Analysis of a satellite-derived record of sea ice age for 1980 through March 2011 shows continued net decrease in multiyear ice coverage in the Arctic Ocean, with particularly extensive loss of the oldest ice types. The fraction of total ice extent made up of multiyear sea ice in March decreased from about 75% in the mid 1980s to 45% in 2011, while the proportion of the oldest ice declined from 50% of the multiyear ice pack to 10%. These losses in the oldest ice now extend into the central Arctic Ocean and adjacent to the Canadian Archipelago; areas where the ice cover was relatively stable prior to 2007 and where long-term survival of sea ice through summer is considered to be most likely. Following record-minimum multiyear ice coverage in summer 2008, the total multiyear ice extent has increased to amounts consistent with the negative trend from 2001–2006, with an increasing proportion of older ice types. This implies some ability for the ice pack to recover from extreme conditions. This recovery has been weakest in the Beaufort Sea and Canada Basin though, with multiyear ice coverage decreasing by 83% from 2002 to 2009 in the Canada Basin, and with more multiyear ice extent now lost in the Pacific sector than elsewhere in the Arctic Ocean.
 The large decreases in summer sea-ice extent in recent years [e.g., Comiso et al., 2008; Stroeve et al., 2008] not only affect total ice coverage but also introduce a fundamental change in the nature of the Arctic ice itself - the change from a largely perennial ice cover where ice persists from year to year to a seasonal coverage with open water during summer [e.g., Kwok, 2007; Kwok and Cunningham, 2010], and the introduction of younger ice within the areas that survive summer melt [Fowler et al., 2003; Rigor and Wallace, 2004; Maslanik et al., 2007]. Here, we consider changes in the distribution of ice of different ages within the multiyear ice pack (i.e., ice that has survived at least one summer melt cycle) and focus on four particular aspects of the series: decadal-scale change, increases in coverage and aging of multiyear ice since the record multiyear ice minimum in 2008, regional variability, and the changing role of the Pacific sector of the Arctic Ocean. These changes in ice age serve as a proxy for ice thickness [e.g., Maslanik et al., 2007].
2. Sea Ice Age Data and Methods
 The data used here are a reprocessed and extended version of the 1980–2007 series of sea ice age introduced by Fowler et al.  and described further by Maslanik et al. , Tschudi et al. , and Stroeve et al. . In brief, using satellite data and drifting buoys, it is possible to observe the formation, movement, persistence, and disappearance of sea ice. This history can then be used to estimate age [Rigor and Wallace, 2004]. In the Fowler et al.  approach, ice movement is calculated using gridded satellite-derived ice motion vector fields and International Arctic Buoy Program buoy position data for 1979 onward. Ice age is estimated by treating each grid cell that contains ice as a discrete, independent Lagrangian parcel, and then transporting the parcels at weekly time steps. In cases where particles of different ages fall within a single grid cell, the cell's age is assigned that of the oldest particle. If ice concentration at the corresponding grid cell, as estimated using the NASA Team Algorithm [Cavalieri et al., 1984] applied to SMMR and SSM/I passive microwave data, remains at or above 15% throughout the melt season, then that particle is assumed to have survived the summer, and the particle's age is incremented by one year. For example, first-year ice is ice that has yet to survive a melt period, while fifth-year ice is ice that has survived four melt cycles. This also has the effect of “re-setting” the multiyear ice extent each summer, thereby minimizing any cumulative errors in multiyear ice extent over the time series. Because the grid cell is assigned the age of the oldest ice present, a cell with a total concentration as low as 15% at the end of the melt period is coded as multiyear ice even though the majority of ice present after freeze-up is first-year ice. The effects of higher thresholds have been tested, and while a higher value is useful when the goal is to study areas where older ice is dominant [e.g., Stroeve et al., 2011], the 15% value was used here as it captures greater detail within the marginal ice zone and is the more conservative approach in terms of assessing net loss of areas where some multiyear ice is present.
