Sea ice plays an important role in the climate and ecology of the Arctic. Numerous studies have identified a reduction in Arctic sea ice cover that has occurred over the past several decades, the period for which we have remotely sensed sea ice concentration data. The regional and seasonal expressions of this reduction have not been as extensively studied. In this paper, we describe the reduction in sea ice concentration that has occurred around southern Baffin Island, one of the Arctic's most biologically active regions. We show that a reduction in sea ice concentration, statistically significant at the 95% level or higher in the presence of temporally correlated noise, in the range of 10–20% per decade has occurred in the region with the largest reduction occurring during the early winter. This reduction is consistent with a recent statistically significant surface warming in the region during the fall and early winter.
 Sea ice is an important component of the Arctic climate system as a result of its high albedo [Budyko, 1969] and its role as an insulative medium that restricts the exchange of heat, moisture and momentum between the atmosphere and ocean [Randall et al., 1998]. Its growth and decay also act to modify the salinity of the surface waters thereby playing a role in the thermohaline circulation of the ocean [Dickson et al., 1988]. Sea ice is also important in the ecology of the Arctic in that it provides a medium for organisms to live within as well as providing a substrate for animals, including humans, to walk and hunt on [Hsiao, 1992; Ferguson et al., 2000; Henshaw, 2003].
 It follows that changes in the Arctic sea ice cover have the potential to impact on many aspects of the climate and ecology both in the Arctic and elsewhere. For example, variability in sea ice cover in the Greenland, Iceland and Norwegian Seas has been proposed as a modulator in the intensity and location of deep water formation in the North Atlantic [Manabe et al., 1992; Schaeffer et al., 2004]. With regard to the ecology of the Arctic, the timing of sea ice breakup in southern Hudson Bay has been shown to impact on the health and distribution of polar bears in the region [Stirling et al., 1999, 2004]. Changes in sea ice cover can also have an impact on the Inuit who rely on it for hunting, fishing and transportation [Riewe, 1991; Nichols et al., 2004].
 A number of studies over the past decade have documented a reduction in Arctic sea ice cover [Maslanik et al., 1996; Bjørgo et al., 1997; Johannessen et al., 1999; Parkinson et al., 1999; Parkinson and Cavalieri, 2002]. All of these studies make use of passive microwave observations from satellites that exploit the difference in emissivity between the solid and liquid phases of water to retrieve sea ice concentration. For the Arctic as a whole, there appears to be a reduction in sea ice concentration of approximately 3–5% per decade over the period for which we have passive microwave observations, 1979–present [Bjørgo et al., 1997; Johannessen et al., 1999]. There is a pronounced seasonal cycle to this reduction which has a strong regional expression. For example, in the Sea of Okhotsk the reduction is largest during the months of April and May, while over the Arctic Ocean, the reduction is largest in September [Parkinson and Cavalieri, 2002].
 There is however much still to be documented with regard to the regional and seasonal characteristics of this reduction in sea ice. In this paper, we describe the characteristics of the reduction in sea ice concentration that has occurred around southern Baffin Island, one of the Arctic's most biologically active regions [Dayton, 1990; Stirling, 1997; Clarke and Harris, 2003].
2. Data and Methods
 We will make use of the sea ice concentration data derived using both the NASA Team [Cavalieri et al., 1999] and the Bootstrap [Comiso, 2005] algorithms. Both datasets are based on the same set of passive microwave measurements from the Scanning Multichannel Microwave Radiometer (SSMR) on Nimbus 7 and the Special Sensor Microwave/Imager (SSM/I) onboard satellites of the U.S. Defense Meteorological Satellite Program (DMSP). The SMMR operated from late 1978 to 1987, while the SSM/I has been operational in a sequential manner since 1987. There was a six week overlap in 1987 that allowed for an intercomparison of the SSMR and SSM/I instruments [Bjørgo et al., 1997; Cavalieri et al., 1999]. Both instruments collect microwave radiances at multiple frequencies at both horizontal and vertical polarizations that allow one to distinguish between different surface types. The two datasets use different combinations of these channels to estimate sea ice concentration. Both algorithms produce sea ice concentration data on a 25 km horizontal scale.
