Changes in the extent of seasonal ice were investigated using historical and satellite observations for the period 1870–2003. The seasonal ice zone (SIZ) has been gradually expanding since 1870, with a marked acceleration over the past three decades, and has migrated north to encompass all peripheral Arctic seas. The expansion of the SIZ may be increasing the salinity of the upper Arctic Ocean, consistent with recent observations. The migration of the SIZ over continental shelves may also be enhancing the formation rate and salinity of Arctic deep waters, which are subsequently advected to the convective region of the Greenland-Iceland-Norwegian Sea, thereby influencing the formation of North Atlantic deep waters and related global thermohaline circulation.
 Various studies have reported on the declining Arctic sea ice cover using the recent microwave satellite record available since 1972 [e.g., Cavalieri et al., 2003], historical observations [Kinnard et al., 2006] or a combination of both [Walsh and Chapman, 2001]. A consistent observation from these studies is the faster rate of ice extent decline during summer than winter, with recently reported trends of −2.8% and −8.6% per decade for March and September, respectively [Serreze et al., 2007].
 Sea ice plays a key role in the surface radiation balance of the Arctic Ocean. A reduced ice cover decreases the surface albedo and allows the ocean to absorb more solar radiation, promoting further ice melt and surface warming. This positive ice-albedo feedback may amplify global warming, and it has been an important incentive for studying recent sea-ice changes.
 One aspect which has received less attention is the potential change in the extent and configuration of the seasonal ice zone (SIZ) associated with recent ice cover decline. In this paper we use the term “seasonal ice” to designate any ice that melts after the timing of maximum winter extent. This is mainly first-year ice but can include some multiyear ice.
 The freeze-thaw cycle which operates within the SIZ plays an important role in the stratification and convection of Arctic Ocean waters [Aagaard and Carmack, 1989]. Moreover, exchanges of water masses between the Arctic and the North Atlantic oceans strongly affect the strength of the global thermohaline circulation (THC), which acts as an important climate regulator [Holland et al., 2001]. Hence future changes in the characteristics of the SIZ have the potential to impact global climate.
 Our objectives here are to quantify the changes in the extent of seasonal ice that have occurred over the period of sea-ice observations 1870–2003, and to discuss the potential consequences of these changes on the Arctic Ocean circulation.
2. Data and Methods
 We use the historical grids of Northern Hemisphere (NH) sea ice cover from the University of Illinois for the period 1870–2003 [Walsh and Chapman, 2001; hereafter termed WC dataset]. The grids provide mid-month values of ice concentration (percent ice cover per grid square) on a standard 1 × 1 degree cylindrical projection. The dataset was compiled from various sources, which vary in quality and availability over time [Walsh, 1978]. Reliable ice concentrations are only available from historical sources after 1953, and from satellite imagery since 1972. Prior to 1953, only the ice edge position is reliable. For this reason we used grids of ice extent (i.e. ocean area within the ice edge) to derive the SIZ extent since 1870, but we also calculated the total seasonal ice area (sum of concentrations) from satellite-derived time series after 1978 [Cavalieri et al., 2003]. The southern Sea of Okhotsk, the Gulf of St-Lawrence and the Baltic Sea were excluded from the analysis as data coverage in these regions is inconsistent in the WC dataset.
 We created mid-month ice extent grids for the period 1870–2003 by attributing a value of one (presence of ice) to gridpoints with at least 15% ice concentration and a value of zero (no ice) elsewhere. The threshold of 15% is in accordance with previous studies of ice extent [e.g., Serreze et al., 2007]. Grids of maximum annual ice extent were created by assigning, for any given cell, a value of one if ice was present at any time during the year, and a value of zero otherwise. Conversely, in minimum extent grids a value of zero was assigned to cells where open water was present at any time during the year. Our method differs from the common practice of evaluating minimum and maximum ice extent at fixed dates (typically in September and March). Following Comiso , the minimum ice extent is taken as equivalent to the perennial, predominantly multi-year ice extent. The extent of the SIZ in any given year was then calculated as the difference between the maximum and minimum ice extent grids. Calculated in this manner, the extent of the SIZ approximates the area where new (first year) ice may be produced [e.g., Zhang and Walsh, 2006]. This does not take into account ice advection, but this effect may be considered negligible if we assume that during the freezing period, any area from which ice is exported refreezes. Our method may however underestimate the extent of new ice formed over coastal polynyas where continued ice export and refreezing occurs.
 Time series of total minimum, maximum and seasonal ice extent were calculated by summing the respective area-weighted gridded fields. In order to examine the spatio-temporal variability of seasonal ice, we calculated anomaly fields of the probability of occurrence of seasonal ice over discrete 20-year time windows. The anomalies are then defined as departures between the 20-year and the 1870–2003 averages.
