Climate change and sea ice: Shipping in Hudson Bay, Hudson Strait, and Foxe Basin (1980–2016)

The marine waters of Hudson Bay, James Bay, Hudson Strait, and Foxe Basin in the eastern Canadian Arctic (the “study area”; Figure 1) are covered by sea ice for 5 to 10 months of the year (Hochheim and Barber, 2014). The seasonal duration of sea ice in the area is declining in response to climate change (Hochheim and Barber, 2014; Stroeve et al., 2014; Kowal et al., 2017). As a result, the relationships between sea ice and society are evolving, which has provided the motivation for our research. This paper consists of two parts: first, a review of the literature covering sea ice and shipping in the study area; and second, a scientific assessment of shipping accessibility in the area, based on our analysis of ice timing in offshore waters and in the local waters near the communities of Rankin Inlet, Churchill, Kuujjuarapik/Whapmagoostui, and Salluit.

Hudson Bay, James Bay, Hudson Strait, and Foxe Basin are frequently grouped for scientific discussion in the academic and gray literatures (e.g., Government of Canada, 2011; Galbraith and Larouche, 2011; Stroeve et al., 2014). There is good reason for this grouping: although each of the four regions is unique, they all share a close oceanographic relationship (see, for example, Joly et al., 2011; Hochheim and Barber, 2014). Unfortunately, there is no consensus on the term used to refer to the four marine regions collectively. Examples in the literature include the Hudson Bay system, the Hudson Bay Complex, the Greater Hudson Bay Marine Region, and others. Throughout this paper we simply refer to Hudson Bay, James Bay, Hudson Strait, and Foxe Basin as the “study area” (Figure 1).

The study area encompasses five major administrative regions within Canada: Nunavik and Eeyou Istchee of northwestern Quebec, Nunavut, Manitoba, and Ontario. The area is home to 39 formal communities dotted along the coast (Figure 1), with a total population of roughly 50,000 (Statistics Canada, 2016). Indigenous people make up the vast majority of the population in these 39 communities. The majority of residents in the communities of Nunavut and Nunavik are Inuit, and the majority of residents in the coastal communities of Eeyou Istchee and northern Ontario are Cree (Statistics Canada, 2011). Of the 39 communities, only Churchill, Manitoba, and the nine communities of James Bay are accessible by rail or road (all-weather or winter). The remaining 29 communities (and roughly 27,000 people) can be accessed only by sea or air.

Sea ice is fundamental to the culture and sustenance of the Inuit people in the study area: sea ice is a seasonal extension of their homeland that enables travel and wildlife harvesting (Inuit Circumpolar Council – Canada, 2008; Aporta, 2010). Inuit people have spent generations learning the nuances and subtleties of their local sea ice, and the seasonal ice covering forms a familiar, well-travelled topography (Laidler and Elee, 2008; Laidler et al., 2009; Aporta, 2010). Furthermore, many residents of the study area obtain a considerable portion of their food from wildlife harvested on sea ice (Chan et al., 2006; Ford, 2009; Laidler et al., 2009). On the other hand, sea ice is the greatest barrier to shipping in the study area (Kelley and Ljubicic, 2012; Engler and Pelot, 2013). For shippers it is an obstacle and a hazard that prevents access and reduces shipping potential (Aporta, 2011; Brooks and Frost, 2012; Kelley and Ljubicic, 2012). It is important to be mindful of these contrasting perceptions of sea ice.

The waters of Hudson Bay, James Bay, Hudson Strait, and Foxe Basin all undergo a complete freeze-and-melt cycle each year (Hochheim and Barber, 2014). Put simply, the study area is seasonally covered with first-year ice (FYI), though small extents of second-year ice may be produced in Foxe Basin and migrate into northeastern Hudson Bay (Gagnon and Gough, 2006; Tivy et al., 2011; Environment and Climate Change Canada, 2013). In recent decades, the seasonal timing of ice in the study area has typically run as follows. Sea ice grows from north to south between September and December and melts in a complicated pattern (though generally south to north) between May and August (Environment and Climate Change Canada, 2013; Hochheim and Barber, 2014). Foxe Basin is typically ice-free for parts of August and September, Hudson Strait is ice-free from July to November/December, and Hudson Bay and James Bay are ice-free from July/August until November/December (Environment and Climate Change Canada, 2013; Hochheim and Barber, 2014). There is, however, considerable spatiotemporal variation within these broad patterns.

