Atmospheric forcing of sea ice in Hudson Bay during the fall period, 1980–2005



[1] The principal objective of this study is to describe the autumn sea ice regime of Hudson Bay in the context of atmospheric forcing from 1980 to 2005. Both gridded Canadian Ice Service (CIS) data and Passive Microwave (PMW) data are used to examine the freezeup period for weeks of year (WOY) 43–52. Sea ice concentration (SIC) anomalies reveal statistically significant trends, ranging from −23.3% to −26.9% per decade, during WOY 43–48 using the CIS data and trends ranging from −12.7% to −16.8% per decade during WOY 45–50 using the PMW data. Surface air temperature (SAT) anomalies are highly correlated with SIC anomalies (r2 = 0.52–0.72) and with sea ice extents (r2 = 0.53–0.72). CIS data show that mean sea ice extents based on SICs ≥80% (consolidated ice) have decreased by 1.05 × 105 to 1.17 × 105 km2 for every 1°C increase in temperature in late November; PMW data show similar results. Regression analysis between SAT and standardized climate indices over the 1951–2005 period show that the East Pacific/North Pacific index is highly predictive of interannual SATs followed by the North Atlantic Oscillation and Arctic Oscillation indices. The data show that the Hudson Bay area has recently undergone a climate regime shift, in the mid 1990s, which has resulted in a significant reduction in sea ice during the freezeup period and that these changes appear to be related to atmospheric indices.

1. Introduction

[2] Over the past several decades Arctic sea ice has undergone significant changes in ice extent and concentration. In this paper we define sea ice extent (SIE) as the geographic distribution of sea ice (presence/absence) within the study region and sea ice concentration (SIC) as the percentage concentration of sea ice within a particular subset of the study area. From 1953 to 2006 the total SIE at the end of the summer melt season in September declined at a rate of −7.8% per decade [Stroeve et al., 2007]. The trends in SIC vary depending on the time period examined and the geographic location. Passive microwave (PMW) data show that trends in SIC during the 1979–1996 period were relatively small throughout the Arctic, −2.2 and −3.0% per decade, in contrast to the 1997–2007 period, which showed that declines in SIC accelerated to −10.1 and −10.7% per decade [Comiso et al., 2008].

[3] Deser and Teng [2008] showed that during the early part of the PMW period (1979–1993), ice trends in the ice marginal zones within the polar seas varied geographically. During the winter the Labrador and Bering seas had large positive trends in SIC; the Greenland and Barents seas and the Sea of Okhotsk had large negative trends. In 1993–2007 SIC trends were consistently negative throughout the Arctic and subarctic seas. Summer trends during the first half of the satellite record showed negative trends in the eastern Siberian Sea and positive trends in the Barents, Kara, and eastern Beaufort seas, in contrast to the second half of the satellite record, which was dominated by negative trends throughout the Arctic.

[4] In Hudson Bay (HB) a number of studies have examined trends in SIE. Parkinson et al. [1999] showed that during 1979–1996, only very slight negative trends were detectable within HB (including Foxe Basin): annual trends were −1.4 × 103 ± 1.4 × 103 km2/yr; autumn trends were larger, at −2.9 × 103 ± 3.6 × 103 km2/yr; and none of the seasonal trends were statistically significant. Gough et al. [2004] found no significant trends in freezeup dates for the fall period in southwestern HB (1971–2003) using Canadian Ice Service (CIS) data (Environment Canada, CIS daily analysis ice charts; available at

[5] Gagnon and Gough [2005], on the contrary, found statistically significant trends in freezeup dates using point observations. Of the 25 points used throughout HB during the freezeup period, only 6 points, located in the northern reaches of HB, showed statistically significant freezeup date trends (based on an SIC ≥50%); results indicated that freezeup was occurring 0.32–0.55 day/yr earlier (1971–2003). Kinnard et al. [2006] showed no significant trends in SICs based on CIS data from 1980 to 2004. The most recent work by Parkinson and Cavalieri [2008] showed statistically significant annual trends for SIE in HB (including Foxe Basin), with decreases of −4.5 × 103 ± 0.9 × 103 km2/yr (or −5.3% ± 1.1% per decade); fall trends were −8.5 × 103 ± 1.9 × 103 km2/yr (or −12.93% ± 2.9% per decade).

[6] Gagnon and Gough [2006] used ice thickness data from the CIS to examine trends in thickness. The data used in their study were collected from the early 1960s to the early 1990s (the data collection program was terminated in ∼1990). Temperature trends were predominantly negative and ice thickness trends were predominantly positive in HB during the fall and winter periods.

[7] The variations in SIC and SIE throughout the Arctic and sub-Arctic have been variously attributed to some combination of anthropogenic forcing due to greenhouse gases and low-frequency oscillations in atmospheric circulation and associated positive feedback mechanisms [Johannessen et al., 2004; Holland et al., 2006]. Interannual variations in SIC anomalies in the Arctic from 1960 to the mid 1990s are partly explained by variations in the Arctic Oscillation (AO) and North Atlantic Oscillation (NAO) [Venegas and Mysak, 2000; Deser, 2000; Polyakov and Johnson, 2000; Comiso et al., 2008; Deser and Teng 2008; Overland et al., 2008] and their effects on ice circulation (ice export) [Rigor et al., 2002], air temperature [Polyakov et al., 2003], and oceanic heat transport [J. Zhang et al., 2004]. In addition to the gradual warming of the Arctic over the last 50 years, Lindsay and Zhang [2005] have also suggested that the temporary phase change associated with the Pacific Decadal Oscillation (PDO) together with the AO in 1988 may have contributed significantly to the flushing of older ice out of the Arctic. More recently, warming in the high Arctic has accelerated, independent of any indices, even beyond worst-case scenarios using greenhouse gas forcing, suggesting that factors such as the sea ice-albedo feedback mechanism are contributing significantly to recent decreases in SIE [Lindsay and Zhang, 2005; Holland et al., 2006].

[8] The HB region differs from the Arctic Ocean and adjacent seas in that it is essentially a closed system and, therefore, isolated from the effects of open-ocean circulation [Wang et al., 1994] (e.g., warm-water intrusions and sea ice export) and more reflective of atmospheric forcing, specifically changes in air temperature and winds. Interannual variations in SIE in HB have been attributed largely to a number of standardized hemispheric indices that are associated with characteristic wind, temperature, and precipitation patterns. Wang et al. [1994] and Mysak et al. [1996] showed that both the NAO and the Southern Oscillation Index (SOI) were associated with peak SIEs in HB (1953–1993). Strong positive NAOs were associated with a deepened Icelandic Low, northerly winds, and lower temperatures over eastern Canada, whereas negative NAOs were associated with southerly winds and warmer temperatures. Years with strong negative SOIs during the spring/summer/fall period were associated with more ice production during the freezeup period, with the largest negative SAT anomalies occurring in August (cool summer); years with strong positive SOIs tended to have positive temperature anomalies. The largest sea ice anomalies within HB were associated with strong negative SOIs during the summer and strong positive NAOs during the winter. Prinsenberg et al. [1997], Kinnard et al. [2006], and Qian et al. [2008] all showed that NAO variability is the main factor controlling temperature variation in the winter season over eastern Canada, with positive NAO indices coinciding with early formation of sea ice in HB. In addition to the NAO, Kinnard et al. [2006] showed that the ice regime in HB was significantly correlated with the East Pacific/North Pacific oscillation (EP/NP) index during the spring (r = 0.63) and summer (r = 0.57), both being significant at the p < 0.05 level. A positive phase of the EP/NP index corresponds to a high pressure located over Alaska/western Canada and a low pressure over the central North Pacific and eastern North America. This configuration acts to draw cool Arctic air south to eastern North America including the HB region.

