The Hudson Bay System of the Canadian Arctic includes Hudson Bay, Hudson Strait and Foxe Basin, with Foxe Basin being the least studied from a climatological perspective. We examined the temporal and spatial variation of seasonal sea ice in Foxe Basin, using time series and spatial clustering analyses. For the period of 1971 to 2018, time series of sea-ice breakup and freeze-up dates and ice-free season length at 24 grid points were generated from sea-ice charts. The temporal analyses indicated a spatially nuanced response to a warming climate with statistically significant earlier breakup dates, later freeze-up dates and longer ice-free seasons, consistent with previous work. Freeze-up dates and ice-free season length correlated strongly with coincident air temperatures. This link was weaker for breakup dates, as also found in nearby Hudson Bay and Hudson Strait, and likely reflects dependence on antecedent sea-ice and ocean temperature conditions. The spatial analysis revealed patterns in sea-ice behaviour consistent with the ocean flow regime in the basin and the presence of polynyas along its west coast. The spatial clustering was not as predictably coherent as in Hudson Bay and Hudson Strait, which does not bode well for navigation in this region.
Introduction
The Arctic region, as a result of increased greenhouse gas emissions, is experiencing an increase in air temperature at double the rate as the rest of the world since about the mid-twentieth century as a result of the ice albedo feedback (Comiso, 2003; Serreze and Barry, 2011; Comiso and Hall, 2014; Cao and Liang, 2018; Carvalho and Wang, 2020; Marquardt Callow et al., 2020; Zheng et al., 2021; Nielson-Englyst et al., 2023). Directly linked to the Arctic Ocean is Foxe Basin, an inland sea and part of the Hudson Bay System, which also includes Hudson Bay and Hudson Strait. The ice platform of Foxe Basin is central to the hunting and well-being of local communities (Ford et al., 2009; Laidler et al., 2009). Of the three basins within the Hudson Bay System, Foxe Basin is the least studied, both temporally and spatially, from a climatological perspective.
The physical geography of Foxe Basin has been the subject of a number of studies exploring changes in the sea-ice cycle (Gagnon and Gough, 2006; Laidler et al., 2009; Galbraith and Larouche, 2011; Hochheim and Barber, 2014; Andrews et al., 2018), sea-ice thickness (Gagnon and Gough, 2006; Landy et al., 2017), atmospheric temperature (Gagnon and Gough, 2006; Laidler et al., 2009; Galbraith and Larouche, 2011; Hochheim and Barber, 2014; Leung and Gough, 2016; Bello and Higuchi, 2019) and deep water production (Defossez et al., 2008, 2010). The changes in the sea-ice cycle have led to the analysis of observed and potential changes to ship navigation (Tivy et al., 2011; Pizzolato et al., 2014; Pizzolato et al., 2016; Andrews et al., 2017; 2018), an important consideration for this work.
Central to Foxe Basin’s climatology is a full seasonal sea-ice cycle. Sea ice is mainly produced locally, although some sea ice enters Foxe Basin from the Arctic archipelago via Hecla and Fury Strait at the northwest corner of the basin. Sea ice forms in the basin in late October/early November, and the breakup of the sea ice occurs in July (Gagnon and Gough, 2006; Laidler et al., 2009; Galbraith and Larouche, 2011; Hochheim and Barber, 2014; Andrews et al., 2018). Sea ice is dynamic following atmospheric and oceanic currents in the basin. This dynamic allows for the periodic creation of polynyas (open water), in particular at the northwest part of the basin near Hecla and Fury Strait and along the western coast of the basin including near Hall Beach, south of Hall Beach and further south near Lyon’s Inlet (Barber and Massom, 2007; Defossez et al., 2008; 2010).
Most relevant from this literature to the purposes of this paper, we review in more detail the published rates of change of the two main metrics of the seasonal sea-ice cycle: the timing of breakup and the timing of freeze-up in Foxe Basin. Gagnon and Gough (2006) examined the sea-ice cycle, ice thickness and concurrent temperatures at Hall Beach, along the western coast of Foxe Basin (Figure 1). Using sea-ice data that spanned from 1959 to 1998, they found a statistically significant trend for earlier breakup dates of approximately 0.5 days per year with a concurrent spring temperature that exhibits a statistically significant warming trend of 0.5°C/decade. While the freeze-up period was warming at the same rate, no trend in sea-ice freeze-up was detected. Laidler et al. (2009) examined the sea-ice conditions, concurrent air temperature and wind conditions at and near Igloolik. Igloolik is located on an island in the northwest part of Foxe Basin. For the time period of 1982–2005, using a 5/10 sea-ice concentration threshold, they found that breakup was occurring 0.6 days per year earlier and freeze-up was later by the same amount. Concurrent surface air temperatures were warming at a rate of 1–2°C/decade. Galbraith and Larouche (2011) using two different time periods, 1971–2009 and 1990–2009, found an acceleration of the rates for the latter period to 0.9 days per year with a clear linkage to air temperature. Hochheim and Barber (2014) compared the breakup dates for two time periods, 1980–1995 and 1996–2010, and found the latter time period to have ice breakup occurring 1.5 weeks earlier or 0.7 days per year earlier. In a similar analysis for freeze-up dates, these dates were later by 2 weeks for the latter time period, or 0.9 days per year later, consistent with both Laidler et al. (2009) and Galbraith and Larouche (2011). Andrews et al. (2018) updated the climate record by adding temperature and sea-ice data up to and including 2014 in Foxe Basin. For the 1980–2014 period, breakup occurred 0.7 days per year earlier on average and freeze-up 0.62 days per year later. While consistent with previous work, given the differing methodologies and timeframes there is nonetheless some indication that the rate of change may be decreasing in intensity, especially for freeze-up, a point further explored in this work while updating the sea-ice and temperature data to 2018.
Foxe Basin within the Hudson Bay System of the Canadian Arctic. Local climate stations for obtaining surface air temperatures are located at Igloolik (69.37°N, 81.82°W), Hall Beach (68.79°N, 81.24°W) and Kinngait (64.23°N, 76.54°W).
Foxe Basin within the Hudson Bay System of the Canadian Arctic. Local climate stations for obtaining surface air temperatures are located at Igloolik (69.37°N, 81.82°W), Hall Beach (68.79°N, 81.24°W) and Kinngait (64.23°N, 76.54°W).
Of the five papers reviewed above on the changing sea-ice conditions in Foxe Basin, four of them (Gagnon and Gough, 2006; Laidler et al., 2009; Galbraith and Larouche, 2011; Hochheim and Barber, 2014) establish clear linkages with the concurrent local temperatures. Not explored is the nature of the changes in the surface air temperatures. Is surface air temperature the driver of change, as generally assumed in a globally warming climate, or is the change, or part of the change, in sea-ice conditions a result of changes in the sea ice itself (part of the ice-albedo feedback)? Leung and Gough (2016) looked to air mass theory to explain the changes in air temperature in the region. They found that the dominance of the coldest air mass, dry polar, had weakened for Hudson Bay and Foxe Basin and that for both regions the dominant air mass is now moist polar, reflective not only of the greater access to open water but also larger shifts in the interplay between these air masses. They also established, however, that the change in air mass dominance did not fully explain the changes in local temperatures and that a warming of both air masses was also likely occurring (Leung and Gough, 2016).