The formation, persistence, and melting of sea ice are not simple processes. There is some consensus that, on a broad scale, the timing of sea ice in the study area is largely determined by atmospheric temperatures and wind (i.e., thermodynamic and dynamic “forcing”; Tivy et al., 2011; Hochheim and Barber, 2014; Ogi et al., 2016). Sea ice is influenced by both the immediate impact of external forcing and by the legacy of recent forcing, via the ice-albedo feedback loop and “climate memory” (Gough and Houser, 2005; Serreze and Barry, 2011). Climate memory refers to a circular process wherein the timing of ice at one point in the annual cycle influences the amount of heat stored in the surface waters, which in turn affects ice timing at a subsequent point in the cycle, and so on (Gough and Houser, 2005; Serreze and Barry, 2011; Stroeve et al., 2016). Numerous studies have shown relationships between the timing of breakup and freeze-up in waters of the study area (e.g., Houser and Gough, 2003; Gough and Houser, 2005; Stroeve et al., 2016; Andrews et al., 2017).

The first-year sea ice that covers the study area each year is not uniform. This first-year ice typically varies in thickness from 30 to 120 cm, with an average maximum ice thickness of 160 cm in Hudson Bay, as reported by Stewart and Lockhart (2005). The ice is influenced by local variations in air temperatures, wind, and current, and by other relatively local factors such as freshwater input and precipitation (Gagnon and Gough, 2006; Laidler and Elee, 2008; Galley et al., 2012). As a consequence, the ice of the study area grows, melts, rafts, and ridges in a variable pattern (Hochheim and Barber, 2014; Mussells et al., 2016).

The great majority of sea ice in the study area is “mobile pack ice” – floating sea ice that moves according to ocean currents and winds (Rampal et al., 2009). Mobile pack ice follows distinct movement patterns in different regions of the study area. For example, the ocean currents in Hudson Bay flow counter-clockwise and push the mobile ice with them, which can result in dynamic thickening of ice in the eastern Bay during the ice season and raised ice concentrations in the southwest and south during spring breakup (Gagnon and Gough, 2006; Galbraith and Larouche, 2011; Hochheim and Barber, 2014).

During the ice season some sections of coastal ice become “landfast”. Landfast ice is defined as any relatively immobile and continuous sheet of ice that is grounded or anchored to land in some way (Yu et al., 2014). In the study area, coastal pack ice typically becomes landfast several weeks after freeze-up. During the ice season, landfast ice may extend only a short distance or up to tens of kilometres offshore (Environment and Climate Change Canada, 2013). This landfast ice typically becomes detached from land (and no longer landfast) in the early stages of breakup.

The previous section described the “typical” timing of sea ice in the study area, but a volume of research indicates that this timing has been changing over recent decades in response to warming temperatures driven by climate change. In a pan-Arctic study using passive microwave-based sea ice data, Parkinson (2014) found significant trends of –10 to –20 days decade^–1 in the length of the ice season (ice concentrations > 15%) in our study area between 1979 and 2013. Stroeve et al. (2014) used passive microwave-based data to examine changes in the pan-Arctic timing of melt and freeze onset between 1979 and 2013 and reported significant trends for our study area of roughly –3 days decade^–1 in melt onset and roughly +6 days decade^–1 in melt season length (no significant trends for freeze onset). Tivy et al. (2011) used ice charts from the Canadian Ice Service (CIS) to examine changes in average summer sea ice extent in the Canadian Arctic between 1968 and 2008 and calculated significant trends of –8.9 ± 2.3% decade^–1 for Foxe Basin (July–September), –16.0 ± 3.4% decade^–1 for Hudson Strait (July–October), and –10.4 ± 3.1% decade^–1 for Hudson and James Bays (July–October). The authors further reported that within Hudson Bay, trends in summer ice extent were strongest in the northwest, followed by the central Bay and then the northeast, with no significant trends along the east coast (Tivy et al., 2011). Finally, Tivy et al. (2011) remarked that reductions in the ice cover of the “Hudson Bay region” (Hudson Bay and Hudson Strait in this case) were amongst the greatest in the circumpolar Arctic.