 Since each grid cell is considered to be a single, discrete age category and is not assigned an ice concentration value, the maps are best considered as extent maps that indicate areas where ice of different ages exists. For the results presented here, “extent” is defined as the sum of all grid cells that contain ice of the specified age; it is not the actual areal coverage of ice of particular ages. The age estimates are restricted to open ocean areas only, where ice motion can be resolved in the microwave data (see the inset map in Figure 1). As a result, some areas with ice cover, such as the passages in the Canadian Archipelago, are excluded from the analyses and totals.
3. Decadal Variability in Sea Ice Age
Figures 1–4 illustrate the basic changes in the distributions of ice extent of different ages, comparing conditions in early spring (third week of March; the most recent data available through 2011), late spring (mid May, through 2010), and at the end of summer melt. Multiyear ice cover overall has declined over time and become younger, with the oldest ice types confined to a relatively small portion of the Arctic Basin. In terms of total coverage over the entire study area (regions 1–9; Figure 1), multiyear ice extent during March has declined by 33% (from 4.57 × 106 km2 to 3.04 × 106 km2) between 1980 and 2011, while extent at the September minimum has declined by 50%. As a result, the fraction of the total ice extent (first-year plus multiyear ice) in the Arctic Ocean (regions 1–7) made up of multiyear ice in March decreased from about 75% in the mid 1980s to 45% in 2011. The fraction of this remaining multiyear ice extent that consists of areas containing particularly old ice (fifth-year and older, abbreviated as “5+” ice) declined from 50% to 10%.
 The effects of the record minimum in total ice extent in summer 2007 [e.g., Stroeve et al., 2008] are clear in the ice age time series (Figures 2 and 3), with nearly twice as much multiyear ice lost through the summer (0.94 × 106 km2) as was typical over the full time period. The loss over summer 2008 was smaller (0.67 × 106 km2), but was still the second greatest decrease observed and resulted in a record minimum of 1.5 × 106 km2 of multiyear ice, or 37% of the 1980–2006 mean of 3.9 × 106 km2. Over the period from 2007–2010, the mean survival rate of multiyear ice (defined as the difference between extent in mid May and at the September minimum) calculated over regions 1–9 declined to 74% versus the mean for 1980–2006 of 90%.
 The maps and time series suggest nominal periods of change superimposed on the net downward trend, particularly for the oldest ice types. Applying the piecewise linear fitting and break-point calculation method of [Tome and Miranda, 2004] to the March extents for multiyear and 5+ ice and specifying a minimum 5 years per period yields four time periods for each series (Table 1 and Figure S1 of the auxiliary material). As done by Tome and Miranda, we examined the robustness of the breakpoint locations by testing the dependency on the specified minimum duration of the years between breakpoints. The breakpoints vary by one or two years depending on the interval specified, but are consistent in centering around those given in Table 1. The general pattern is one of increasing multiyear ice cover through the mid 1980s, followed by loss through the early 1990s, stable coverage through the early 2000s, and then another period of accelerated decline beginning in 2002 for multiyear ice and 2004 for 5+ ice. Annual mean (October to September) sea level pressure (SLP) anomalies from NCEP/NCAR reanalysis fields (Figure S1) for periods 1 and 2 of the 5+ ice series show different phases (negative and positive, respectively) of a somewhat zonal, Arctic Oscillation (AO)-like pattern, representing the first leading mode of EOF SLP variability [Rigor and Wallace, 2004; Stroeve et al., 2011]. During the relatively stable period from 1994–2003, SLP anomalies show considerable interannual variability but with a relatively weak positive Arctic Dipole (AD) pattern on average (the second mode of EOF SLP variability) consisting of below-normal pressure over Canada and above-normal pressure over Siberia, as defined by Overland and Wang , or alternatively, the negative Dipole Anomaly (DA) pattern [Wang et al., 2009]. Period 4 corresponds to a strengthening in the negative AD phase, leading to accelerated loss of multiyear and 5+ ice, through large-scale changes in sea ice transport in the Arctic Basin, such as a reduced Beaufort Gyre and enhanced ice export out of Fram Strait.