Also contributing to sea-ice variability are large scale oscillations of the atmosphere, such as El Nino/Southern Oscillation, North Atlantic Oscillation, Arctic Oscillation and Pacific Decadal Oscillation (Wang et al., 1994; Mysak et al., 1996). These oscillations are communicated from region of origin through atmospheric teleconnections and realized in Foxe Basin as a temperature anomaly, often associated with changes in air mass distribution (Leung and Gough, 2016). These oscillations also contribute to inter-annual variability of atmospheric temperatures and by extension to sea ice in Foxe Basin.
Following on these considerations, we introduce one final thread that contributes to regional temperature and sea-ice variability. While concurrent local and regional temperatures largely explain the sea-ice cycle as noted in several studies reviewed above, there is evidence in other locations of the Hudson Bay System that elements of ‘climate memory’ may play an important role. Gough and Houser (2005) introduced the term in examining the nature of Hudson Strait sea ice and used this memory to forecast future ice conditions in the Strait. The ‘memory’ is climate information passed from year to year via the sea ice. For example, breakup is a function of the air temperature required to melt the ice but the total amount of ice is a function of the both air and water temperatures of the previous season, a form of preconditioning (Gough and Houser, 2005; Serreze and Barry, 2011; Stroeve et al., 2012; 2014; Stroeve et al., 2016; Andrews et al., 2017; 2018; Kowal et al., 2024a; 2024b). This concept also can be applied to freeze-up and its dependence on breakup timing (Serreze and Barry, 2011; Stroeve et al., 2016; Kowal et al., 2024a; 2024b).
In this paper we explore the relative roles of direct warming and preconditioning mechanisms in Foxe Basin. We examine, in a more comprehensive fashion than previous work, the temporal and spatial nature of sea ice in the basin, extending the sea-ice times series to 2018 to facilitate both times series analysis and spatial cluster analysis. The former contributes to a broader assessment of the impacts of climate change in the eastern Arctic. The latter provides the spatial granularity that is key from the perspectives of local indigenous communities and ship navigation.
Following themes introduced above, we address two main research questions. First, how statistically significant are the temporal changes in the breakup, freeze-up and ice-free seasons for the 24 grid points representing Foxe Basin for three time periods, 1971–2018, 1971–1994, 1995–2018, and their relationships to concurrent temperature change? Second, using a spatial cluster analysis, what is the spatial nature of breakup, freeze-up and ice-free season length for Foxe Basin for these three time periods?
Methods
Temperature data
Surface temperature data were obtained from the Environment and Climate Change Canada archive (https://climate.weather.gc.ca/historical_data/search_historic_data_e.html) for three local climate stations: Igloolik (69.37°N, 81.82°W), Hall Beach (68.79°N, 81.24°W) and Kinngait (64.23°N, 76.54°W) (Figure 1). Annually averaged mean temperatures of the day were used to assess the temporal evolution of temperature in the Foxe Basin region. Data for Igloolik were available for the following years: 1978–2002, 2009 and 2010. For Hall Beach the data were available for 1971–2006, 2009, 2012 and 2014. Kinngait had data for 1971, 1980–1993, 1995–1998 and 2000–2006.
Sea ice data
A grid of 24 points was developed to represent Foxe Basin (Figure 2). In order to determine the breakup and freeze-up dates from 1971 to 2018 for each of the sampling locations for Foxe Basin, the same methodology described below and utilized by others (Houser and Gough, 2003; Gagnon and Gough, 2005; Chambellant et al., 2012; Clark et al., 2016; Archer et al., 2017; Kowal et al., 2017; 2023; 2024a; 2024b) was employed.
Foxe Basin grid points. The distribution of 24 grid points used in the temporal and spatial analyses of sea-ice breakup, freeze-up and length of ice-free season.
Foxe Basin grid points. The distribution of 24 grid points used in the temporal and spatial analyses of sea-ice breakup, freeze-up and length of ice-free season.
The Canadian Ice Service has reported sea-ice concentration data for the Arctic region since 1971 on a weekly basis, with the exception of the winter months of January to May when the frequency is bi-weekly. The sea-ice concentration data are expressed in tenths (from 0 to 10/10), indicating the fraction of the surface area that is covered with ice at a given location. The Canadian Ice Service maps are produced by using a range of available information such as satellite imagery, ship and aircraft reconnaissance, shore observations and climate data. The maps are available online at https://iceweb1.cis.ec.gc.ca/Archive/page1.xhtml?lang=en. Here, the sea-ice concentrations were obtained for every sampling point superimposed over the study area (Figure 2), and the dates of the ice breakup, ice freeze-up and the derived ice-free season were catalogued for each year (from 1971 to 2018) with an accuracy of ±1 week, as has been done in other work (Etkin, 1991; Stirling et al., 1999; Houser and Gough, 2003; Gough et al., 2004; Gagnon and Gough, 2005; Chambellant et al., 2012; Clark et al., 2016; Archer et al., 2017; Kowal et al., 2017; 2023; 2024a; 2024b). The ice breakup date is defined as the first date when the ice concentration is 5/10 or less during the spring and summer months, while the ice freeze-up date is the earliest date when the ice concentration has reached 5/10 or more between October and December. These specific thresholds, navigationally motivated, were used to determine the breakup and freeze-up dates in accordance with those utilized by both the Canadian Sea Ice Service and the World Meteorological Organization, although other thresholds have been used (Laidler et al., 2009). The breakup and freeze-up dates were expressed numerically as the ordinal day of the year, where January 1 is day 1 and December 31 is day 365, unless there was a leap year in which case December 31 would be the day 366 of the year (Gagnon and Gough, 2005; Kowal et al., 2017; 2023; 2024a; 2024b).
Time series analysis
The Kendall correlation is a non-parametric measure of correlation that assesses monotonicity of trends without assuming a normal distribution of data. When this correlation is with time, it is referred to as the Mann-Kendall correlation that tests (statistically) whether a trend is unchanging against an alternative hypothesis that the trend is increasing or decreasing (Neave and Worthington, 1988; Gagnon and Gough, 2005; Mohsin and Gough, 2010; Kowal et al., 2017; 2023; 2024a; 2024b). The test is based on the assumption of independence among the observations. However, observations in a time series can be autocorrelated, which can influence the detection of a statistically significant trend. Thus, checking for autocorrelation and adjusting the test if necessary is crucial (Gagnon and Gough, 2005; Mohsin and Gough, 2010; Kowal et al., 2017; 2023; 2024a; 2024b).
In order to account for the possibility of autocorrelation within each of the three data sets (for the three metrics), the adjustment was performed using the adjustment method of Hamed and Rao (1998). R’s fume package’s mkTrend function carries out the adjusted and unadjusted Mann-Kendall tests. Only the adjusted values are presented in this work. The Mann-Kendall tests for trend significance were performed for each of the three metrics for each of the three separate time periods, 1971–2018, 1971–1994 and 1995–2018. A p-value threshold of less than 0.05 is considered significant (Kowal et al., 2017).
The Theil-Sen method is a non-parametric slope estimator of the rate of change of the considered variables that is not affected by outliers (Sen, 1968; Gagnon and Gough, 2005; Kowal et al., 2017; 2023; 2024a; 2024b). The slope represents the rate of change of the considered variables (breakup date, freeze-up date and ice-free season) in days per year. The Theil-Sen method was used for the three metrics for each of the three time periods, namely the full time period, 1971–2018, and the two distinct sub-periods, 1971–1994 and 1995–2018, using the statistical program R.