Several recent articles have examined the timing of ice specifically in our study area or regions within it. Galbraith and Larouche (2011) used CIS ice charts to examine the timing of ice breakup (below concentrations of 50%) in our study area between 1971 and 2009. Their data had a resolution of one-quarter degree (1/4°) of latitude and longitude, roughly equivalent to 28 km north/south and 10–17 km east/west in the study area. The authors reported trends of –4.9 days decade^–1 in Foxe Basin, –5.6 in Hudson Strait, and –3.2 in Hudson Bay. The declining trend in breakup date was stronger for 1991–2009 versus 1971–1990 for Foxe Basin (–9.0 versus –0.9 days decade^–1) and Hudson Strait (–13.5 versus +2.3 days decade^–1) but became non-significant for Hudson Bay (Galbraith and Larouche, 2011). On a more local scale, the authors noted that the only significant trends in Hudson Bay occurred on the western side and that the southern coast of the Bay averaged the latest breakup over the time period (Galbraith and Larouche, 2011).

Hochheim and Barber (2014) examined the timing of breakup and freeze-up in our study area between 1980 and 2010 using a 2011 version of the Comiso (2000) passive microwave-based sea ice concentration dataset, which is overlaid on a 25 km × 25 km pixel grid. The authors defined the open water threshold as the point where 50% of the pixels in an area have ice concentrations below 60% (Hochheim and Barber, 2014). The authors’ comparison of the open water seasons of 1996–2010 versus 1980–1995 provided the following results: an average growth of 3.5 weeks for Foxe Basin, with breakup 1.5 weeks earlier and freeze-up 2 weeks later; an average growth of 4.9 weeks for Hudson Strait, with breakup 2.5 weeks earlier and freeze-up 2.4 weeks later; and an average growth of 3.1 weeks for Hudson Bay, with breakup 1.5 weeks earlier and freeze-up 1.6 weeks later (Hochheim and Barber, 2014). The results of Hochheim and Barber (2014) suggest that the open water season has been lengthening most quickly in Hudson Strait, in the region at the centre of our study area, and in eastern Hudson Bay.

Kowal et al. (2017) examined sea ice timing in Hudson Bay between 1971 and 2011 using CIS ice chart data for 36 points spaced across the Bay. This research built upon work by Gagnon and Gough (2005), who established the 36 points and examined ice timing for 1971–2003. Kowal et al. (2017) used an ice concentration threshold of 50% to calculate breakup and freeze-up at each point. The authors found that between 1971 and 2011, 23 of 36 points had a significant trend towards earlier breakup and the average trend across all 36 points was –0.49 days year^–1; 34 of 36 points had a significant trend towards later freeze-up and the average trend across all points was 0.46 days year^–1; and 31 of 36 points had a significant trend towards a longer open water season and the average trend across all points was 0.91 days year^–1 (Kowal et al., 2017). Breakup trends were least significant in the eastern Bay, while freeze-up trends were fairly uniform (Kowal et al., 2017). Finally, the authors’ results indicate a strengthening in the magnitude and significance of trends for 1971–2011 versus 1971–2003, particularly in the case of freeze-up (Kowal et al., 2017).

The articles discussed above do not provide small-scale analysis of coastal ice conditions, nor are their data appropriate for that purpose. Other articles have considered near-shore ice conditions by using sea ice information from finer-resolution data or traditional knowledge. For example, Laidler et al. (2009) used both of these information sources to examine sea ice changes near Igloolik, a community in northern Foxe Basin. First, the authors used non-gridded data from the CIS ice charts to examine sea ice timing near the community (precise area unspecified): using data for 1982 to 2005 and with “open water” defined using an ice concentration threshold of 5/10, the authors found significant trends of –0.6 days year^–1 for breakup, +0.6 days year^–1 for freeze-up, and +1.19 days year^–1 for the open water season (Laidler et al., 2009). Second, the authors present community members’ observations of a considerable constriction of the sea ice season and more volatile ice conditions during fall and spring over recent decades (Laidler et al., 2009). Other articles have presented similar traditional knowledge-based observations of sea ice change for other communities in the study area, including Cape Dorset at the western end of Hudson Strait (Laidler et al., 2010) and Churchill (Ford et al., 2008).

There are relatively few articles in the scientific literature discussing the timing or trends in landfast ice in our study area. On a broader scale, research suggests that the duration of landfast ice is declining in many regions of the Arctic (Galley et al., 2012; Yu et al., 2014). However, in a pan-Arctic study of landfast sea ice based on ice charts from the U.S. National Ice Center, Yu et al. (2014) found no significant trend in winter (January–May) landfast ice area nor in the length of the landfast ice season for our study area between 1977 and 2007.