3. Trends in Seasonal Ice Cover
 Figure 1 shows the spatial probability of occurrence of the ice edge over the 1870–2003 period. Regions with pronounced meridional contrast in probability (wide color spectrum), such as the Greenland Sea in winter and the Kara Sea in summer, show large temporal changes in ice extent. Time series of minimum and maximum ice extent are shown in Figure 2. The maximum extent was relatively stable until the early 1960s, after which a gradual decline is observed. The minimum extent is more variable on inter-annual to decadal timescales. A declining trend, more pronounced than that of the maximum extent, is apparent after the early 1950s, with a rate increase over the last decade or so.
 The total extent of the SIZ is shown in Figure 3. Large variability is evident on inter-annual to decadal scales. Overall, the extent of the SIZ has increased over the 1870–2003 period, with the greatest increase occurring after the 1950s, in agreement with the trend in minimum ice extent. Also noteworthy are periods of relative increases in the SIZ extent, in the early 1900s and 1930s. Figure 3 also shows the total northern hemisphere SIZ extent and area since 1979, calculated from the NASA team SMMR-SSM/I satellite-derived ice concentrations. Linear trends were computed by ordinary least-square regression with 95% confidence intervals adjusted for autocorrelation in the residuals [Weatherhead et al., 1998], and converted to percent change relative to the 1979–2003 average. Decadal trends for the SIZ extent computed over the 1979–2003 period were 2.14 ± 1.93 × 105 km2 (or 2.6 ± 2.3%) using the WC dataset, and 2.22 ± 2.15 × 105 km2 (or 2.6 ± 2.5%) using the NASA team dataset. Thus the omission of unreliable regions in the WC dataset did not significantly affect the estimated trends, at least over the recent satellite era. When ice concentrations are taken into account, the decadal trend in the SIZ area is 3.42 ± 2.44 × 105 km2 (or 4.1 ± 2.9%), thus more pronounced than for the SIZ extent. This means that a non-negligible amount of seasonal ice forms within the ice edge. Over longer time periods, the SIZ extent exhibits decadal trends of 1.03 ± 0.91 × 105 km2 (or 1.2 ± 1.1%) for 1953–2003, and 0.90 ± 0.26 × 105 km2 (or 1.1 ± 0.3%) for 1870–2003. These results suggest that the increase in the total extent of the SIZ has accelerated during the past 130 years.
 In a recent study by Zhang and Walsh , modeled sea ice projections from the Intergovernmental Panel on Climate Change Fourth Assessment Report (IPCC AR4) were synthesized for various global warming scenarios. The prescribed scenarios, taken from the IPCC Special Report on Emission Scenarios (SRES), included: (1) unconstrained emissions leading to an increase in atmospheric CO2 concentrations to over 800 ppm by the year 2100 (SRES-A2); (2) constrained emissions leading to an increase to 770 ppm (SRES-A1B); and (3) as in SRES-A1B but with a CO2 increase of 550 ppm (SRES-B1). The authors found that most models predicted an increase in seasonal ice extent over the next century. Their projected trends for the three emission scenarios are shown in Figure 3. These were converted to percent change relative to the WC 1979–2003 average for ease of comparison. The steepest projected decadal trend, that of the SRES-A2 scenario (1.99 ± 1.47 × 105 km2 or 2.4 ± 2.0%), is close to that observed in our study. Hence in the last 24 years the areal extent of the SIZ has increased at a rate comparable to that forecasted under the “worst case” (SRES-A2) scenario. We may conclude that the magnitude and rate of changes in SIZ extent over recent decades are consistent with IPCC AR4.
4. Spatio-temporal Variability
 Changes in the extent of the SIZ have not occurred equally in space. During the late 19th and early 20th centuries, seasonal ice appeared more frequently than average in the southern portion of the Barents and Greenland-Iceland-Norwegian (GIN) Sea, and less frequently in the Eurasian coastal region, and in Baffin Bay (Figures 4a–4c). Beginning in the mid-century the areas of preferred seasonal ice formation shifted north (Figures 4d–4e), closer to the mean long-term position (Figure 4h). The northward shift accelerated in the early 1970s, when the decline in sea ice cover intensified and multi-year ice began to melt around the Arctic Ocean to be replaced by first-year ice (Figure 4f). In the last 13 years there has been an Arctic-wide retreat of the summer ice cover, dominating over the winter decline, which has resulted in a net increase in the extent, and a migration of, the SIZ to the peripheral Arctic seas (Figure 4g).