Spatial analysis
Cluster analysis (Johnson and Wichern, 2002; Kowal et al., 2023; 2024a; 2024b) is used to divide observations into groups or clusters, based on the values of several observed variables. There are two main types of cluster analysis, hierarchical (Ward’s) and non-hierarchical (K-means). While both were used on the Foxe Basin data, there were no significant differences between the two analyses and so only the Ward’s results are presented. Hierarchical cluster analysis, such as Ward’s, begins with one cluster containing all of the observations that is repeatedly subdivided (Johnson and Wichern, 2002). The process is described below:
Start with each point in a cluster by itself (sum of squares = 0).
Merge two clusters, in order to produce the smallest increase in the sum of squares (merging cost).
Keep merging until k clusters is reached. The merging cost is the increase in sum of squares when two clusters are merged.
The cluster analysis was applied separately to the breakup dates, freeze-up dates and the ice-free season length in order to examine the spatial coherency of these sea ice metrics. The Ward’s method can handle missing data; specifically, the dissimilarity between any pair of locations is calculated using any years for which that pair of locations has non-missing data. The benefit of this approach is that years in which there are missing values are not deleted entirely and therefore more of the data is employed for the analysis. The Ward’s method was run for each of the three metrics for the Foxe Basin grid points for each of the three separate time periods (1971–2018, 1971–1994 and 1995–2018) using the statistical program R (Kowal et al., 2023; 2024a; 2024b). Six clusters were chosen for the Foxe Basin data set.
Results and discussion
Temperature analysis
Three climate-data time series for mean temperature were used to assess the temporal evolution of temperature in the Foxe Basin region: Igloolik, NU (69.37°N, 81.82°W); Hall Beach, NU (68.79°N, 81.24°W); and Kinngait (Cape Dorset), NU (64.23°N, 76.54°W) (Figure 1). These three stations provide a spatial representation of the Foxe Basin region with Igloolik and Kinngait at the north and south ends of the basin, respectively, and Hall Beach in between, although it is nearer to Igloolik.
Resulting time series are shown in Figure 3. All three time series show an increase in temperature during the study period (1971–2014), at a rate of 0.6°C per decade for Kinngait, 0.8°C per decade for Hall Beach and 1.3°C per decade for Igloolik. The temperatures at Kinngait were considerably warmer than those at Igloolik and Hall Beach, which came as no surprise as Kinngait is at the southern extent of the study region (Figure 1). These results are roughly consistent with Laidler et al. (2009) and Ford et al. (2009), who found statistically significant increases in mean temperature at Igloolik (1977–2002) for all seasons except winter. Gagnon and Gough (2006), as part of a broader Hudson Bay system study, did an analysis of the Hall Beach climate record as noted earlier. Using data spanning from 1959 to 1998, they found statistically significant temperature increases for spring and autumn, of approximately 0.5°C per decade, at the low end of the rates found in the current work which included more recent data and thus is suggestive of an accelerated warming in recent decades. The year 2010 was anomalous at Igloolik, as can be seen in Figure 3 (data not available at the other two stations), as well as the final point in that time series, thus likely leading to an overestimate of the trend. A regional warm anomaly in 2010 was also detected in Hudson Strait (Statham et al., 2015; Kowal et al., 2024a).
Atmospheric temperature time series for Foxe Basin. Temperature record (°C) at Igloolik, NU (69.37°N, 81.82°W); Hall Beach, NU (68.79°N, 81.24°W); and Kinngait (Cape Dorset), NU (64.23°N, 76.54°W). The dotted lines indicate the linear regression trend lines.
Atmospheric temperature time series for Foxe Basin. Temperature record (°C) at Igloolik, NU (69.37°N, 81.82°W); Hall Beach, NU (68.79°N, 81.24°W); and Kinngait (Cape Dorset), NU (64.23°N, 76.54°W). The dotted lines indicate the linear regression trend lines.
Sea ice time series analysis
The Mann-Kendall correlation and Theil-Sen slope estimator were used to quantify the temporal trends for each of the 24 grid points. These results are presented in Table 1 for the ice breakup for the three time periods, 1971–2018, 1971–1994 and 1995–2018. The latter two are included to determine if there has been a significant change in behaviour throughout the 48 year period.
Sea-ice breakup in Foxe Basin, quantified by the Mann-Kendall correlation test and Theil-Sen slope estimator, for three time periods
. | 1971–2018 . | 1971–1994 . | 1995–2018 . | |||
---|---|---|---|---|---|---|
Grid Pointsa . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . |
1 | ** | −0.778 | *** | −2.111 | NS | 0.364 |
2 | *** | −0.747 | NS | −0.519 | NS | 0.000 |
3 | *** | −0.692 | NS | −0.455 | NS | −0.845 |
4 | * | −0.516 | NS | 0.800 | NS | −1.000 |
5 | *** | −0.838 | *** | −0.962 | NS | 0.000 |
6 | * | −0.663 | NS | −1.314 | NS | 0.000 |
7 | NS | −0.036 | *** | −1.300 | NS | −0.692 |
8 | ** | −0.281 | NS | −0.098 | NS | −0.118 |
9 | ** | −0.494 | *** | −1.500 | NS | 0.536 |
10 | NS | 0.111 | NS | −0.679 | NS | 1.000 |
11 | ** | −0.710 | *** | −1.833 | NS | −0.858 |
12 | *** | −0.507 | NS | 0.265 | NS | 0.000 |
13 | ** | −0.479 | NS | −0.066 | * | 0.500 |
14 | NS | 0.000 | ** | −1.293 | NS | 0.500 |
15 | NS | −0.309 | *** | −2.425 | NS | 0.407 |
16 | NS | −0.357 | NS | −1.369 | *** | −2.472 |
17 | ** | −0.421 | NS | −0.333 | NS | 0.058 |
18 | NS | −0.083 | NS | −0.580 | NS | 0.133 |
19 | *** | −0.745 | ** | −1.500 | NS | 0.000 |
20 | NS | −0.355 | NS | −1.222 | NS | −0.444 |
21 | NS | 0.000 | NS | −0.757 | NS | −0.083 |
22 | NS | −0.385 | * | −1.783 | NS | 1.111 |
23 | *** | −0.756 | ** | −1.817 | NS | 0.000 |
24 | * | −0.265 | NS | −0.640 | NS | 0.250 |
. | 1971–2018 . | 1971–1994 . | 1995–2018 . | |||
---|---|---|---|---|---|---|
Grid Pointsa . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . |
1 | ** | −0.778 | *** | −2.111 | NS | 0.364 |
2 | *** | −0.747 | NS | −0.519 | NS | 0.000 |
3 | *** | −0.692 | NS | −0.455 | NS | −0.845 |
4 | * | −0.516 | NS | 0.800 | NS | −1.000 |
5 | *** | −0.838 | *** | −0.962 | NS | 0.000 |
6 | * | −0.663 | NS | −1.314 | NS | 0.000 |
7 | NS | −0.036 | *** | −1.300 | NS | −0.692 |
8 | ** | −0.281 | NS | −0.098 | NS | −0.118 |
9 | ** | −0.494 | *** | −1.500 | NS | 0.536 |
10 | NS | 0.111 | NS | −0.679 | NS | 1.000 |
11 | ** | −0.710 | *** | −1.833 | NS | −0.858 |
12 | *** | −0.507 | NS | 0.265 | NS | 0.000 |
13 | ** | −0.479 | NS | −0.066 | * | 0.500 |
14 | NS | 0.000 | ** | −1.293 | NS | 0.500 |
15 | NS | −0.309 | *** | −2.425 | NS | 0.407 |
16 | NS | −0.357 | NS | −1.369 | *** | −2.472 |
17 | ** | −0.421 | NS | −0.333 | NS | 0.058 |
18 | NS | −0.083 | NS | −0.580 | NS | 0.133 |
19 | *** | −0.745 | ** | −1.500 | NS | 0.000 |
20 | NS | −0.355 | NS | −1.222 | NS | −0.444 |
21 | NS | 0.000 | NS | −0.757 | NS | −0.083 |
22 | NS | −0.385 | * | −1.783 | NS | 1.111 |
23 | *** | −0.756 | ** | −1.817 | NS | 0.000 |
24 | * | −0.265 | NS | −0.640 | NS | 0.250 |
aSee Figure 3 for grid point locations.
b***Indicates p < 0.01; **, p < 0.05; *, p < 0.1; and NS, p > 0.1.