[9] In summary, previous work has shown that the distribution of sea ice anomalies throughout the Arctic and subarctic seas have not been uniform over the PMW satellite record (1978 to now). This observation is significant for the HB region and eastern Canada in general. Whereas much of the Arctic was warming, the HB region was actually cooling (1979–1993), hence the positive sea ice anomalies early in the PMW record [Deser and Teng, 2008], the lack of significant statistical trends in SIE from 1979 to 1996 [Parkinson et al., 1999], and the increasing sea ice thickness from 1960 to the early 1990s [Gagnon and Gough, 2006]. More recent data have shown that warming has occurred in HB since 1999–2003 [Gagnon and Gough, 2005; Ford et al., 2009; Laidler et al., 2009] and that statistically significant negative SIC trends are now evident in the Foxe Basin and HB [Parkinson and Cavalieri, 2008].

[10] This paper seeks to build on previous work as it relates to the HB region by examining both SIE and SIC and then examining the possible atmospheric forcing mechanisms linked to these sea ice metrics. In this paper we (1) provide detailed gridded representations of SAT trends of the land surrounding HB to provide a context for the observed changes in SIC and SIE; (2) show the weekly evolution of sea ice cover during the fall period from 1980 to 2005, provide gridded maps of SIC trends over 1980–2005, and provide SIC difference maps comparing the “cool period” (1980–1995) to the “warm period” (1996–2005); (3) quantify the relationship between SAT anomalies and SIC anomalies and SIE; and (4) examine the relationships between SAT anomalies and standardized atmospheric indices relevant to the fall period in HB.

2. Methods

2.1. Study Area

[11] HB is a large, shallow, inland sububarctic sea; it covers approximately 804,000 km2, and its mean depth is <150 m [Prinsenberg, 1986] (Figure 1). HB is 95%–100% ice covered during the winter months and typically ice-free during August–September. It has two openings: one to the northwest via Roes Welcome Sound and the other east of South Hampton Island into the Hudson Strait. HB is isolated from open ocean circulation, therefore variations in sea ice cover are largely a function of atmospheric forcing [Wang et al., 1994]. Currents within HB are dominantly wind driven and cyclonic at all depths, reaching a maximum in November when the winds are strongest [Prinsenberg, 1986; Saucier et al., 2004]. The circulation pattern in James Bay is also cyclonic, driven by a combination of winds and runoff dilution [Prinsenberg, 1986]. The HB basin drains an area of 3.7 × 106 km2 in North America and its freshwater discharge of ∼950 km3/yr represents 20% of the total annual runoff to the Arctic Ocean [Déry and Wood, 2004]. During the fall period SSTs are highest in the James Bay area and southeastern HB, extending north along the east coast of HB [Saucier et al., 2004]. This area is typically the last to freeze up.

Figure 1.

Study site map.

2.2. Surface Air Temperature (SAT) Data

[12] We use a SAT product known as CANGRID, developed for climate change studies by the Climate Research Division of Environment Canada. It uses adjusted historical Canadian climate data [Vincent and Gullet, 1999] that account for changes resulting from reporting station system changes. A full description of the Canada-only data set is provided by McKenney et al. [2005]. The CANGRID grid data have a spatial resolution of 50 km and cover land surfaces only.

[13] The CANGRID data used in this study consist of monthly air temperature anomalies dating back to 1950, a period when most of the stations in the region were observing on a regular basis (E. Milewska, Environment Canada, personal communication, 2009). The bounds used to compute the mean HB regional temperature anomalies (per month per year) were 50°–65°N and 72.5°–100°W (Figure 2). The use of temperature anomalies in gridding data has the advantage of removing location, physiographic, and elevation effects. Monthly temperature anomalies were computed for each month per year relative to the 1980–2005 mean to match the normals computed for sea ice data. A 3 month running mean was applied to the monthly SAT anomaly data ending in (including) the month of interest; the intent here was to incorporate lead-up SATs to obtain a (moving) seasonal temperature index (anomaly) value. We tested both normality and autocorrelation (assumptions of the general linear model) and we found each to be sufficiently low to allow for use of parametric analysis. SAT anomaly trends and their statistical significance (p; at 0.10, 0.05, and 0.01) were mapped based on the least-squares fit per grid point (n = 1128). The trend maps intend to show the regional distribution of SAT anomalies around HB.

Figure 2.

Surface air temperature (SAT) stations used to create the CANGRD data of Environment Canada. The dashed line delineates the area used to generate the regional air temperature anomaly index for Hudson Bay (HB).

[14] These temperature data were used (1) to examine general temperature trends from 1950 to 2005, (2) to establish relationships between SAT anomalies and HB-wide mean SIC anomalies and SIEs per week(s) of year (WOY; 1980–2005), and (3) to examine the relationship between SAT anomalies and atmospheric indices.

2.3. Sea Ice Data

[15] The SIC and SIE data were obtained from two sources: CIS digital ice charts (available at and PMW data processed at the National Snow and Ice Data Center [Cavalieri et al., 1996].

2.3.1. Canadian Ice Service (CIS) Data

[16] CIS ice charts are produced weekly from a variety of sources, including aerial reconnaissance data, NOAA AVHRR, RADARSAT-1, and ENVISAT ASAR. GIS information from the U.S. National Ice Center and spatial data from other national and international partners may be integrated to produce the final product. Although the CIS data go back to 1970, the charts produced since the early 1980s are of more consistent quality, owing to improvements in Earth observation technology. Data used in the study are from 1980 to 2005. For the HB area the CIS data have temporal limitations in terms of doing ice climatology work, especially during the fall period. Only WOY 43–48 have a consistent set of weekly observations for the 26 year period being examined (Table 1).

Table 1. Week of Year and Associated Datesa
  • a

    WOY, week of year.

4322–28 Oct
4429 Oct to 4 Nov
455–11 Nov
4612–18 Nov
4719–25 Nov
4826 Nov to 3 Dec
494–9 Dec
5010–16 Dec
5117–23 Dec
5224–30 Dec
011–7 Jan
028–14 Jan

[17] Each CIS data file was converted from its .e00 GIS format to a 2.5 km2 resolution grid (n = 128,656) encompassing only those areas within HB (including James Bay) that were consistently observed during the 26 year period (see Figure 1).

[18] Sea ice anomalies for each grid point per year per WOY were computed by subtracting the weekly SICs from the 26 year means. To determine trends in sea ice concentration anomalies, a least-squares linear regression was calculated for each grid point over the 26 year period, where the slope of the regression indicates the trend per year following [Parkinson et al., 1999; Galley et al., 2008; Parkinson and Cavalieri, 2008]. Data were tested for normality and autocorrelation (assumptions of the general linear model) and we found each to be sufficiently low to allow for use of parametric analysis. We thus opted for the parametric general linear model rather than a nonparametric equivalent. The statistical significance of each trend per grid point was computed and trends meeting the p = 0.1, 0.05, and 0.01 levels of significance were mapped.

[19] We noted a natural demarcation point in this time series, and as a result we also subset this time series into 1980–1995 and 1996–2005. The 1996 segmentation was chosen for two reasons: (1) the period prior to this year was representative of a relatively cooler period dominated by positive SIC anomalies and therefore provided a good contrast to the warmer period following 1995, dominated by negative sea ice anomalies; and (2) there was a significant change in technology with the introduction of RADARSAT-1 data in 1996, which allowed for improved mapping of nearshore areas owing to increased resolution and improved detectability of new and young ice. The change in technology explains the positive nearshore anomalies that appear during the relatively warmer period (1996–2005).