Table 1. Individual Periods of Change Within the March Time Series for Multiyear Ice and 5+ Ice (see Figure 3)
Years (multiyear ice)
Multiyear ice trend (106 km2/yr.)
Years (5+ ice)
5+ ice trend (106 km2/yr.)
4. Changes Since 2008
 Since the 2008 minimum, total multiyear ice extent has increased back to values in line with the amount that would have been expected if the general downward trend over the last decade had not been interrupted by the extreme losses in 2007 and 2008. For example, based on a linear least-squares fit to data for 2002–2006 for March (included in Figure 3), multiyear ice extent in 2011 was 14% greater than would have been predicted from the 2002–2006 trend. In addition to increasing in overall extent since 2008, there was more 3rd.-year ice in March 2011 than in any prior year except 1998 (shown earlier in Figure 2). The total extent of 2nd. through 4th.-year ice, at 2.73 × 106 km2 in 2011, was 8% greater than the 1983–2011 average and greater than the amount present in 22 of the 28 years. Nevertheless, the total multiyear extent in March 2011 was still less than in any year prior to 2008, and 39% below the 1983–1986 mean, largely influenced by continued decline in the amount of 5+ ice.
 One likely reason for the recent increase in multiyear ice extent was a shift toward a more “ice favorable” atmospheric circulation. An increase in high pressure over Siberia and extending into the Barents Sea together with a weakening of high pressure over the Canada Basin yielded reduced northward ice transport in regions 1 and 6 during the ice year periods (Oct.–Sep.) 2009–2010 compared to 2007–2008. This pattern contrasts with the strong high-pressure area seen over the Canada Basin in 2007 [e.g., Maslanik et al., 2007] and the negative AD pattern in 2008 [Wang et al., 2009; Overland and Wang, 2010], and is somewhat akin to the positive AD pattern but rotated within the Arctic by about 45 degrees (e.g., with the low and high pressure regions centered over the Pacific and Atlantic sectors, respectively). Strong and sustained negative AO periods during winter [Stroeve et al., 2011] not seen since the late 1960s were present in 2010, with an annual mean AO index of −1.04. This was the lowest annual mean from 1950 through 2010, over two standard deviations from the mean of −0.14, and one of only three years with a negative AO in 11 of the 12 months. The year 2010 was also unusual as the only year over the same period with a negative North Atlantic Oscillation (NAO) in all months, with an annual mean NAO index of −1.30; the lowest observed over the 61-year period and more than three standard deviations less than the long-term mean of −0.10. As a result, while mean ice transport was cyclonic in 2009 (see http://www.arctic.noaa.gov/reportcard), transport was more anticyclonic in 2010. During the first quarter of 2011, however, SLP patterns reverted to a positive AO-like pattern, which if it persists would lead to enhanced ice loss in summer 2011 [e.g., Rigor et al., 2002] and particularly in the central Arctic Basin where losses were large during the late 1980s-early 1990s (Figure 1) and were the multiyear pack has partially recovered since 2008.
5. Regional Variability and Ice Loss in the Pacific Sector
 The regional changes seen in Figure 1 are summarized in Figures 4 and S2. Of particular note are inroads of ice loss into the most northern portions of the Arctic Basin. Within the central Arctic (Canada and Nansen basins; regions 1 and 6), the multiyear pack became considerably younger following the losses in the oldest ice types from the late 1980s through mid-1990s. The proportion of 5+ ice decreased to about 37% of the multiyear pack in 1994 from a mean during 1983–1987 of 65%. In the Nansen Basin the oldest ice types have essentially disappeared, decreasing from about 35% of the multiyear extent to less than 1% in December 2010. Through 2003, multiyear extent in the Canada Basin had remained relatively unchanged, but in summer 2004 the extent decreased below that observed during any previous period, with further decreases in the multiyear coverage during each consecutive year, reaching a minimum in 2009 and decreasing by 83% from 2002. The loss of the oldest ice was even more extreme, with 5+ ice reaching a minimum in 2010 of just 6% of the 1983–2002 mean, with a decrease of 92% since 2002.