For the 1971–2018 time period, 12 of the 24 points had statistically significant (p < 0.05) earlier breakup dates using the Mann-Kendall test. In comparison, Hudson Strait and Hudson Bay had a higher percentage of grid points that were statistically significant (Kowal et al., 2017; 2024a). Figure 4 depicts the spatial distribution of the statistical significance. The non-significant points tended to be clustered in the north and eastern parts of the basin. Of the significant points, the trend rate ranged from −0.28 to −0.84 days per year which translates to 13 to 40 days earlier over the 48-year study period, a less intense result than for Hudson Strait (Kowal et al., 2024a).
Statistical significance of sea-ice breakup times series (1971–2018) in Foxe Basin. The spatial distribution of the results of the Mann-Kendall test for statistical significance of the temporal trend during the period of 1971–2018 for ice breakup dates for Foxe Basin. The strength of the statistical significance is colour-coded, where blue indicates p < 0.01 (***); red, p < 0.05 (**); yellow, p < 0.1 (*); and green, p > 0.1 (NS).
Statistical significance of sea-ice breakup times series (1971–2018) in Foxe Basin. The spatial distribution of the results of the Mann-Kendall test for statistical significance of the temporal trend during the period of 1971–2018 for ice breakup dates for Foxe Basin. The strength of the statistical significance is colour-coded, where blue indicates p < 0.01 (***); red, p < 0.05 (**); yellow, p < 0.1 (*); and green, p > 0.1 (NS).
These results are consistent with the rate of −0.5 days per year found by Gagnon and Gough (2006) near Hall Beach for the period of 1959 to 1999, which is almost identical to point 19 (−0.49 days per year) in this study, the nearest point to Hall Beach. Similarly, Laidler et al. (2009) found a rate of −0.6 days per year for Igloolik for the years 1982–2006. Point 23 of this study, the nearest one to Igloolik, had a slightly higher rate of −0.76 days per year, reflective of the longer, and more recent, times series.
Six of the 24 breakup points had autocorrelation in the time series for the 1971–2018 period. The autocorrelation was more widespread for freeze-up (11 of 24 points) and ice-free season (10 of 24). While the trend analysis accounted for this autocorrelation, the presence of autocorrelation is noteworthy. Unlike the similar analysis for Hudson Strait (Kowal et al., 2024a) there is no obvious spatial coherency to the locations of the autocorrelated points. The only point that has autocorrelation for all three metrics for the 1971–2018 time period is point 16 located along the western coast of Foxe Basin, south of Hall Beach. This point is located at a persistent polynya (Barber and Massom, 2007; Defossez et al., 2008; 2010), which may contribute to the autocorrelation. The sea-ice platform is an effective insulator, cutting off the thermal inertia of the ocean waters. The open water of a polynya enables the ocean water thermal inertia to propagate from year to year.
The first subdivided period, 1971–1994, had nine statistically significant grid points, and the second, 1995–2018, had just one. Of the nine, six were coincident with statistically significant grid points found for the longer period of 1971–2018. However, three were not, which is suggestive of shifting sea-ice patterns in Foxe Basin. The fact that the second period had only one statistically significant point for earlier breakup indicates that changes in Foxe Basin were stronger in the earlier period, a signal that did not emerge in Hudson Strait (Kowal et al., 2024a) and Hudson Bay (Kowal et al., 2017) but was suggested in Andrews et al. (2018). This result may be related to large scale oscillations in the atmospheric temperature (Wang et al., 1994; Mysak et al., 1996).
In summary, the temporal analysis of breakup dates demonstrated consistency of earlier breakup dates in Foxe Basin with previous work that used shorter time series. There was evidence, however, that the rate of change may have been slowing in the most recent decade. No coherent conclusion could be reached by examining autocorrelation in the time series. The strength of the change signal was less (fewer statistically significant grid point time series) compared to Hudson Bay (Kowal et al., 2017) and Hudson Strait (Kowal et al., 2024a). Next, we consider the freeze-up analysis.
The Mann-Kendall correlation and Theil-Sen slope estimator for freeze-up are presented in Table 2 for the three time periods. For the 1971–2018 period, 20 of the 24 points were statistically significant using the Mann-Kendall test, with p-values less than 0.05. The four points that did not achieve statistical significance were located in the northeast part of the basin, in a region reminiscent but more restricted than that for the breakup date analysis. The rate of change for the statistically significant grid points ranged from 0.23 to 1.4 days per year, which indicates 11 to 67 days later in freeze-up over the study period. The more statistically significant behaviour with a north/south orientation, strikingly different from the breakup behaviour, suggests that the freeze-up is more of a response to direct thermal forcing and less dependent on the advection of sea ice into or out of the region. This result is as expected, because the starting point is the end of the ice-free season whereas the breakup may be influenced by the movement of sea ice into the basin through Hecla and Fury Strait as well as redistribution of sea ice within the basin during breakup. This behaviour was also found to be the case in Hudson Strait (Kowal et al., 2024a) and Hudson Bay (Kowal et al., 2024b). It contrasts to what occurs in Hudson Bay where the strongest change signal was first detected during the breakup period (Gagnon and Gough, 2005), although there have been recent significant changes in freeze-up (Kowal et al., 2017; Andrews et al., 2018).