[20] The SIC trend maps were supplemented with SIC difference maps showing the mean differences in SIC over 1980–1995 versus 1996–2005. The statistical significance of the differences between the two time periods was assessed per grid point using a two-tailed Student's t test. Significant differences were mapped at p = 0.1, 0.05, and 0.01 probability levels for each WOY (43–48). Again, normality assumptions were tested and the parametric approach was selected over the nonparametric equivalent.

[21] Ice probability maps were also computed for SICs ≥20% and ≥80% per grid point. Each grid point per year/week was classified as meeting (1) or not meeting (0) the preceding criteria; those meeting the SIC criteria per grid point/week were summed and divided by the number of years within the observational window. The ≥20% SIC probability maps depict the leading ice edge during freezeup, while the ≥80% SIC probability maps are intended to depict “consolidated ice” [after Galley et al., 2008]. Probability maps were produced for each WOY for the entire time series (1980–2005), in part to describe the freezeup sequence. Ice probability difference maps were also generated using the ≥80% SIC data per WOY. Mean differences (and significance) in SIEs using SICs ≥80% were computed for 1980–1995 versus 1996–2005.

2.3.2. Passive Microwave (PMW) Data

[22] Because of significant gaps in the observational record of the CIS data during the freezeup period, from WOY 49 to WOY 02, SIC data from PMW data [Cavalieri et al., 1996] were used to supplement the CIS data, thus providing a second estimate of change for the full freezeup period (WOY 43 to 02) These data are provided in a polar stereographic projection and have a spatial resolution of 25 × 25 km.

[23] By use of the daily SIC data, a weekly data set was created for WOY 43–02. Sea ice anomaly maps were computed per WOY using the 26 year mean (1980–2005) as the baseline. SIC trends and significance were computed using the anomaly data as they were for the CIS data. Although the SICs computed from the PMW data are internally consistent, it is well understood that these data tend to seasonally underestimate SICs relative to the CIS data [Agnew and Howell, 2003], especially in ice marginal zones and during freezeup and melt conditions. Our use of anomalies rather than absolute concentrations minimizes this problem of underestimation, since we are in fact presenting relative (rather than absolute) change. Even with these limitations, the PMW data set is one of the best data sources available to monitor seasonal ice cover on a weekly basis, as CIS data are not always consistently available. As with the CIS data, differences in SIE were computed for WOY 46–52 based on SICs ≥60%, 1980–1995 versus 1996–2005, and their statistical significance was determined.

2.3.3. Sea Ice Thickness Data

[24] Ice thickness data have been collected in HB by the CIS (Environment Canada; available at from the late 1950s to the early 1990s, when data collection ended. Work published thus far [Gagnon and Gough, 2006] has not included the recent warming trend. Data collection in HB started again in 2002. The only station collecting ice thickness is Coral Harbour in northern HB (R. Chagnon, CIS, personal communication, 2008). Because of gaps in the data, mean ice thickness, and SATs, comparisons were made between the following time periods: 1980–1989 and 2002–2007. Statistical significance of the mean differences was computed using a two-tailed Student's t test.

2.4. Hemispheric Teleconnections

[25] Hemispheric teleconnections were examined in the context of interannual regional SATs during the fall period in HB. Various climate indices have previously been identified as potentially significant in relation to HB SATs, including the NAO, AO, SOI, EP/NP index, and PDO. Details of how each index functions are well presented in the literature and, as such, are not repeated here. Each index has an associated seasonal pressure and SAT pattern. A correlation map of each index (in its positive phase) showing its associated 500 mb geopotential heights and SATs were generated using Web tools at the National Oceanic and Atmospheric Administration Earth System Research Laboratory ( based on National Centers for Environmental Prediction/National Center for Atmospheric Research reanalysis data [Kalnay et al., 1996] for the period 1980–2005. The observed pressure and temperature patterns are discussed in relation to the HB area.

[26] The monthly standardized teleconnection data were downloaded from NOAA's National Weather Service Climate Prediction Centre ftp site (NAO, EP/NP, SOI) and from the Climate Diagnostics Center (National Oceanic and Atmospheric Administration; for the AO index and from the Joint Institute for the Study of the Atmosphere and Ocean ( for the PDO index.

[27] Since the indices fluctuate on a monthly basis, longer-term seasonal means were computed leading the month of interest. The AO and NAO means were computed based on a 4 month lead (ending in the month of interest); indices related to the Pacific region were computed based on a 5 month lead (SOI, PDO, and EP/NP). Recall that SAT anomalies used in this study were based on a 3 month moving average, so the 4–5 month leads to establish the dominant seasonal phase of an index and hence the dominant atmospheric circulation pattern are reasonable.

[28] Correlations between standardized climate indices and SAT anomalies were made interannually over several time periods, 1951–2005, 1980–2005, and the “cool” and “warm” episodes within 1980–2005. We tested the interannual data for both normality and autocorrelation (assumptions of the general linear model) and we found each to be sufficiently low to allow for use of parametric analysis. Because of the inherent variability of the indices and the varying periodicity of each of them (e.g., the AO (and NAO) operates at 2 to 3.5, 5.7 to 7.8, and 12 to 20 year scales [Venegas and Mysak, 2000; Jevrejeva et al., 2003], and the PDO index displays a periodicity at scales of 20 to 30 years [Lindsay and Zhang, 2005]), 5 year running means for both the index and SAT anomalies were also used to look at more general trends, thus complementing the interannual statistics. Although results based on the running means meet most of the assumptions of linear regression, the data are by definition autocorrelated (Table 10). We therefore caution the reader to use the statistical relationships as evidence for the underlying processes controlling these relationships rather than for hypothesis testing. Using a running mean is consistent with the 5 year running mean used by Déry and Wood [2004] and the 7 year running mean used by Polyakova et al. [2006] to assess long-term trends in indices, versus precipitation, SATs, etc.

3. Results

[29] Results are presented in the following order: (1) a review of SAT trends in the HB region from 1950 to 2005, to provide a context for the observed sea ice anomalies and trends; (2) sea ice conditions and trends in HB from 1980 to 2005 and their relationship to basin-wide SAT anomalies; and (3) correlation of longer-term fall SAT anomalies in HB with observed variations in standardized teleconnections.

3.1. Hudson Bay Air Temperature Trends

[30] The trends in SAT anomalies (Figure 3) are based on a 3 month running mean ending in the month of interest. The temperature trends throughout HB and the surrounding region are positive, indicating a warming of 0.2 to 1.8°C per decade, depending on the month and location. In general, the largest increases are on the eastern half of HB and the lowest are along the southern coast of HB between the Nelson River Estuary and James Bay.

Figure 3.

SAT anomaly trends (β) based on 3 month running means ending in (including) the month of interest. Significance (p) of trends at 0.01, 0.05, and 0.10 levels.