 While the area directly north of the Canadian Archipelago (Region 7) has remained relatively stable in terms of multiyear ice extent, the fraction of the oldest ice types decreased considerably beginning in 2006. In addition to less of the oldest ice available to be transported into the region, this change is due to the continued flow into the Beaufort Sea (Region 2), including above normal transport observed during winter 2009/2010 [e.g., Stroeve et al., 2011], likely combined with increased export through the Nares Strait in recent years [Kwok et al., 2010]. Some multiyear ice therefore continues to be present in the Beaufort region during winter and spring, including some of the oldest ice. However, in recent years, relatively little of this ice has survived through summer [Kwok and Cunningham, 2010]. The mean survival rate of multiyear ice extent in the Beaufort Sea as a whole declined from 93% over 1981–2005 to 73% during 2006–2010, with most of the surviving ice located in the eastern portion of the region (e.g., Figure 1). Regions 1 and 2 now account for the majority of summertime reductions in multiyear ice extent, exceeding the losses elsewhere in the Arctic Ocean (Figure 5). Whereas the mean loss in this area prior to 2004 was 0.11 × 106 km2 per year, this increased to 0.34 × 106 km2 for 2004–2010. During 2005–2008, these two regions accounted for 67% of the total ice lost within regions 1–8. Kwok and Cunningham  show similar results for a domain comparable to the area covered by our regions 1 and 2 but using different data and ice concentration thresholds; finding that the area contributed 32% of the net loss for 2005–2008, versus our estimate of 42%. While convergence and deformation or transport to adjacent regions between mid May and the September minimum contribute to some of the loss in extent, the short time period and the directions of transport in the regions suggest that nearly all of the loss is due to melt. This changing significance of the Pacific sector is particularly apparent for the oldest ice types. Most of the reduction in coverage of the 5+ ice over summer now occurs within the Arctic Basin itself, with relatively little loss through the Fram Strait.
 The 32-year record of estimated sea ice age points to a considerably younger ice pack than existed as recently as a decade ago, highlighting a net long-term decrease in multiyear ice extent and the extent of the oldest ice types. Since older multiyear ice is typically thicker than younger ice [e.g., Maslanik et al., 2007], this translates into an overall thinner ice cover. The recovery in multiyear ice extent through March 2011 from the extreme reductions in 2007 and 2008 along with the continued aging of the surviving ice through multiple melt seasons is consistent with an ice pack that has not passed a tipping point across the Arctic Ocean as a whole [Amstrup et al., 2010], and reflects favorable large-scale ice transport patterns conducive to retaining multiyear ice.
 On the other hand, a basic shift has occurred in recent years, with the Beaufort Sea and Canada Basin now accounting for the largest portion of the multiyear ice extent lost within the Arctic Basin. A variety of factors may be impeding recovery in this area, including effects of ocean heat transport [Polyakov et al., 2010] and increased solar heat absorption in open water areas [Perovich et al., 2011; Steele et al., 2010; Jackson et al., 2010]. Accelerated ice motion in the Beaufort Sea, which we find to be associated with the negative Arctic Dipole, may also help foster ice melt [Shimada et al., 2006]. It could be that these processes, taken together, represent evidence of a “regional tipping point”. A key question then is whether the increases in the extent and age of the multiyear ice seen over the last three years can be sustained - continuing the recovery observed since the 2007/2008 record minima and thickening enough to become less vulnerable to another extreme melt season [e.g., Lindsay et al., 2009].
 This work was supported by the NASA Cryospheric Sciences grant NNX07AR21G and NASA Measures grant NNX08AP34A. We thank the NSIDC for microwave products, the IABP for buoy data, NOAA ESRL and the CDC for reanalysis access and plotting and provision of AO and NAO indices, and I. Rigor at the University of Washington for assistance with the IOBP products.
 The Editor thanks three anonymous reviewers for their assistance in evaluating this paper.