Sea-ice freeze-up in Foxe Basin, quantified by the Mann-Kendall correlation test and Theil-Sen slope estimator, for three time periods
. | 1971–2018 . | 1971–1994 . | 1995–2018 . | |||
---|---|---|---|---|---|---|
Grid Pointsa . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . |
1 | *** | 0.714 | NS | 0.200 | NS | 0.323 |
2 | ** | 0.381 | NS | −0.154 | NS | 0.112 |
3 | *** | 1.440 | NS | 2.273 | NS | 0.243 |
4 | *** | 1.056 | *** | 0.786 | * | 0.600 |
5 | *** | 0.826 | NS | 0.000 | NS | 0.500 |
6 | *** | 0.846 | NS | −0.385 | NS | 0.183 |
7 | ** | 0.569 | NS | 0.633 | NS | 0.333 |
8 | ** | 0.353 | NS | −0.714 | NS | 0.211 |
9 | *** | 0.448 | NS | 0.500 | NS | −0.175 |
10 | *** | 0.600 | NS | −0.167 | NS | −0.154 |
11 | ** | 0.571 | NS | −0.556 | NS | −0.240 |
12 | *** | 0.460 | NS | 0.431 | NS | 0.000 |
13 | *** | 0.383 | NS | 0.071 | NS | 0.000 |
14 | *** | 0.500 | NS | −0.273 | NS | −0.146 |
15 | NS | 0.074 | NS | 0.297 | NS | 0.000 |
16 | *** | 0.536 | NS | 0.000 | NS | 0.000 |
17 | *** | 0.500 | NS | 0.473 | NS | 0.000 |
18 | * | 0.333 | NS | −0.788 | NS | −0.143 |
19 | *** | 0.364 | NS | 0.000 | NS | 0.121 |
20 | * | 0.415 | * | −1.823 | NS | −0.400 |
21 | NS | −0.030 | NS | 0.059 | NS | −0.301 |
22 | *** | 0.714 | NS | 0.950 | NS | −0.160 |
23 | ** | 0.227 | NS | 0.191 | NS | 0.191 |
24 | *** | 0.561 | NS | 0.348 | NS | 0.500 |
. | 1971–2018 . | 1971–1994 . | 1995–2018 . | |||
---|---|---|---|---|---|---|
Grid Pointsa . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . |
1 | *** | 0.714 | NS | 0.200 | NS | 0.323 |
2 | ** | 0.381 | NS | −0.154 | NS | 0.112 |
3 | *** | 1.440 | NS | 2.273 | NS | 0.243 |
4 | *** | 1.056 | *** | 0.786 | * | 0.600 |
5 | *** | 0.826 | NS | 0.000 | NS | 0.500 |
6 | *** | 0.846 | NS | −0.385 | NS | 0.183 |
7 | ** | 0.569 | NS | 0.633 | NS | 0.333 |
8 | ** | 0.353 | NS | −0.714 | NS | 0.211 |
9 | *** | 0.448 | NS | 0.500 | NS | −0.175 |
10 | *** | 0.600 | NS | −0.167 | NS | −0.154 |
11 | ** | 0.571 | NS | −0.556 | NS | −0.240 |
12 | *** | 0.460 | NS | 0.431 | NS | 0.000 |
13 | *** | 0.383 | NS | 0.071 | NS | 0.000 |
14 | *** | 0.500 | NS | −0.273 | NS | −0.146 |
15 | NS | 0.074 | NS | 0.297 | NS | 0.000 |
16 | *** | 0.536 | NS | 0.000 | NS | 0.000 |
17 | *** | 0.500 | NS | 0.473 | NS | 0.000 |
18 | * | 0.333 | NS | −0.788 | NS | −0.143 |
19 | *** | 0.364 | NS | 0.000 | NS | 0.121 |
20 | * | 0.415 | * | −1.823 | NS | −0.400 |
21 | NS | −0.030 | NS | 0.059 | NS | −0.301 |
22 | *** | 0.714 | NS | 0.950 | NS | −0.160 |
23 | ** | 0.227 | NS | 0.191 | NS | 0.191 |
24 | *** | 0.561 | NS | 0.348 | NS | 0.500 |
aSee Figure 3 for grid point locations.
b***Indicates p < 0.01; **, p < 0.05; *, p < 0.1; and NS, p > 0.1.
The subdivided periods, 1971–1994 and 1995–2018, are largely not statistically significant (only one in the first period and none in the second), and in this respect are similar to a contemporaneous Hudson Strait analysis (Kowal et al., 2024a). Laidler et al. (2009) found a later freeze-up of 0.6 days per year at Igloolik for the period of 1982–2006, consistent with points 22 and 23 of the current analysis. Gagnon and Gough (2006) found at Hall Beach no change in the freeze-up dates for the period of 1960–1994. The current analysis using point 19, the nearest one to Hall Beach, in contrast shows a statistically significant change to later freeze-up dates using the more recent sea-ice data. Consistent with Gagnon and Gough (2006) the earlier period of 1971–1994 showed no trend.
In summary, the temporal analysis of freeze-up dates showed stronger and more coherent later freeze-up dates in Foxe Basin than previous work that used shorter time series. In contrast with the breakup analysis most of the grid points had trends that were statistically significant (20 of 24). This result is likely due to a more thermally driven freeze-up, without the complication of dynamically driven, redistributed remnant sea ice that characterizes the breakup, as was also seen in Hudson Bay and Hudson Strait. The strength of the change signal was as strong or stronger (more statistically significant grid point time series) compared to Hudson Bay (Kowal et al., 2017) and Hudson Strait (Kowal et al., 2024a). Next, we consider the analysis of ice-free season length.
The Mann-Kendall correlation and Theil-Sen slope estimator for the ice-free season are presented in Table 3 for the three time periods. For the 1971–2018 period, 18 of the 24 points were statistically significant using the Mann-Kendall test, with p-values less than 0.05. This number of significant points falls between the breakup and freeze-up results, understandably as this metric is derived from the other two. The rate of change ranges from 0.52 to 2.42 days per year which corresponds to a longer ice-free season of 25 to 116 days over the study period. As expected, this result is largely additive to the earlier breakup and later freeze-up of seasonal sea ice, reported above. The largest values occur at the northern entrance of the basin (Hecla and Fury Strait, points 19, 22 and 23) and the southern exit from Foxe Basin (points 1, 3, 4 and 5) with the lowest magnitude in the eastern part of the Basin (points 14 and 15). The subdivided periods, 1971–1994 and 1995–2018, were largely not statistically significant (p < 0.05), with only two points for the first period and four for the latter period.
Length of ice-free season in Foxe Basin, quantified by the Mann-Kendall correlation test and Theil-Sen slope estimator, for three time periods
. | 1971–2018 . | 1971–1994 . | 1995–2018 . | |||
---|---|---|---|---|---|---|
Grid Pointsa . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . |
1 | *** | 1.750 | * | 1.750 | NS | 1.025 |
2 | *** | 1.077 | NS | 0.000 | NS | 0.369 |
3 | *** | 2.419 | NS | 2.818 | ** | 1.781 |
4 | *** | 1.944 | NS | 1.125 | ** | 1.810 |
5 | *** | 1.609 | NS | 0.800 | NS | 0.563 |
6 | ** | 1.667 | NS | 0.357 | NS | 1.151 |
7 | ** | 1.025 | NS | 1.478 | NS | 1.167 |
8 | NS | 0.273 | NS | −1.500 | NS | 0.000 |
9 | *** | 0.824 | ** | 2.000 | ** | −0.975 |
10 | NS | 0.625 | NS | 0.538 | NS | −0.379 |
11 | *** | 1.094 | NS | 0.778 | NS | 0.351 |
12 | *** | 1.037 | NS | 0.000 | NS | 0.000 |
13 | ** | 1.000 | NS | 0.000 | NS | −0.359 |
14 | *** | 0.516 | NS | 0.941 | NS | −0.778 |
15 | NS | 0.359 | *** | 2.721 | NS | 0.000 |
16 | ** | 0.875 | NS | 0.875 | *** | 3.000 |
17 | ** | 1.077 | NS | 0.849 | NS | 0.000 |
18 | NS | 0.000 | NS | 0.000 | NS | 0.000 |
19 | *** | 1.077 | * | 1.286 | NS | 0.415 |
20 | * | 0.769 | NS | 0.000 | NS | 0.000 |
21 | NS | 0.000 | NS | 1.336 | NS | 0.000 |
22 | *** | 1.266 | * | 2.267 | NS | −1.199 |
23 | *** | 1.077 | * | 1.400 | * | 0.200 |
24 | *** | 0.840 | NS | 0.933 | NS | 0.000 |
. | 1971–2018 . | 1971–1994 . | 1995–2018 . | |||
---|---|---|---|---|---|---|
Grid Pointsa . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . | Mann-Kendallb . | Theil-Sen (days per year) . |
1 | *** | 1.750 | * | 1.750 | NS | 1.025 |
2 | *** | 1.077 | NS | 0.000 | NS | 0.369 |
3 | *** | 2.419 | NS | 2.818 | ** | 1.781 |
4 | *** | 1.944 | NS | 1.125 | ** | 1.810 |
5 | *** | 1.609 | NS | 0.800 | NS | 0.563 |
6 | ** | 1.667 | NS | 0.357 | NS | 1.151 |
7 | ** | 1.025 | NS | 1.478 | NS | 1.167 |
8 | NS | 0.273 | NS | −1.500 | NS | 0.000 |
9 | *** | 0.824 | ** | 2.000 | ** | −0.975 |
10 | NS | 0.625 | NS | 0.538 | NS | −0.379 |
11 | *** | 1.094 | NS | 0.778 | NS | 0.351 |
12 | *** | 1.037 | NS | 0.000 | NS | 0.000 |
13 | ** | 1.000 | NS | 0.000 | NS | −0.359 |
14 | *** | 0.516 | NS | 0.941 | NS | −0.778 |
15 | NS | 0.359 | *** | 2.721 | NS | 0.000 |
16 | ** | 0.875 | NS | 0.875 | *** | 3.000 |
17 | ** | 1.077 | NS | 0.849 | NS | 0.000 |
18 | NS | 0.000 | NS | 0.000 | NS | 0.000 |
19 | *** | 1.077 | * | 1.286 | NS | 0.415 |
20 | * | 0.769 | NS | 0.000 | NS | 0.000 |
21 | NS | 0.000 | NS | 1.336 | NS | 0.000 |
22 | *** | 1.266 | * | 2.267 | NS | −1.199 |
23 | *** | 1.077 | * | 1.400 | * | 0.200 |
24 | *** | 0.840 | NS | 0.933 | NS | 0.000 |
aSee Figure 3 for grid point locations.