[31] In October temperatures are warming from 0.6 to 0.8°C/decade around the northern and eastern coasts of HB (at 95%–99% probability); lower SAT trends are evident on the western side of HB (0.4 to 0.6/decade), with trends at 0.4°C generally being nonsignificant. November trends increase to 1.0°C per decade to the north of HB and remain statistically significant (95%), while the highest trends are observed in Hudson Strait to the east (1.2°C/decade). The highest SAT anomaly trends occur in December, ranging from 1.1 to 1.4°C per decade (90%–95%) in northwestern HB to 1.2 to 1.6°C per decade in the eastern portion of the HB region (95%–99%). In January temperature trends decrease to 0.4 to 0.8°C/decade in the north and northwest (not statistically significant) and to 0.8 to 1.2°C per decade along the southeastern coast of HB including James Bay (significant at 90%–95% probability).

[32] These results show that the air temperature around HB has been warming, particularly in the northern and eastern portions. Figure 4 puts the gridded temperature trends into context relative to longer-term (1950–2005) mean SAT anomalies around HB for the months of October to December. It is evident from the graphs that (1) SAT anomalies for a given month vary significantly interannually; (2) the temperature fluctuations have a cyclical nature (smoothing spline fit λ = 0.04778; minimal smoothing); and (3) temperatures in the past have been relatively cooler, especially in the 1970s to the mid 1990s, and have warmed significantly since the mid 1990s, which is particularly evident in November and December data (stiff smoothing spline λ = 1612.676).

Figure 4.

SAT anomalies surrounding HB (1951–2005) using 3 month averages ending in (a) October, (c) November, and (e) December, with smoothing splines, i.e., (i) flexible spline fit (λ = 0.047) and (ii) stiff spline fit (λ = 1612.676), and interannual SAT anomalies trends per month for 1980–2005 (shaded line) and 1989–2005 (bold line). (b) October semidecadal mean temperature comparisons, with 1996 to 2005 identified as being statistically different. (d) November semidecadal mean temperature comparisons with 1996 to 2005 identified as being statistically different. (f) December semidecadal mean temperature comparisons, with 1996 to 2005 identified as being statistically different. Means comparison (diamonds) shows the mean (centerline) and the upper and lower 95% confidence limits, delineated by the tips of the diamonds.

[33] Comparing all semidecadal mean temperature anomalies for October (Figure 4b), only the last decade (1996–2005) is identified as being statistically different from the other periods based on both the Student t test (two tailed) and the Tukey-Kramer honestly significant difference (HSD) test. The mean temperature difference in HB for October, 1980–1995 versus 1996–2005, is 0.99°C; the mean regional temperature trend computed over 1980–2005 is 0.5°C per decade (p = 0.025); and the trend computed from the hinge point (∼1989) to 2005 is 1.1°C per decade (p = 0.0098).

[34] In November the semidecadal mean temperature anomalies (Figure 4d) identify the 1996–2005 period as being statistically different from the two preceding periods, spanning 1970–1995, with the Tukey-Kramer HSD test identifying 1996–2005 as the only statistically different period. The mean temperature difference between the latter two periods is 1.44°C. The temperature trend averaged over the HB region from 1980 to 2005 for November (Figure 4c) is 0.71°C per decade (p = 0.056), computed from the inflection point (∼1989); the temperature trend is 1.8°C per decade (p = 0.005). SAT anomalies show a slight negative trend in SAT from 1950 to 1989 (−0.12°C/decade) but the trend is nonsignificant.

[35] In December both the Student t test and the Tukey-Kramer test show 1996–2005 to be statistically different from the two preceding periods; 1996–2005 is 1.94°C warmer than 1970–1979 on average and 1.85°C warmer than 1980–1995 (Figure 4f). The regional temperature trend computed over HB for December (Figure 4e) is 1.0°C per decade from 1980 to 2005 (p = 0.024) and 2.3°C per decade from 1989 to 2005 (p = 0.008). SAT anomalies show a negative trend in SAT from 1950 to 1989 (−0.28°C/decade) but the trend is nonsignificant.

3.2. Fall Sea Ice Distribution and Trends

3.2.1. Fall Freezeup Sequence, 1980–2005

[36] The early freezeup sequence for the study period is represented by CIS data (WOY 43–48) showing mean SICs and sea ice probability maps (for SICs ≥80% and ≥20%) for 1980–2005 (Figure 5). Freezeup starts in the northern portion of HB around the shores of South Hampton Island and along the northwestern coast of HB (WOY 43). The probabilities of ≥20% ice cover are highest within the northern inlets and bays, with about a 10% probability of freezeup occurring along the coast extending down to Cape Churchill. During WOY 43 there is <30% probability of “consolidated ice” (≥80% SIC) occurring in northern HB.

Figure 5.

Fall freezeup sequence for HB based on Canadian Ice Service (CIS) data per week of year (WOY), 1980–2005, using (a) mean weekly sea ice concentrations (SICs), (b) ice probabilities based on SICs ≥80%, and (c) probabilities based on SICs ≥20%.

[37] During WOY 44, mean nearshore SICs increase and start to expand offshore from the north and northwest. Ice development begins to extend southward along the coast to the Nelson Estuary and in a narrow band along the southern coast toward James Bay. The probability of consolidated ice remains very low (10%–30%) for the most part, with higher probabilities (40%–60%) of consolidated ice in the northern coastal regions and inlets. During WOY 45 ice development progresses south and southeastward, with pronounced ice development from Cape Churchill and the Nelson River estuary to James Bay; probabilities are high (60%–100%) that the SIC along the north and northwest coasts is consolidated, and probabilities of consolidated ice remain low (≤40%) along the southern coast to James Bay. In WOY 46–48 consolidated sea ice (≥80% SIC) extends well into the HB in the north and west. During WOY 47–48 consolidated ice extends along the southern coast of HB into western James Bay, eventually encompassing Akimiski Island in WOY 48. The central portion of HB remains fairly open (SIC ≤50%).

[38] The east coast of HB starts to freeze much later (WOY 46); ice first develops along the northeastern portion of the coast and then extends southward toward James Bay in the following weeks. The probability of nearshore consolidated ice along the east coast of HB remains low in WOY 48 (40%).

[39] The remaining freezeup sequence is shown using SIC data derived from PMW data (Figure 6). For purposes of comparison, WOY 43–48 are shown again. Despite the absolute differences between the data sets, the general pattern of freezeup is consistent with the CIS data. It shows that the northern and northwest portions of HB start to freeze first, followed by the extension of ice along the south shore of HB into James Bay (WOY 46–47). The central portion of HB freezes from the north to the south and southeast, with the southeastern portion of the Bay freezing last. The PMW data show that HB is consolidated by late December to early January. Evidence of early winter latent heat polynyas in James Bay and northwestern HB, formed as a result of persistent westerly and northwesterly winds, is apparent in WOY 02.

Figure 6.

Fall freezeup sequence based on passive microwave (PMW) data using mean SICs (1980–2005) per WOY.

3.2.2. Trends in Sea Ice Concentration (SIC)

[40] Trends in sea ice anomalies were computed for WOY 43–48 using the CIS data (Figure 7) and for WOY 43–02 using the PMW data (Figure 8). Both data sets show that significant negative trends in sea ice anomalies occur throughout the fall period, indicating a decrease in SICs. Some positive trends appear along coastal regions using the CIS data. The CIS database was queried and it was found that, since 1996, nearshore new and young ice has been mapped, despite warmer air temperatures. The improved capability of detecting and mapping new and young nearshore ice since 1996 coincides with the introduction of high-resolution RADARSAT-1. Positive nearshore anomalies are therefore considered unrepresentative in the context of the historical data and are excluded from any statistical summaries, despite their appearance in CIS data.

Figure 7.

Linear trends (β) in SIC anomalies using CIS data (1980–2005) and statistical significance (p) of trends at 90%–99% probability (WOY 43–48).