5. Implications for Arctic Ocean Circulation
 The change of multiyear to seasonal ice may have strong implications for the energy and freshwater budgets in the Arctic Ocean [Zhang and Walsh, 2006]. The influence of freeze-thaw processes on the water properties of the Arctic Ocean is the focus of intense study by oceanographers [e.g., Aagaard and Woodgate, 2001]. The freezing of seawater in winter releases salts and forms dense brines which sink and drive haline convection. If seasonal freezing and melting are balanced, this process typically results in shallow convection (∼50 m), but this depth may vary spatially according to regional differences in stratification [Aagaard et al., 1985]. However because wind-driven ice divergence, freezing and melting rates vary spatially, so does net sea ice production. On a large scale, the Arctic Ocean is an area of net ice production, predominantly balanced by ice export through Fram Strait and melting in the GIN Sea [Steele and Flato, 2000]. Hence one might expect that the increased extent of seasonal ice, which is occurring primarily in the Arctic Ocean, should lead to a net salinification of its surface waters, winter ice growth rates and all other fluxes of salt and freshwater being equal. This occurs mainly because ice growth and brine production rates are greater under thin first year-ice (18–70 mm ice/day) than under thicker multi-year ice (∼0–4 mm ice/day) [Koerner, 1973]. Observations suggest that most of the upper Arctic Ocean has become saltier since the late 1970s [Swift et al., 2005], even though freshwater fluxes to the Arctic have increased [Peterson et al., 2006]. The salinity increase has affected mostly the upper 50 m of the ocean, leading Swift et al.  to suggest that the decreasing perennial ice cover has proportionally increased the area available for new ice formation and brine production. Increased surface salinity was also reported in the upper Eurasian Basin during the early 1990s. This was explained by eastward diversion of Siberian inflow waters [Steele and Boyd, 1998], but also by extensive freezing of open water and thinner ice conditions [Johnson and Polyakov, 2001]. In this case, open water conditions resulted from an anomalous atmospheric cyclonic circulation over the eastern Arctic, linked with a more positive North Atlantic Oscillation (NAO), which caused strong southerly winds and ice divergence along the Eurasian coast [Hu et al., 2002]. These observations are consistent with our own: over the past three decades the SIZ has expanded in most peripheral Arctic seas and the Eurasian coast, as well as the Chukchi Sea and Baffin Bay, are now prime locations for the formation of seasonal ice (Figures 4f and 4g).
 The influence of ice export rates on the freshwater budget of the GIN Sea and its effect on the global THC have been well studied [Aagaard and Carmack, 1989; Holland et al., 2001; Peterson et al., 2006]. However the effect of increased salinification of the Arctic Ocean on water exchanges with the GIN Sea and North Atlantic Ocean is less clear. Two main mechanisms are known to cause thermohaline convection in the Arctic Ocean: boundary current deepening and shelf convection. In the first case the two branches of the North Atlantic current entering the Arctic, the Fram Strait and Barents Sea branches, lose heat by atmospheric cooling and freshen slightly by melting sea ice, ultimately increasing in density, sinking and circulating anti-clockwise around the Arctic Ocean sub-basins at intermediate depths (∼200–1000 m) [Aagaard et al., 1985]. In the shelf convection process, brine rejection during ice freezing leads to accumulation of high salinity waters on shelf bottoms, which eventually cascade down the shelf slope as gravity plumes and replenish intermediate and deep (>∼1000 m) waters of the Arctic Basin. Shelf convection is most active in areas with frequent ice removal and sustained freezing [Aagaard et al., 1985]. Both ventilation processes are affected by sea-ice. In particular, shelf-driven ventilation will be enhanced under an accentuated sea-ice seasonal cycle. The picture is less clear for boundary currents but as sea-ice retreats in the Barents Sea, the density of the Barents throughflow may increase as dilution by ice melt ceases, resulting in even deeper ventilation [Aagaard and Woodgate, 2001]. Intermediate waters are well connected with the ventilating regions of the GIN Sea and have relatively short replacement times (10–30 years) compared to deep waters (>100 years), making them susceptible to environmental changes at the surface [Aagaard et al., 1985; Macdonald and Bewers, 1996].
 We thus anticipate that the expanding SIZ and its migration to the shelf areas will likely increase the ventilation of intermediate and deep waters of the Arctic Basin. A possible consequence of this could be an increased salinity or advection of Arctic deep/intermediate waters to the convective gyre of the GIN Sea and an increasing contribution of these waters to NADW formation [Aagaard et al., 1991]. No convection to the bottom of the GIN Sea has occurred recently and the GIN Sea has gradually been filled with Arctic Ocean deep waters [Rudels, 1995; Meincke et al., 1997]. Enhanced ventilation of the Arctic Basin as we evolve toward a summer ice-free Arctic Ocean may then represent a negative feedback on the possible slowing down of the global THC by freshwater dilution in the GIN Sea. For example, increased ice export from the Arctic under a more positive NAO could result in a freshwater pulse in the GIN Sea, reduced convection and slowing down of the THC, whereas the resulting thinner ice and increased open water conditions in the Arctic Ocean would drive southward advection of deep waters to sustain NADW formation and the global THC. Critical to this discussion is the extent to which shelf-driven convection ventilates the Arctic Ocean, as this remains a debated topic and one that deserves further elucidation [Aagaard and Woodgate, 2001].