b***Indicates p < 0.01; **, p < 0.05; *, p < 0.1; and NS, p > 0.1.
In summary, the temporal analysis of ice-free season length was similar to the freeze-up analysis, with 18 of 24 grid points that were statistically significant. The longest ice-free seasons occurred at opposite ends of the basin. The longer season at the southern end of Foxe Basin occurs where it is warmest. The longer season at northern season is the result of a wind-generated polynya that advects sea ice away from Fury and Hecla Strait.
Above, the analysis of air temperature at three climate stations (Figure 3) showed the general warming of the region over the study period. We now explore the direct impact of these changing temperatures at each of the grid points. A correlation analysis was performed for the three sea-ice metrics for the years of coincident surface air temperature data (from Hall Beach, Igloolik and Kinngait). Hall Beach temperature data correlated significantly (p < 0.05) with 22 of the 24 grid points for freeze-up, with lower values for Igloolik (15) and Kinngait (14), likely the result of data paucity (Igloolik) and location (Kinngait). For the ice-free season, there were similar results, with 23, 16 and 14 grid points for Hall Beach, Igloolik and Kinngait, respectively. The signal was weaker for breakup dates, especially for Kinngait in which only seven of the 24 points had a statistically significant correlation with higher, but still modest, values for Igloolik (8 points) and Hall Beach (14 points). This metric is likely more dependent than other metrics on antecedent conditions from the previous year, as was found in Kowal et al. (2024a) for Hudson Strait, suggesting some short term thermal preconditioning or ‘climate memory’ at play. The coefficient of determination (R2) for Hall Beach, the highest of the three locations, varied from 11% to 73% for the ice-free season and with somewhat lower values for the freeze-up date analysis (7% to 56%). As expected, lower values were found for the breakup date analysis, ranging from 1% to 40%. These results indicate the importance of ambient temperature but also point to other factors that need consideration, such as antecedent temperatures from the previous year and the advection of sea ice, especially during the breakup. We note for the Hall Beach temperature and ice-free season length correlation that the only point that was not significantly correlated was point 16, at the location of a well-known polynya (Barber and Massom, 2007; Defossez et al., 2008; 2010).
The preconditioning argument, which was invoked to explain the breakup results, could also be active within the sea-ice cycle season. An earlier breakup date has been associated with later freeze-up dates (Stroeve et al., 2016; Kowal et al., 2024a) in other areas of the Arctic. Earlier breakup enables more absorption of more energy into the ocean waters during the ice-free season, and this thermal inertia in and of itself leads to later freeze-up dates, notwithstanding other mechanisms at play. To tease out this effect, the breakup and freeze-up time series were detrended and then the two were correlated. A negative correlation would support the hypothesis that earlier breakup dates lead to later freeze-up dates. A negative correlation proved to be the case for 19 of 24 grid points (Table 4). These correlations were weaker than those generated from the original data (not detrended), indicating that the more general warming of the region was contributing to the changes in sea-ice conditions, as well as the seasonal thermal inertia indicated by the detrended analysis which amplified the impact of regional temperature change.
Linear correlation between sea-ice breakup and freeze-up date time series and correlation between breakup and freeze-up dates for detrended time series
Grid Pointa . | Correlation . | Detrended . |
---|---|---|
1 | −0.17 | −0.09 |
2 | −0.29 | −0.26 |
3 | −0.25 | −0.06 |
4 | −0.43 | −0.28 |
5 | −0.51 | −0.36 |
6 | 0.11 | 0.17 |
7 | 0.00 | 0.02 |
8 | 0.06 | 0.07 |
9 | −0.28 | −0.15 |
10 | −0.16 | −0.21 |
11 | −0.35 | −0.25 |
12 | −0.44 | −0.20 |
13 | −0.50 | −0.45 |
14 | −0.37 | −0.35 |
15 | −0.15 | −0.10 |
16 | 0.17 | 0.20 |
17 | −0.41 | −0.30 |
18 | −0.26 | −0.28 |
19 | −0.30 | −0.21 |
20 | 0.11 | 0.13 |
21 | −0.06 | −0.05 |
22 | −0.36 | −0.32 |
23 | −0.47 | −0.47 |
24 | −0.43 | −0.34 |
Grid Pointa . | Correlation . | Detrended . |
---|---|---|
1 | −0.17 | −0.09 |
2 | −0.29 | −0.26 |
3 | −0.25 | −0.06 |
4 | −0.43 | −0.28 |
5 | −0.51 | −0.36 |
6 | 0.11 | 0.17 |
7 | 0.00 | 0.02 |
8 | 0.06 | 0.07 |
9 | −0.28 | −0.15 |
10 | −0.16 | −0.21 |
11 | −0.35 | −0.25 |
12 | −0.44 | −0.20 |
13 | −0.50 | −0.45 |
14 | −0.37 | −0.35 |
15 | −0.15 | −0.10 |
16 | 0.17 | 0.20 |
17 | −0.41 | −0.30 |
18 | −0.26 | −0.28 |
19 | −0.30 | −0.21 |
20 | 0.11 | 0.13 |
21 | −0.06 | −0.05 |
22 | −0.36 | −0.32 |
23 | −0.47 | −0.47 |
24 | −0.43 | −0.34 |
aSee Figure 2 for grid point locations.