Figure 8.

Linear trends (β) in SIC anomalies using PMW data (1980–2005) and statistical significance (p) of trends at 90%–99% probability (WOY 43–02).

[41] The sea ice anomaly trends per WOY follow the ice marginal zone. The trends identified as being statistically significant (90%–99% level) are summarized in Tables 2 and 3. The statistically significant trends based on the CIS data estimate reductions in SICs ranging from −23.3% to −26.9% per decade, implying mean reductions in SICs over the last 26 years of −61% to −71% (Table 2, a). Mean trends within HB, regardless of significance, for the CIS data range anywhere from −13.8% to −19.2% per decade, depending on the WOY, indicating more general reductions in SIC concentrations of −36% to −50% over broad areas of HB during the last 26 years.

Table 2. Summary of Mean Sea Ice Concentration Anomaly Trends per Decade in Hudson Bay Using Canadian Ice Service Dataa
  • a

    WOY, week of year; SIC, sea ice concentration; β, anomaly trends; HB, Hudson Bay.

Trends Based on 90%–99% Probability
β, 10 years (90%–99% prob.)−26.9−23.1−23.7−25.0−23.7−23.3
% of HB area0.111.288.5113.5215.226.18
Trends in SIC Regardless of Significance (1980–2005), Including Percent Area of HB Affected
β, all−19.2−16.4−16.6−17.0−14.3−13.8
% of HB area0.496.2622.343.9369.3276.12
Table 3. Summary of Mean Sea Ice Concentration Trends per Decade Using Passive Microwave Data Based on 90%–99% Probability, Including Percent of Hudson Bay Area Affecteda
  • a

    WOY, week of year; β, mean sea ice concentration trends; HB, Hudson Bay. Trends are from 1980–2005.

β, 10 yr (90%–99% prob.)−12.7−16.1−16.8−14.9−14.3−15.5−12.1−09.0−05.7−00.8
% of HB area9.434.052.050.357.436.841.533.414.810.0

[42] The statistically significant trends computed with the PMW data are lower, but cover a broader area, compared to the statistically significant trends of the CIS data. During WOY 45–50 the PMW data estimate SIC trends ranging from −12.7% to −16.8% per decade, indicating changes in SICs in the past 26 years ranging from −33% to −44% (Table 3). As HB sea ice consolidates late in the freezeup period (WOY 51–02), interannual variation in anomalies decrease and trends become progressively smaller, from −12.1% to −0.8% per decade.

[43] When the mean CIS anomalies (meeting 90%–99% probability) are plotted by year per WOY (Figure 9), it becomes evident that SIC anomalies from 1980 to 1995 are typically positive (20% to 60%), with a number of negative anomaly events. From 1996 to 2005 mean SIC anomalies in WOY 43–45 are exclusively negative (−20% to −40%); during WOY 46–48 almost all years have negative anomalies except for 2002 and 2004, where anomalies were slightly positive.

Figure 9.

Mean SIC anomalies per year computed from grid points with significant (p = 0.1 − 0.01) linear trends using CIS data (WOY 43–48).

3.2.3. SIC Difference Mapping CIS Data

[44] On the basis of SAT and SIC anomaly data we have identified two periods or climate regimes within the 1980 to 2005 time series. The cool period (1980–1995) shows positive anomalies and the warmer period (+0.90 to +1.94°C; 1996–2005) represents negative anomalies within the time series.

[45] Using the CIS data, change between these two periods is illustrated in two ways: (1) by a means comparison, to identify statistically significant changes (at 90%–99% levels) in mean SICs per grid point (Figure 10); and (2) by a probability difference map of SICs ≥80% (Figure 11c), to illustrate change in the probability of “consolidated ice.” Both products are functionally equivalent, with the former illustrating statistically significant changes in SIC and the latter illustrating shifts in probability.

Figure 10.

(a) SIC difference mapping using CIS data per WOY: (a) change (Δ) in mean SIC anomalies (%), 1980–1995 versus 1996–2005; (b) statistical significance (p) of change based on Student's t test.

Figure 11.

Probabilities of SICs ≥80% (a) for the “cool” period (1980–1995) and (b) for the “warm” period (1996–2005). (c) Change in probability of SICs ≥80%, 1980–1995 versus 1996–2005.

[46] Table 4 summarizes the statistically significant changes in mean SIC (%) between the two periods for each WOY. Table 4 also reports the mean sea SIE (based on ≥20% SIC) over 1980–2005 per WOY expressed as a percentage of the total HB area and lists the percentage area of HB that has undergone statistically significant change in mean SIC (% HB (ΔSIC)). The differences in mean SIC between the two periods per WOY has decreased consistently on average between −35% and −38% over each WOY within statistically significant areas, and this change has occurred over a significant portion of the mean SIE. For example, early in the season (WOY 43) the mean SIE is 0.57% of the HB area (or 4.58 × 103 km2); nevertheless, 72% of that area (3.29 × 103 km2) has shown statistically significant change, from a mean SIC of 45% to one of 8% (Table 4). Ending in WOY 48, the mean SIE is typically 92% of the HB area (or 7.39 × 105 km2); ∼42% of that area (or 3.11 × 105 km2) has undergone significant change, from a mean SIC of 69% down to one of 30%.

Table 4. Summary of Mean Sea Ice Concentration Differences Using Canadian Ice Service Data for 1980–1995 Versus 1996–2005 Within the Areas Identified as Being Statistically Different (90%–99% Level), Mean Sea Ice Extent in Hudson Bay for 1980–2005, and Percent Area of HB That Has Undergone Significant (90%–99% Probability) Change in SIC for Weeks of Year 43–48a
WOYPeriodMean SIC (%)SD (%)SIE, 1980–2005 (%)% of HB Area (ΔSIC)
  • a

    SIC, sea ice concentration; SIE, sea ice extent; HB, Hudson Bay; WOY, weeks of year. Mean SIE for HB is based on SICs ≥20%.

 Diff. (Δ)−37.36 0.570.41
 Diff. (Δ)−36.63 6.45.45
 Diff. (Δ)−38.03 23.1417.08
 Diff. (Δ)−34.94 46.1825.4
 Diff. (Δ)−36.45 76.3338.81
 Diff. (Δ)−38.48 92.1738.7

[47] A different representation of change within the HB is provided by the sea ice probability map, showing, in this case, changes in SICs ≥80% (defined hereinafter as consolidated ice) (Figure 11). The differences between the two time periods are quite dramatic for each week. In WOY 43 the probability of any “consolidated ice” has almost been eliminated, with the exception of some sheltered bays and inlets along the southern coast of South Hampton Island and the northwestern coast of HB. The same is true in WOY 44 along the southeastern portion of South Hampton Island, where in 1980–1995 a high probability of consolidated ice is reduced to a 0%–10% probability. In WOY 45 the percentage area where one would expect a high probability (60%–100%) of SIC ≥80% is reduced from 9% to 0.87% of the HB area (Δ 6.54 × 105 km2); in WOY 46 the area is reduced from 19.1% to 6.3% of the HB area (Δ 1.03 × 105 km2); in WOY 47 the area is reduced from 37.5% to 20.3% of HB (Δ 1.35 × 105 km2); and in WOY 48 the area is reduced from 75% to 38.3% of the HB area (Δ 2.95 × 105 km2). Some of the largest changes in probability of consolidated ice are evident in WOY 48 along the southern coast of HB, from the Nelson River estuary down into James Bay, and along the northeastern coast of HB, extending into the central basin. Here probabilities of consolidated ice have decreased by −50%, to >more than −70%, thus often reducing the probabilities of consolidated ice to 0%–10% during the 1996–2005 period.

