were first described by Gudkovich  and have since been noted in research [e.g., Rogers and McHugh, 2002; Wu et al., 2006] using approaches that differ somewhat from the original AO identification. Overland and Wang [2005b] describe the relationship between the dipole pattern discussed above and spring temperature anomalies, and emphasize that multiple climate states exist in the Arctic. In fact, Wu et al.  conclude that the dipole pattern is more important than the AO for driving ice transport within the Arctic Basin as a whole; a finding supported by the results presented here. The central Arctic pattern is essentially what Rogers and McHugh  and Wang et al.  call an “AO-like” pattern. It is important to recognize though that the AO index does not necessarily capture the presence of this central Arctic pattern, and the two are not interchangeable in terms of describing SLP within the western and central Arctic.
 Consistent with research pointing out the relationships between the AO and the NAO and a lack of statistical correlation with the north Pacific [e.g., Deser, 2000; Ambaum et al., 2001], the AO index correlates most strongly with the frequency of occurrence of SOMs that depict below-normal SLP in the north Atlantic, but the correlation is divided among two types of SOMs – one type with low pressure in the north Pacific (the largest positive correlation; Figure 3a) and a second with high pressure in the same area (not shown). The AO is negatively correlated with the positive dipole pattern (Figure 3c).
 Using the time series of SOMs for each day from 1948–2004 and the mean ice drift fields for each of these SOMs, a proxy of daily ice drift was reconstructed for the full time period. WAT index values were then calculated from the proxy motion fields for each day (Figure 4). The resulting WAT time series indicates that northward and west-to-east ice transport is likely to have been mostly above average since 1988 relative to the 1948–2004 record. The increase has been greatest in spring, particularly during 2002–2003 (not shown). In contrast, winds and ice drift were consistently more anticyclonic and/or northerly during 1978–1987. The ice cover may therefore have been unusually heavy prior to the observed large losses of old ice in 1989 and subsequent decreases in ice extent. If so, this needs to be kept in mind when interpreting the satellite data record since the large decreases during the early 1990s could in part represent an adjustment from a period of abnormally heavy ice.
 The frequency of occurrence (Figure 5) of SOMs corresponding to the three circulation patterns discussed above and shown in Figure 3 suggests that during the peak positive AO years of 1989–1993, these three general patterns were all relatively frequent. The correlations of the summed frequency (Figure 5d) of these SLP patterns with the WAT index and the observed ice concentration anomalies for 1979–2004 are 0.68 and −0.69 respectively at 0 lag (both significant at p = 0.01), supporting the conclusion that these patterns are strongly linked to observed changes in ice cover.
 From the mid 1990s through 2004, positive NAO conditions were less common, and the frequency of the central Arctic low pressure and positive dipole patterns initially decreased. However, the latter two patterns then increased from 2000–2004, consistent with cyclonic motion during that period as determined by Proshutinsky . Frequency of the dipole pattern was greater for 2000–2003 than any previous period (Figure 5c). This latter situation fosters northward transport of relatively warm Bering Sea water through the Bering Strait [e.g., Woodgate et al., 2006], and helps explain large climatic changes in the northern Bering Sea area that Grebmeier et al.  showed to be associated with regional distribution of high and low atmospheric pressure.
 The apparent lack of correlation of the AO with the continued warming and loss of Arctic sea ice has led to speculation that atmospheric circulation is no longer playing a dominant role in affecting Arctic conditions. However, we find that winds and ice transport patterns that favor reduced ice cover in the western and central Arctic have in fact continued to be present since the late 1980s, but the AO index is not a reliable indicator of these patterns. Polyakova et al.  reach similar conclusions regarding use of the NAO index to describe conditions in the North Atlantic. This lessens the need for explanations such as lag effects from the peak positive AO years to help account for observed ice conditions during recent years with reduced AO index, and points to the importance of understanding factors that affect regional atmospheric circulation and multidecadal variability.
 This work is supported by NASA grant NNG04GP50G and NSF grant OPP-0100120. Data were provided by NOAA and the National Snow and Ice Data Center.