Spatial analysis
The spatial clustering of the 24 Foxe Basin grid points was explored using two methods, Ward’s and K-means. The two methods provided only minor differences in the clustering of the grid points; only the Ward’s results are presented. In addition, because the bifurcation into two shorter time series did not result in any differences, only results for the full time period of 1971–2018 are presented.
For sea-ice breakup, the six clusters of grid points are presented in Figure 5. In addition, Figure 6 displays the clusters ordered by average breakup date to help facilitate the discussion. The clusters follow a temporal progression, although there are overlaps among them. This progression shows that the average breakup date is an important factor but not exclusively so; the inter-annual variability also plays an important role in determining the clusters.
Cluster analysis for sea-ice breakup dates in Foxe Basin. Spatial representation of the cluster analysis using Ward’s method for breakup dates for the period of 1971–2018. The six clusters generated by the cluster analysis are represented by six colours: magenta (grid points 19, 22 and 23), black (points 1–3, 5, 9 and 15), red (points 4, 8, 10, 12–14, 17, 21 and 24), green (point 6), turquoise (points 16 and 20) and blue (points 7, 11 and 18).
Cluster analysis for sea-ice breakup dates in Foxe Basin. Spatial representation of the cluster analysis using Ward’s method for breakup dates for the period of 1971–2018. The six clusters generated by the cluster analysis are represented by six colours: magenta (grid points 19, 22 and 23), black (points 1–3, 5, 9 and 15), red (points 4, 8, 10, 12–14, 17, 21 and 24), green (point 6), turquoise (points 16 and 20) and blue (points 7, 11 and 18).
Cluster analysis of the sea-ice breakup dates in Foxe Basin. Breakup clusters ordered by breakup dates for the period of 1971–2018. The x axis is a counter for the 24 points with a blank between each cluster. See Figure 5 for colour-coded locations of the grid points.
Cluster analysis of the sea-ice breakup dates in Foxe Basin. Breakup clusters ordered by breakup dates for the period of 1971–2018. The x axis is a counter for the 24 points with a blank between each cluster. See Figure 5 for colour-coded locations of the grid points.
The earliest breakup clusters are located at opposite ends of Foxe Basin: the magenta, green and black clusters (Figure 6). The Black cluster (grid points 1, 2, 3, 5, 9 and 15) is located at the southern end of the basin, as expected with warmer temperatures in the southern part of the basin (Kinngait). In contrast, and unexpected from a thermal/radiative perspective, the magenta cluster in the northwest (grid points 19, 22, 23) is located at Fury and Hecla Strait. The Strait is a source of incoming flow (and sea ice) from the rest of the Arctic, and the location of a frequently observed polynya (Barber and Massom, 2007; Defossez et al., 2008; 2010). The single green cluster also occurs where polynyas form. This association with polynyas suggests that as ice melts in these areas (green, magenta), it is not replenished from mobile ice from other parts of the basin. Mobile ice appears to be at play with the central clusters in the basin (red, blue and turquoise), particularly the blue cluster with the latest breakup dates. For these three clusters there is an east/west stratification, with turquoise and blue clusters occurring in the more dynamic side of the basin and the red cluster occurring in the eastern, more quiescent part of the basin. The flow in the basin is weakly cyclonic (counter clockwise), with the strongest flow starting with incoming waters from Fury and Hecla Strait, continuing along the western side of the Basin, and exiting partially into Hudson Bay and Hudson Strait, with some return flow heading north into the eastern part of the basin. This latter flow is quite weak, and the clustering of red and black grid points suggests a bit of local backwater that supplies mobile sea ice as the melting process takes place. As a result, there is some interplay at the boundary of these clusters (red, black, blue and turquoise). This interplay appears reflective of the less defined flow compared to Hudson Bay (Kowal et al., 2024b) and Hudson Strait (Kowal et al., 2024a). In Hudson Bay the breakup is characterized by a clearly recognizable and annually persistent sea-ice platform in the southwestern part of the bay (Gough et al., 2004; Gagnon and Gough, 2005; Kowal et al., 2024b). In Hudson Strait the strong dynamic flow of the seawater along the north shore and exiting along the south shore with an Ungava Bay backwater clearly emerges in the clustering process (Kowal et al., 2024a). Foxe Basin, while having the overall patterns noted above, does not have the more clearly defined behaviours seen in the other two water bodies. Hence, the clusters are less coherent, particular the red, blue and turquoise clusters.
Turning to the cluster analysis for freeze-up in Foxe Basin (Figure 7), a different picture emerges. In addition, Figure 7 displays the clusters ordered by average freeze-up date to help facilitate the discussion. These clusters, as with breakup clusters, follow a temporal progression, although there are overlaps among them. The overlaps are greater than observed for the breakup clusters (Figure 6). These results show that the average freeze-up date is important but that inter-annual variability plays a key role in determining the clusters.
Cluster analysis for sea-ice freeze-up dates in Foxe Basin. Spatial representation of the cluster analysis using Ward’s method for freeze-up dates for the period of 1971–2018. The six clusters generated by the cluster analysis are represented by six colours: magenta (grid points 15, 17, 19 and 21–24), black (points 1, 2, 7 and 8), red (point 3), green (points 4, 5, 9 and 12–14), turquoise (points 10, 11, 16, 18 and 20) and blue (point 6).
Cluster analysis for sea-ice freeze-up dates in Foxe Basin. Spatial representation of the cluster analysis using Ward’s method for freeze-up dates for the period of 1971–2018. The six clusters generated by the cluster analysis are represented by six colours: magenta (grid points 15, 17, 19 and 21–24), black (points 1, 2, 7 and 8), red (point 3), green (points 4, 5, 9 and 12–14), turquoise (points 10, 11, 16, 18 and 20) and blue (point 6).
The earliest freeze-up clusters are the isolated ones: red grid point 3 and blue grid point 6 (Figure 7). Both are close to shore and less impacted by the general flow of the basin, with blue point 6 influenced by a persistent polynya at Lyon’s Inlet. This result was surprising from a radiative standpoint in that the freeze-up is anticipated to begin in the northern reaches of the basin and spread south. This expectation is indeed reflected in the general behaviour of the rest of the clusters, with the progression from magenta to turquoise and then green and black with similar timing (Figure 8). However, there is significant spatial interplay at the boundaries of the clusters leading to clusters that are not spatially contiguous (points 10 and 15 are conspicuous examples). This interplay picks up on the less coherent behaviour found for breakup compared to the sister seas of Hudson Bay and Hudson Strait.
Cluster analysis of the sea-ice freeze-up dates in Foxe Basin. Freeze-up clusters ordered by freeze-up dates for the period of 1971–2018. The x axis is a counter for the 24 points with a blank between each cluster. See Figure 7 for colour-coded locations of the grid points.
Cluster analysis of the sea-ice freeze-up dates in Foxe Basin. Freeze-up clusters ordered by freeze-up dates for the period of 1971–2018. The x axis is a counter for the 24 points with a blank between each cluster. See Figure 7 for colour-coded locations of the grid points.
Finally, we examine the clustering for the ice-free season in Figure 9. In addition, Figure 10 displays the clusters ordered by the ice-free season length to help facilitate the discussion. These clusters, as with the breakup and freeze-up clusters, follow a temporal progression, particularly notable for the magenta, red and black clusters, although there are significant overlaps among them, as noted for freeze-up. Because the ice-free season length is derived from the other two metrics, this progression is not surprising. As with the other two, this analysis shows that while the average length of the ice-free season is important, the inter-annual variability plays an important role in determining the clusters.