[48] Table 5 summarizes the differences in SIE between the two periods based on SICs ≥80%. The mean differences in ice extents between each period are statistically significant at 95% levels except for WOY 46 (90%). In WOY 47 to 48 the extent of consolidated ice between the two periods decreased by 1.71 × 105 to 1.82 × 105 km2. On the basis of the results shown, there is at least a 1 week delay in the formation of consolidated ice.

Table 5. Summary of Mean Differences in Sea Ice Extent Based on Sea Ice Concentrations ≥80% for 1980–1995 Versus 1996–2005 Using Canadian Ice Service Data for Weeks of Year 45–48a
DataWeekYearSIE (% of HB Area)SD (%)Area (km2)p
  • a

    SIE, sea ice extent; CIS, Canadian Ice Service; WOY, weeks of year.

CIS451980–199512.8510.821.03 × 105 
  1995–20053.412.652.74 × 104 
  Diff. (Δ)−9.44 −7.59 × 1040.013
 461980–199526.6220.142.14 × 105 
  1995–200512.9212.421.04 × 105 
  Diff. (Δ)−13.69 −1.10 × 1050.066
 471980–199546.0427.773.70 × 105 
  1995–200523.3712.661.88 × 105 
  Diff. (Δ)−22.66 −1.82 × 1050.02
 481980–199567.1525.725.40 × 105 
  1995–200545.8821.043.69 × 105 
  Diff. (Δ)−21.27 −1.71 × 1050.038 PMW Data

[49] Because of temporal limitations of the CIS data, PMW data are used to document relative changes in SIE beyond WOY 48. As PMW data tend to underestimate SICs [Agnew and Howell, 2003], change detection is based on SIEs using SICs ≥60%. We start with WOY 46 to provide some overlap with the CIS data. The mean differences in SIE between the two periods for each WOY were statistically significant (at the 95%–99% level) (Table 6). For example, in WOY 46 SIE is reduced from ∼14% (1.19 × 105 km2) to 0.8% (6.2 × 103 km2) of the HB area. The maximum differences in SIE occur in WOY 47 to 50, with differences in extent ranging from −1.74 × 105 to 2.41 × 105 km2, depending on the week. In late December the relative differences in SIE between the two periods become progressively smaller, as the sea ice is typically more consolidated late in the season.

Table 6. Summary of Mean Differences in Sea Ice Extent Based on Sea Ice Concentrations ≥60% for 1980–1995 Versus 1996–2005 Using Passive Microwave Data for Weeks of Year 46–52a
WOYYearSIE (% of HB Area)SD (%)Area (km2)p
  • a

    HB, Hudson Bay; SIE, sea ice extent; WOY, weeks of year.

461980–199514.7416.661.19 × 105 
 1995–20050.770.966.20 × 103 
 Diff. (Δ)−13.97 −1.12 × 1050.015
471980–199530.3722.952.44 × 105 
 1995–20057.327.145.89 × 104 
 Diff. (Δ)−23.05 −1.85 × 1050.005
481980–199551.9123.754.17 × 105 
 1995–200525.1214.782.02 × 105 
 Diff. (Δ)−26.79 −2.15 × 1050.004
491980–199573.2520.35.89 × 105 
 1995–200543.2822.693.48 × 105 
 Diff. (Δ)−29.98 −2.41 × 1050.002
501980–199587.7715.437.06 × 105 
 1995–200566.1425.315.32 × 105 
 Diff. (Δ)−21.63 −1.74 × 1050.012
511980–199595.1810.077.65 × 105 
 1995–200579.9319.16.43 × 105 
 Diff. (Δ)−15.25 −1.23 × 1050.013
521980–199599.292.257.98 × 105 
 1995–200592.6711.037.45 × 105 
 Diff. (Δ)−6.61 −5.32 × 1040.027

3.2.4. Air Temperature Versus SIC Anomalies and SIE

[50] The results presented thus far have shown that the SAT of the land surrounding HB within the time series (1980–2005) has warmed significantly since 1995, accompanied by a significant reduction in SIC and, ultimately, SIE. Here we quantify the dependence of weekly SIC anomalies computed from CIS data (WOY 45–48) and from the PMW on interannual air temperature anomalies.

[51] The relationships between SIC anomalies computed from CIS and PMW data versus SAT anomalies are summarized in Figure 12 and Table 7. Coefficients of determination (r2) range from 0.50 to 0.60 for CIS data and from 0.54 to 0.72 for PMW data, suggesting that interannual sea ice anomalies are dependent on SAT anomalies. The data show that, over WOY 45–48, a 1°C increase in SAT results in a decrease in SICs by −14% on average using CIS data. The trends in SIC anomalies are somewhat lower using the PMW data (Table 7). In week 45 the relationship between SIC anomalies and SAT anomalies is curvilinear, because it is very early in the freezeup period so positive SIC anomalies are favored; the same occurs in week 52, where negative anomalies are favored, as ice is typically consolidating at this point. During WOY 46–51 all the relationships are linear; the highest correlations occur during weeks 47–49 (r2 = 0.62–0.72; p < 0.0001), when SIC anomalies are more evenly distributed (period of maximum interannual variation). SIC anomaly trends during WOY 47–49 range from −9.6% to −12.6%. The correlation between air temperature anomalies and SIC anomalies remains high (r2 = 0.60–0.72; p < 0.0001) during WOY 50–52, when slopes gradually decrease from −8.08 to −3.29.

Figure 12.

Relationships between SAT anomalies surrounding HB versus SIC anomalies based on CIS and PMW data.

Table 7. Regression Parameters for Sea Ice Concentration Anomalies Versus Air Temperature Anomalies for Canadian Ice Service Data for WOY 45–48 and Passive Microwave Data for WOY 45–52a
SourceWOYSlope β1β2RMSEr2p
  • a

    RMSE, root mean square error; CIS, Canadian Ice Service; WOY, week of year; PMW, passive microwave. Polynomial fits are italicized. See Figure 12.

CIS45−14.7849 20.430.52<0.0001
 46−14.6069 21.000.50<0.0001
 47−13.9777 18.890.54<0.0001
 48−13.5203 16.140.60<0.0001
PMW459.09192.721311.310.67<0.0001; 0.0166
 46−10.9469 14.530.54<0.0001
 47−12.2011 13.770.62<0.0001
 48−12.6211 11.670.71<0.0001
 49−9.6249 11.900.67<0.0001
 50−8.0852 11.710.60<0.0001
 51−5.6422 7.780.62<0.0001
 523.29330.93224.470.72<0.0001; 0.0063

[52] The degree to which SAT anomalies are predictive of interannual SIE is illustrated in Figure 13 for CIS data (WOY 47–48) and PMW data (WOY 48–49) These are periods of maximum interannual variation for each data set; regression coefficients are summarized in Table 8. For CIS data the areal extent was based on SICs ≥80%, and for PMW data it was based on SICs ≥60% to generally approximate the CIS extents. The areal extent of the ice is expressed as a percentage of the HB area. For the CIS data mean ice extents over 1980–2005 were 37.3% (or 3.00 × 105 km2) for week 47 and 58.9% (or 4.58 × 105 km2) for week 48, with slopes ranging from a −13.1% to a −14.5% (or −1.05 × 105 to −1.17 × 105 km2) decrease in areal extent per 1°C increase.