 In coastal regions, problems exists in both datasets that arise from the combination of the relatively coarse resolution of the underlying microwave sensors and the large difference in brightness temperature between ocean and land [Maslanik et al., 1996; Cavalieri et al., 1999]. A number of approaches have been developed to help minimize this effect by identifying the problem grid points in the sea ice concentration datasets. For example, the NASA Team algorithm identifies near shore grid points that have the potential to be contaminated by this effect and then adjusts the sea ice concentration at these points if open water is present in their vicinity [Cavalieri et al., 1999]. On the other hand, the Bootstrap algorithm uses the higher spatial resolution microwave radiances to identify areas impacted by this effect [Comiso, 2005]. For this paper, no additional processing of the data was done, other than that described above, to identify areas impacted by this effect. In this paper, we have made use of the monthly mean sea ice concentrations from the two datasets for the 26-year period from January 1979 to December 2004.
 A comparison of the NASA Team and Bootstrap algorithms show only small differences in the central Arctic during the winter but more widespread differences during the summer and in regions of seasonal ice cover [Comiso et al., 1997]. A conclusion reached from this comparison is that users should be aware of the strengths and weaknesses of the various algorithms and select one that is best suited to the task at hand. In the area of interest, the difference in sea ice concentration as computed by the two algorithms was on the order of 15–20% with the Bootstrap sea ice concentration generally being higher than that of the NASA team [Comiso et al., 1997]. An exception occurred during the fall where the NASA team algorithm typically indicated the presence of sea ice in Frobisher Bay while the Bootstrap algorithm did not. In this instance, a comparison with an independent climatology [Canadian Ice Service, 2002] showed greater agreement with the Bootstrap sea ice concentration. For this reason, we have chosen to use the Bootstrap sea ice concentration dataset in our analysis. Similar results were however obtained using the NASA Team sea ice concentration dataset.
 Our primary interest in this paper is the identification of changes in seasonal sea ice cover surrounding southern Baffin Island. Accordingly, we created a monthly mean climatology of sea ice concentration in the region. The trend in sea ice concentration was computed, using the least squares method, on a monthly basis at each grid point in the region of interest. To assess the statistical significance of the trends obtained, we tested the validity of the null hypothesis that the trend at each grid point is indistinguishable from zero. Geophysical time series typically exhibit temporal autocorrelation that results in a tendency for elevated power at lower frequencies [Mann and Lees, 1996]. This characteristic was incorporated into the significance test through a resampling technique [Gershunov and Barnett, 1998] that made use of 1000 synthetic time series constructed to have the same power spectrum as the monthly mean sea ice concentration time series at each grid point but with randomized phases of the individual Fourier components [Rudnick and Davis, 2003]. The significance of the trends was confirmed with an independent method that used a reduced effective sample size in the calculation of the standard error in the trend and in the application of the t-test [Angell, 1999; Santer et al., 2000a].
 The sea ice cover surrounding southern Baffin Island has a very strong annual cycle with close to 100% ice cover in the winter (January–March) and close to 0% ice cover in the early fall (August–September) [Canadian Ice Service, 2002]. Accordingly, the periods of most rapid changes in ice cover in the region occur during the early summer and early winter. In Figure 1, we present the monthly mean sea ice concentration in the region of interest during representative months in these ‘shoulder’ seasons, July and November. Although these months represent very different sea ice conditions, there nevertheless is a remarkable symmetry in its distribution during these two months. Along the southeast coast of Baffin Island as well as in its two large fiords, Cumberland Sound and Frobisher Bay, the sea ice concentration is on the order of 50–70%. The Hudson Strait to the south of Baffin Island, the strait that connects Hudson Bay to the Labrador Sea, has sea ice concentrations on the order of 10–20% during these months. In the Foxe Basin, sea ice concentrations during these months are typically in the order of 60–80%. Sea ice is also present in Disko Bay along the west coast of Greenland at concentrations of approximately 20%.
Figure 2 provides information on the annual cycle of sea ice cover around southern Baffin Bay, defined to be the oceanic area contained within the region 61°N to 70°N and 80°W to 60°W, for the period 1979–2004. Also shown is the annual cycle for the periods 1979–1991 and 1992–2004. The reduction in sea ice concentration throughout the annual cycle that has occurred over the period for which we have data is apparent. The reduction is largest in magnitude during the ‘shoulder’ seasons of early summer and early winter when the sea ice concentration is undergoing its largest changes.
Figure 3 shows the monthly mean trend in sea ice concentration in the southern Baffin Bay region, as defined above, for the period 1979–2004. Also shown are trends that are significant at the 5% and 1% levels calculated using the method described in section 2. Consistent with the results presented in Figure 2, the trends are largest during the months of July and November when the reduction in sea ice concentration was approximately 8 and 10%/decade respectively. Throughout the shoulder seasons, April to August and October to December, the trends are statistically significant at the 5% level.