Cluster analysis for ice-free season length in Foxe Basin. Spatial representation of the cluster analysis using Ward’s method for ice-free season for the period of 1971–2018. The six clusters generated by the cluster analysis are represented by six colours: magenta (grid points 12–15), black (points 1, 2, 5, 9, 19 and 23), red (points 3, 4 and 22), green (points 6, 8 and 16), turquoise (points 10, 17, 18, 20, 21 and 24) and blue (points 7 and 11).
Cluster analysis for ice-free season length in Foxe Basin. Spatial representation of the cluster analysis using Ward’s method for ice-free season for the period of 1971–2018. The six clusters generated by the cluster analysis are represented by six colours: magenta (grid points 12–15), black (points 1, 2, 5, 9, 19 and 23), red (points 3, 4 and 22), green (points 6, 8 and 16), turquoise (points 10, 17, 18, 20, 21 and 24) and blue (points 7 and 11).
Cluster analysis of the sea-ice season length in Foxe Basin. Ice-free season clusters ordered by ice-free season length for the period of 1971–2018. The x axis is a counter for the 24 points with a blank between each cluster. See Figure 9 for colour-coded locations of the grid points.
Cluster analysis of the sea-ice season length in Foxe Basin. Ice-free season clusters ordered by ice-free season length for the period of 1971–2018. The x axis is a counter for the 24 points with a blank between each cluster. See Figure 9 for colour-coded locations of the grid points.
The longest ice-free seasons (red and black clusters) occur at the northern entry to the basin, Fury and Hecla Strait, reflective of the same dipole that occurred for the breakup dates. This finding also reflects the local dynamics of inflowing and outflowing waters to the basin and the presence of a polynya near the strait. At the other end of the spectrum, the sizeable turquoise cluster (six grid points) has the shortest ice-free season and is situated in the less dynamic part of the basin. The remaining clusters (magenta, blue and green) form the interior clusters of the basin. While coherent as a group of clusters, the individual clusters are not particularly contiguous (especially the green cluster). This limited contiguity reflects the similar lack of coherency found in the breakup and freeze-up results and contrasts with Hudson Bay and Hudson Strait (Kowal et al., 2024a; 2024b).
In summary, a picture emerges of sea-ice behaviour in Foxe Basin that is constrained by the coasts and related circulation (cyclonic with strongest flow south along the west coast of the basin and weakest flows in the eastern part of the basin, and the related polynyas affected by coastal location and prevailing winds). The constraints leave an interior space of the basin that experiences dynamic rearrangement of sea ice during both breakup and freeze-up leading to the shortest ice-free season. The flow dynamics associated with Hecla and Fury Strait at the northwestern reach of the basin lead to the radiatively unexpected result of relatively longer ice-free seasons of similar timing to the most southerly part of the basin.
Conclusion
This paper provides a contemporary assessment of sea-ice conditions in Foxe Basin through temporal and spatial analyses and addresses two research questions. This concluding section is framed by those two questions.
How statistically significant are the temporal changes in the breakup, freeze-up and ice-free seasons for the 24 grid points representing Hudson Strait for three time periods, 1971–2018, 1971–1994, 1995–2018, and their relationships to concurrent temperature change?
These analyses, while providing more spatial nuance than previous work, are consistent with previous work in the temporal trends for sea-ice breakup and freeze-up (Gagnon and Gough, 2006; Laidler et al., 2009; Galbraith and Larouche, 2011; Hochheim and Barber, 2014; Andrews et al., 2018). For the freeze-up and ice-free season, 75% or more of the grid points had statistically significant changes. The signal was less strong for breakup, with 50% of the points having earlier breakups that were statistically significant. These signals are the opposite of the changes observed in Hudson Bay to the south of Foxe Basin (Gagnon and Gough, 2005; Hochheim and Barber, 2014), suggestive of a north/south climate-change dipole, although Kowal et al. (2017) and Andrews et al. (2018) have noted a substantial change in freeze-up in more recent years in Hudson Bay. The correlation with locally measured surface air temperatures produced results similar to those found in Hudson Strait (Kowal et al., 2024a); that is, much stronger correlations with local temperatures were found for the freeze-up dates and ice-free season length than for the breakup dates. This similarity suggests that sea-ice advection and the preconditioning of Foxe Basin by the sea-ice extent and temperatures of the previous year play roles in the breakup timing, as suggested elsewhere (Gough and Houser, 2005; Serreze and Barry, 2011; Stroeve et al., 2016; Andrews et al., 2018; Kowal et al., 2024a). We also found evidence of within-season preconditioning in a correlation analysis of detrended breakup and freeze-up time series. Earlier breakup dates for 19 of the 24 grid points correlated to later freeze-up dates, indicating the impact of greater thermal inertia generated by a longer sea-ice season that delayed the onset of the fall freeze-up.
Using a spatial cluster analysis, what is the spatial nature of breakup, freeze-up and ice-free seasons for Foxe Basin for the three time periods?
For the first time, a spatial clustering analysis was done on gridded sea-ice data for Foxe Basin. The analysis revealed behaviour that is constrained by the prevailing cyclonic flow, with stronger flows along the western extent of the basin (Barber and Massom, 2007; Defossez et al., 2008; 2010) and a more quiescent flow in the eastern part of the basin, essentially forming a backwater. This latter region was largely responsive radiatively, rather than dynamically. The flow from Hecla and Fury Strait and the presence of polynyas (generated along the western extent of the basin by prevailing winds) affected especially the ice breakup and, by extension, the ice-free season behaviour which generated unique behaviour in the northwestern part of the basin (earlier breakup and longer ice-free seasons). The shortest ice-free season occurred in the central western part of the basin, which is the result of dynamic redistribution of sea ice in both the freeze-up and breakup periods. These dynamics seem somewhat chaotic in comparison to a more coherent ice platform that occurs during the breakup of sea ice in Hudson Bay (Gagnon and Gough, 2005; Kowal et al., 2024b). This lack of coherency in Foxe Basin led to more variability among the clusters with more discontiguous cluster behaviour. This finding does not bode well for navigation purposes, specifically the less predictably coherent behaviour in Foxe Basin, in contrast to Hudson Bay and Hudson Strait.
Data accessibility statement
Temperature data were accessed from the national archive (Environment and Climate Change Canada), https://climate.weather.gc.ca/historical_data/search_historic_data_e.html (accessed on July 15, 2023); Sea ice data were obtained from the Canadian Ice Service, https://iceweb1.cis.ec.gc.ca/Archive/page1.xhtml?lang=en.
Funding
This research was funded by NSERC RGPIN-2018-06801.
Competing interests
There are no competing interests.
Author contributions
Contributed to conception and design: SK, WAG.
Contributed to acquisition of data: SK, KB.
Contributed to analysis and interpretation of data: SK, KB, WAG.
Drafted and/or revised the article: SK, WAG, KB.
References
How to cite this article: Kowal, S, Gough, WA, Butler, K. 2024. Seasonal sea ice of Foxe Basin, Canada: Spatial and temporal evolution, 1971–2018. Elementa: Science of the Anthropocene 12(1). DOI: https://doi.org/10.1525/elementa.2024.00033
Domain Editor-in-Chief: Jody W. Deming, University of Washington, Seattle, WA, USA
Associate Editor: Stephen F. Ackley, Department of Geological Sciences, University of Texas at San Antonio, San Antonio, TX, USA
Knowledge Domain: Ocean Science