Figure 13.

Relationships between SAT anomalies surrounding HB and sea ice extent (SIE) expressed as percentage area of HB (total HB area, 804 × 103 km2) for weeks of maximum interannual variation in SIE using (a) CIS data (SIC ≥80%) and (b) PMW data (SIC ≥60%).

Table 8. Regression Parameters for Sea Ice Extent Versus Surface Air Temperature Anomalies for Weeks of Maximum Variation in SIEa
  • a

    CIS, Canadian Ice Service; PMW, passive microwave; RMSE, root mean square error; WOY, week of year; SIE, sea ice extent (% of Hudson Bay area).


[53] For PMW data mean SIEs over 1980–2005 were 41.6% (or 3.35 × 105 km2) for week 48 and 61.7% (4.96 × 105 km2) for week 49. The trends in SIE estimated from PMW using SIC ≥60% for WOY 48 and 49 were −14.42% and −12.04% (or −1.16 × 105 and −9.68 × 104 km2), respectively, for each increase in 1°C.

3.2.5. SAT and Ice Thickness

[54] Recent updates of thickness data from the CIS show that the ice thickness in Coral Harbour (the only reporting ice station on HB) has decreased during the fall period, corresponding to a period of increased SAT anomalies. Data are shown for mid November and mid December (Figure 14). In November the mean difference in ice thickness between 1980–1989 and 2002–2007 was −19.4 cm (p = 0.0458), while the mean difference in air temperature in Coral Harbour during the same period was 1.98°C (p = 0.002). In mid December the mean ice thickness decreased from 72 to 32 cm (−40.9 cm; p = 0.0012), while the mean SAT anomaly increased by 2.54°C (p = 0.0025). Changes in snow cover between the two periods were statistically insignificant for both November and December.

Figure 14.

Mean change (Δ) in sea ice thickness (cm) and SATs for Coral Harbour (1980–1989 versus 2002–2007) for (a) the month of November, which showed a mean change in ice thickness Δ of −19.4 cm, corresponding to an average 1.98°C increase in SAT; and (b) the month of December, when the ice thickness has decreased by 40 cm, with an average increase in SAT of 2.54°C. Snow thickness (not shown) showed no significant differences being the two periods.

3.3. SAT Anomalies Versus Teleconnection Indices

[55] For the fall period a number of indices were examined to determine if any were predictive of fall temperatures and ice conditions in HB. These included the NAO, AO, PDO, SOI, and EP/NP index. The geopotential height and temperature correlation maps for each index in its positive phase are presented for the late summer-early fall period for the 1980–2005 time series (Figures 15a15e) to provide a general spatial context prior to examining the HB region in more detail. Each of the indices shown, with the exception of the SOI, shows that the HB area has a tendency toward cooler air surface temperatures when the indices are in their positive phase.

Figure 15.

Seasonal correlation of indices in their positive phase with (a) 500 mb geopotential heights, using 4 month means for the North Atlantic Oscillation (NAO) and Arctic Oscillation (AO) and 5 month means for the Pacific Decadal Oscillation (PDO), Southern Oscillation Index (SOI), and East Pacific/North Pacific oscillation (EP/NP) index (ending November), and (b) mean SAT for October to November (1980–2005) (

[56] The extent to which the various indices are predictive of interannual SAT anomalies for the region surrounding HB for the periods ending in October to December are summarized in Table 9. Coefficients of determination (r2) were computed for two periods, 1951–2005 and 1980–2005, corresponding to the periods for which ice data are available (see section 2.4).

Table 9. Coefficients of Determination (r2) for Annual Mean Air Temperature Anomalies Versus Hemispheric Indices, EP/NP, NAO, AO, PDO, and SOI, Ending in October, November, and Decembera
  • a

    AO, Arctic Oscillation; EP/NP, East Pacific/North Pacific oscillation index; NA, not available; NAO, North Atlantic Oscillation; NS, not significant (90%–99% confidence interval); PDO, Pacific Decadal Oscillation; SOI, Southern Oscillation Index. A minus sign indicates a negative correlation; p identifies the significance of the relationship; bold characters = 95–99% prob.

1980–2005 0.75<0.00010.01NS0.08NS−0.120.0810.150.047
1980–2005 0.79<0.0001−0.140.0600.00NS0.210.0200.07NS
1951–2005DecNA −0.070.060−0.00NS−0.050.0890.01NS
1980–2005 NA −0.07NS−0.00NS−0.130.0660.02NS

[57] Of all the indices the EP/NP index was consistently predictive of interannual air temperatures during the fall period. During the month of October the interannual EP/NP index was predictive of SAT from 1951 to 2005 (r2 = 0.54, p < 0.0001), and more so from 1980 to 2005 (r2 = 0.75; p < 0.0001). In November (September to November) the relationships held true, with 62% of the variance in SAT surrounding HB being explained by the EP/NP index over 1951–2005 and 79% over 1980–2005. An EP/NP index was not computed for December and is therefore not shown.

[58] The NAO index was not statistically significant in October and was only weakly correlated in November. The AO was not significant at all with the exception of a weak correlation in October (1951–2005). The low correlations between the NAO and the AO very early in the season are consistent with the observation that AO and NAO tend to be strongest in the winter [Barry and Carleton, 2001]. The PDO was significant (at 90%–95%) in both November and December but the coefficients of determination were very weak (r2 = 0.05–0.21).

[59] To show more general tendencies in the indices versus SAT, 5 year running means were applied to the data. Means comparisons were made at 7 year intervals to examine the extent to which mean air temperature and index values varied over time (Figure 16). Three spline fits (λ = 0.01 (no smoothing), 6.20 (moderate), and 1612.7 (high)) were added to the temporal plots for illustrative purposes to highlight the cyclical nature of the indices and SAT anomalies, including the longer-term low-frequency variations exhibited by each of the variables from 1951 to 2005.

Figure 16.

Temporal plots of mean October to November SATs and hemispheric indices using a 5 year running mean. For illustrative purposes three spline fits (λ = 0.01, no smoothing; λ = 6.20, moderate smoothing; λ = 1612.7, stiff spline) are applied to the SATs and indices. Seven year means comparisons for (a) SAT, (b) EP/NP index, (c) NAO, (d) AO, (e) PDO, and (f) SOI show that the latter period (1999–2005), without exception, is statistically different from the preceding period. Means comparison (diamonds) shows the mean (centerline) and the upper and lower 95% confidence limits, delineated by the tips of the diamonds.

[60] The means over the 1999–2005 period (Figure 16), with few exceptions, were statistically different from those in the two preceding intervals (1985–1991, 1992–1998). Also, all indices changed phase in the mid 1990s and showed trends in index values that are typically associated with warmer fall temperatures for the HB region.

[61] In terms of air temperature, the 1999 to 2005 period is statistically warmer (mean SAT anomaly 0.83°) compared to all preceding periods. The first two periods encompassing 1957–1970 are significantly warmer compared to 1992–98 (Δ = 0.36°C) and significantly cooler (Δ = −0.63°C) than the 1999–2005 anomalies.

[62] The temporal plot of the EP/NP index shows that it was positive from 1965–97 with occasional reversals (1970–71, 1983–85, 1991–92), and consistently negative from 1998–2005. Based on the means comparisons the 1999–2005 period is statistically different relative to all preceding periods. Table 10 summarizes the extent to which the various indices exhibit covariance to the mean air temperatures in HB computed for November, the period of maximum sea ice variability. The EP/NP index is shown to be highly predictive of SAT surrounding Hudson Bay back to 1951 (r2 = 0.75) and from 1980 to 2005 period (r2 = 0.89).