 This study has shown that the seasonal ice zone has been gradually expanding since 1870, with a marked increase during the past three decades. This recent increase rate matches model predictions for “worst case” type greenhouse gases emission scenarios. Paleoceanographic evidence for enhanced brine expulsion rates in the western Arctic during the peak warmth of the early Holocene (8–9 ka BP) suggests enhanced seasonal ice formation at the time, and offers a possible analogue for future conditions under a warmer climate [Fisher et al., 2006]. The expansion and migration of the SIZ over the Arctic continental shelves is likely to increase the thermohaline ventilation of the Arctic Basin, eventually promoting NADW formation driving the global THC.
 Financial support to C. Kinnard was provided by the Natural Sciences and Engineering Research Council of Canada.
- 1989), The role of sea ice and other fresh water in the Arctic circulation, J. Geophys. Res., 94(C10), 14,485–14,498.
- 2001), Some thoughts on the freezing and melting of sea ice and their effects on the ocean, Ocean Modell., 3, 127–135.
- 1985), Thermohaline circulation in the Arctic Mediterranean Seas, J. Geophys. Res., 90(C3), 4833–4846.
- 1991), Saline outflow from the Arctic Ocean: Its contribution to the deep waters of the Greenland, Norwegian, and Iceland Seas, J. Geophys. Res., 96(C11), 20,433–20,441.
- 2003), 30-year satellite record reveals contrasting Arctic and Antarctic decadal sea ice variability, Geophys. Res. Lett., 30(18), 1970, doi:10.1029/2003GL018031.
- 2002), A rapidly declining perennial sea ice cover in the Arctic, Geophys. Res. Lett., 29(20), 1956, doi:10.1029/2002GL015650.
- 2006), Natural variability of Arctic sea ice over the Holocene, Eos Trans. AGU, 87(28), 273.
- 2001), The role of ice-ocean interactions in the variability of the North Atlantic thermohaline circulation, J. Clim., 14, 656–675.
- 2002), NAO influence on sea ice extent in the Eurasian coastal region, Geophys. Res. Lett., 29(22), 2053, doi:10.1029/2001GL014293.
- 2001), The Laptev Sea as a source for recent Arctic Ocean salinity changes, Geophys. Res. Let., 28(10), 2017–2020.
- 2006), Climatic analysis of sea-ice variability in the Canadian Arctic from operational charts, 1980–2004, Ann. Glaciol., 44, 391–402.
- 1973), The mass balance of the sea ice of the Arctic Ocean, J. Glaciol., 12(65), 173–185.
- 1996), Contaminants in the Arctic marine environment: Priorities for protection, ICES J. Mar. Sci., 53, 537–563.
- 1997), The Arctic Ocean–Nordic Seas thermohaline system, ICES J. Mar. Sci., 54, 283–299.
- 2006), Trajectory shifts in the Arctic and subarctic freshwater cycle, Science, 313, 1061–1066, doi:10.1126/science.1122593.
- 1995), The thermohaline circulation of the Arctic Ocean and the Greenland Sea, Philos. Trans. R. Soc. London, Ser. A, 352(1699), 287–299.
- 2007), Perspectives on the Arctic's shrinking sea-ice cover, Science, 315, 1533–1536, doi:10.1126/science.1139426.
- 1998), Retreat of the cold halocline layer in the Arctic Ocean, J. Geophys. Res., 103(C5), 10,419–10,435.
- 2000), Sea ice growth, melt, and modeling: A survey, in The Arctic Ocean Freshwater Budget, NATO Sci. Ser., 2, vol. 70, edited by E. L. Lewis, pp. 549–587, Kluwer, Dordrecht, Netherlands.
- 2005), Long-term variability of Arctic Ocean waters: Evidence from a reanalysis of the EWG data set, J. Geophys. Res., 110, C03012, doi:10.1029/2004JC002312.
- 1978), A data set on Northern Hemisphere sea ice extent, Glaciol. Data Rep. GD-2, part 1, pp. 49–51, http://nsidc.org/pubs/gd/, World Data Cent. for Glaciol., Cambridge, U. K.
- 2001), 20th-century sea-ice variations from observational data, Ann. Glaciol., 33, 444–448.
- 1998), Factors affecting the detection of trends: Statistical considerations and applications to environmental data, J. Geophys. Res., 103(D14), 17,149–17,162.
- 2006), Toward a seasonally-ice covered Arctic Ocean: Scenarios from the IPCC AR4 model simulations, J. Clim., 19, 1730–1747.