 The spatial distribution of the trend in sea ice cover is shown in Figure 4. Trends are only shown for July and November, the months during which the reduction in sea ice concentration is largest. From Figure 4, it is clear that a statistically significant reduction in sea ice concentration has occurred in almost all the oceanic areas surrounding southern Baffin Island. Reductions as large as 20%/decade have occurred in Cumberland Sound with reductions of approximately 10–15%/decade taking place in Frobisher Bay, the Hudson Strait and Foxe Basin. There are some seasonal differences in the reduction with Foxe Basin and Cumberland Sound having the largest reduction in November, while the Hudson Strait has the largest reduction in June. Statistically significant reductions in sea ice concentration on the order of 4%/decade have also occurred in Disko Bay during November.
 In this paper, we have examined the seasonality of the trend towards reduced sea ice concentration surrounding southern Baffin Island. Sea ice in the region plays an important role in the climate as well as having a significant impact on the ecology of the region. We have found that a reduction in sea ice cover has occurred in the region with the largest reductions taking place in the early summer when the sea ice cover is retreating and during the early winter when it is expanding. The rate of reduction in sea ice concentration during these periods is in the range of 10–20%/decade and is statistically significant at the 5% level throughout most of the region. Although results from the Bootstrap algorithm were presented, similar trends were observed with the NASA Team algorithm.
 Both the NASA Team and Bootstrap algorithms have difficulty retrieving sea ice concentration during the shoulder seasons [Comiso et al., 1997; Agnew and Howell, 2003] and so there is a natural concern that this may impact on the results presented in this paper. Pending further validation efforts and improvements to the algorithms, it is unclear at present how to quantify this impact. However, assuming that these systematic errors have been constant over time; one can state that the trends that we have presented should be independent of these errors. Maslanik et al.  reached a similar conclusion with respect to reduction in summer ice cover over the Arctic Ocean.
 These reductions are consistent with the experience of Inuit in the region who rely on sea ice cover for transportation and hunting and who have noticed a thinning in the ice as well a lengthening of the ice free season [Nunavut Tunngavik Inc., 2001; Krupnik and Jolly, 2002]. Figure 5 shows the trend in surface air temperature at Iqaluit (Figure 1) for the period of the sea ice concentration data set, 1979–2004, from the Canadian Historical Temperature Database [Vincent and Gullett, 1999]. The statistical significance of the trend was assessed with the resampling method described above. From Figure 5, one can see that a warming on the order of 1–2°C/decade, statistically significant at the 95% level in the presence of temporally correlated noise, has occurred during fall and early winter during the period for which we have sea ice concentration data. Temperature trends are highly sensitive to the choice of record length [Santer et al., 2000b; Lindsay and Zhang, 2005]. Over the entire period for which observations exist at Iqaluit, 1946–2004, only the warming in the fall is statistically significant at the 95% level. For a longer term perspective, a summer temperature reconstruction based on varved sediments from Upper Soper Lake in southern Baffin Island indicate that the warming has been occurring in the region since the turn of the 20th century [Hughen et al., 2000].
 The trend in the surface temperature data suggests that the fall warming is responsible for the decrease in sea ice concentration during this time of the year and that the decrease during the spring is a result of the preceding fall's reduction in sea ice growth. The ice albedo feedback process would of course amplify any warming resulting from a reduction in sea ice cover and so it is difficult to unambiguously identify cause and effect in this coupled system. It is interesting to note that most climate models predict the largest warming in the Arctic will occur during the fall under increased greenhouse gas concentrations, arising in large part from this feedback process [Moritz et al., 2002].
 Given the relative shortness of the sea ice concentration datasets, typically on the order of 25 years, it is unclear at this time if the observed reductions are indicative of a secular trend associated anthropogenic climate change [Serreze et al., 2000] or the result of low frequency variability in the climate system [Deser et al., 2000; Morison et al., 2000]. There is some evidence that the Arctic climate system may have recently undergone a transition in which local thermodynamic processes, such as ice albedo feedback, now control sea ice cover, independent of low frequency climate variability such as the North Atlantic Oscillation or the Arctic Oscillation [Lindsay and Zhang, 2005]. Continued monitoring of the sea ice in the region and the study of its impacts on the climate system as well as the regional ecology should be undertaken in order to better characterize the impacts that arise from this reduction.
 Access to the NASA Team and Bootstrap sea ice concentration data sets was provided by the National Snow and Ice Data Center, Boulder, Colo. Access to the Canadian Historical Temperature Database was provided by the Meteorological Service of Canada. The author thanks D. Barber, J. Hanesiak, and two anonymous reviewers for helpful comments on a version of this paper. Funding was provided by the Canadian Foundation for Climate and Atmospheric Sciences.