Table 10. Matrix of Coefficients of Determination (r2) for Climate Indices Versus Surface Air Temperature Based on the 5 Year Running Mean for 1980–2005, 1950–2005, 1980–1995, and 1995–2005a
  • a

    AO, Arctic Oscillation; AT, air temperature; EP/NP, East Pacific/North Pacific oscillation index; NAO, North Atlantic Oscillation; PDO, Pacific Decadal Oscillation; SOI, Southern Oscillation. A dash indicates a negative correlation. P indicates a second-order polynomial. Boldface italic correlations are significant at 99% level, boldface correlations are significant at 95% level, italic correlations are significant at 90% level, and regular text correlations are nonsignificant.

1950–2005 (n = 51)
1980–2005 (n = 26)
1980–1995 (n = 16)
1995–2005 (n = 11)

[63] Although the NAO index shows considerable variation, it has largely remained positive from 1973 to 1996 (Figure 16c), with a few reversals (1977, 1983–1984, 1988–1989); a positive NAO is associated with cooler temperatures in HB. During the 1999–2005 interval the mean NAO index became strongly negative and is statistically different from that in all preceding periods with the exception of 1964–1970. The most recent trend favors warmer fall temperatures. The NAO index is correlated with regional SAT anomalies from 1951 to 2005 (r2 = 0.39) and from 1980 to 2005 (r2 = 0.65) (Table 10). For 1951–2005 the EP/NP and NAO indices together explained 80% of the variance in SAT, and for 1980–2005 the EP/NP and NAO indices together explained 94% of the variation.

[64] The AO index values have been predominantly negative from 1955 to 1972 and positive from 1973 to 1996, with one reversal from 1980 to 1985 (Figure 16d). From 1997 to 2005 the AO index has been positive. The low-frequency trend shown by λ = 1612.7 indicates that the AO has a long-term periodicity (complete cycle not shown) overlain with shorter-term fluctuations (≤15–20 years). The AO indices are now trending to negative values favoring warmer temperatures in HB. The most recent period (1999–2005) has been consistently negative and statistically different from the two preceding periods, which are positive and associated with cooler November temperatures in HB. The AO index has a weak correlation with SAT from 1951 to 2005 (r2 = 0.14); the correlation improves over the 1980–2005 period (r2 = 0.50). The EN/NP and AO together explained ∼90% of the variance in SAT anomalies based on a 5 year running mean. The PDO index typically has a ≥20–30 year cycle. Through the 1980s and into the late 1990s the fall PDO was positive and is now trending to a negative cycle (Figure 16e). The last period (1999–2005) is statistically different from the three preceding periods (over 1978–1998), which were in the positive phase of the cycle. The negative phase of this index is associated with warmer fall temperatures in HB. Over 1950–2005 the PDO index is not strongly correlated with SAT (r2 = 0.20), although slightly better than the AO. During the 1980–2005 the PDO is more highly correlated (r2 = 0.70).

[65] The SOI is quite variable and generally the most poorly correlated index (interannually) with SAT anomalies in HB (Table 10). Despite that, it is interesting to note that the very low-frequency pattern (λ = 1612.7) appears to be the inverse of the low-frequency decadal pattern exhibited by the other indices, with a notable regime shift around 1976–1977 [Y. Zhang et al., 1997] (Figure 16f). Negative SOIs are loosely associated with cooler fall temperatures in HB and extreme SIE events in HB when in phase with a strong positive NAO [e.g., Wang et al., 1994].

[66] Table 10 also lists correlations within the “cool” and “warm” phases of the standardized atmospheric indices in the 1980–2005 time series. The EP/NP index is the most highly correlated within the 1980–1995 period (r2 = 0.74), followed by the NAO and AO indices, at r2 = 0.38 and 0.36, respectively, with the PDO and SOI not showing any significance. Together, the EP/NP index with either the NAO or the AO explains ∼84% of the variance in SAT anomalies in HB. During the warming phase all indices have changed phase, indicative of warmer fall temperatures for HB. All indices are highly correlated (r2 = 0.50–0.97).

[67] We suggest caution in implying causal relationships to all of the various indices and the observed SAT anomaly trend. What can be stated is that, since 1995, the various indices have changed phase and that the EP/NP, NAO, and AO indices appear to be those most consistently correlated with SAT anomalies over all periods, with the EP/NP index being the single most predictive index during the fall period and the NAO and AO contributing significantly in terms of improving the explained variance in SAT anomalies when using multiple regression.

4. Conclusions

[68] Based on the CANGRID data we have shown that SAT anomaly trends were positive (warming) around HB from 1980 to 2005. The highest and most significant trends occurred in the northern and eastern portions of HB, with overall trends in SAT anomalies increasing from October (0.6–0.8°C/decade) to December (1.1–1.6°C/decade). Although statistically nonsignificant, the regional mean interannual SAT anomalies show a slight cooling period over HB from 1950 to 1989, most evident in November and December, followed by a statistically significant increase in SAT during the mid 1990s to 2005.

[69] Both CIS data and PMW data showed that SIC anomalies were decreasing throughout the fall (WOY 43–01), with the most significant (negative) trends in SIC anomalies following the marginal ice zone. The statistically significant trends in SIC anomalies using the CIS data showed negative trends in SIC ranging from −23.3% to −26.9% per decade for weeks 43–48, resulting in significant reductions in SIE over the last 26 years. Statistically significant trends in SIC anomalies using the PMW data were lower but were more broadly distributed throughout HB, ranging from −14.3% to −16.8% per decade for weeks 46–50.

[70] Interannual SIE was closely related to variations in SAT evidenced by both CIS data and PMW data. The CIS data showed that for every 1°C increase in the mean regional air temperature around HB, the area of SIC ≥80% (consolidated ice) deceased by 1.05 × 105 to 1.17 × 105 km2 for weeks 47–48 (late November). Similar results were shown for changes in SIEs using PMW data based on a slightly lower SIC threshold (SICs ≥60%).

[71] Regional SAT anomalies around HB were shown to be closely related to atmospheric indices. The EP/NP index was predictive of SAT anomalies in HB dating back to 1950. The NAO and AO were much less predictive; they typically exert their strongest influence during the winter period. Five year running means were also applied to the SAT and to the teleconnections data. These data showed that the EP/NP index together with the NAO and AO explained ∼80%–90% of the variance with SAT anomalies in November from 1951 to 2005. The SOI index was consistently the most poorly correlated with SATs (R2 = 0.08) on an interannual basis, whereas the PDO was more predictive of SATs than the AO index over 1951–2005.

[72] Examining the longer-term trends in air temperature and the hemispheric indices using a 5 year running mean, it is apparent that the climate has been undergoing a regime shift in the last 15 years and that this shift in HB during the fall appears to be associated with the low-frequency oscillation pattern inherent in the various indices, particularly the EP/NP, NAO, and AO. The phase change in the mid 1990s coincides with warmer SATs in HB and associated negative SIC anomalies and SIEs. We plan to extend this HB work by examining the winter-to-summer period for these same relationships in a follow-up paper.


[73] This work was funded by the Natural Sciences and Engineering Research Council, Canada Research Chairs program, and ArcticNet Networks of Centers of Excellence program with grants to D.G.B. Thanks go to R. Galley for gridding and extracting the CIS data and to the anonymous reviewers and editors of Journal of Geophysical Research—Oceans for improving the clarity of this presentation.