Shipping traffic in the Canadian Arctic is the product of marine re-supply, trade, fishing, tourism, and government and research activity (Engler and Pelot, 2013; Pizzolato et al., 2014). Marine re-supply (“sealift”) to communities and resource projects typically contributes the majority of traffic (Arctic Council, 2009; Étienne et al., 2013). There is very little shipping during the ice season in the Canadian Arctic, and thus traffic is confined to the months of the open water and “shoulder” (breakup/freeze-up) seasons (Étienne et al, 2013; Pizzolato et al., 2014; Mussells et al., 2016).

At present there are relatively few publicly available data for traffic volumes in the Canadian Arctic. One source of data is the vessel reports collected by the Canadian Coast Guard (CCG) for the Northern Canada Vessel Traffic Zone (or “NORDREG” zone), which encompasses all Canadian waters north of 60°N and all of Hudson, James, and Ungava Bays. Between 1990 and 2010, vessels travelling through the NORDREG zone were requested to submit position reports to the CCG; since 2010, the submission of position reports has been mandatory for vessels over 300 tonnes (Pizzolato et al., 2014; Andrews et al., 2017). Researchers from the University of Ottawa have undertaken to refine the CCG position reports into a dataset that meets academic standards (see Pizzolato et al., 2014). The CCG/University of Ottawa data indicate that roughly 140 vessels completed between 300 and 350 voyages in the NORDREG Zone each year between 2010 and 2013, up from 100–175 voyages per year between 1990 and 2006 (Dawson et al., 2016).

University of Ottawa researchers provided us with the CCG/University of Ottawa shipping traffic data for Hudson Bay, James Bay, Hudson Strait, and Foxe Basin. Although this particular data set has not been released yet, the methodology is largely explained in Pizzolato et al. (2014) and a product of the dataset has been released to the Polar Data Catalogue (Polar Data Catalogue, 2017). The CCG/University of Ottawa data for the study area suggest the following. Vessel traffic doubled from roughly 80 voyages per year for 1990–1995 to 160–180 voyages per year between 2010 and 2015. On average between 1990 and 2015, general cargo vessels were responsible for 27% of vessel traffic, bulk carriers were responsible for 23%, and tankers (likely carrying diesel fuel for re-supply) were responsible for 19%. Finally, significant increases in monthly traffic were observed for each month from June to November between 1990 and 2015.

Vessel traffic is not homogeneous within the study area. Traffic data from numerous sources (Judson, 2010; Étienne et al., 2013; Canadian Coast Guard, 2015; Dawson et al., 2016) and our conversations with shipping stakeholders indicate that most shipping traffic enters and leaves the study area via Hudson Strait. For example, sealift vessels travel from southern Quebec to the study area, make their scheduled stops, and then return to the south. These movement patterns concentrate traffic in Hudson Strait, which has the highest traffic volumes in the study area and had average traffic densities nearly twice as high as any other region in the Canadian Arctic between 1991 and 2008 (Judson, 2010). Other relatively busy waters in the study area historically have included the shipping corridor across Hudson Bay to Churchill and the coastal shipping routes within Hudson Bay, while traffic volumes are typically relatively low in Foxe Basin (Arctic Council, 2009; Judson 2010; Étienne et al, 2013; Oceans North Canada, 2016).

Vessel movement patterns in the study area provide necessary context for an analysis of shipping accessibility. Because nearly all traffic accesses the area via Hudson Strait, the shipping accessibility of a particular community or resource project will typically hinge on both the local ice conditions (local accessibility) and the ice conditions from that locality through the study area to the eastern end of Hudson Strait (offshore accessibility).

Although shipping traffic in the study area is relatively low and is confined to the ice-free and shoulder seasons, shipping nonetheless plays an important socio-economic role in the area. As previously indicated, most of the communities in the region (29 communities, roughly 27,000 people) can be accessed only by sea or air. As a result, marine re-supply (sealift) provides these communities with essential supplies that are too heavy or too costly to be flown, such as fuel, housing materials, construction supplies, vehicles, and non-perishable food items (Inuit Circumpolar Council – Canada, 2008; Arctic Council, 2009; Brooks and Frost, 2012; Andrews et al., 2016). Sealift cargo rates are highly expensive for northern communities (Brooks and Frost, 2012), but they are substantially cheaper than airlift cargo rates. In a report for the Government of Nunavut, The Mariport Group Ltd (2005) wrote that airplane freight rates in the territory can be as much as ten times more expensive than sealift. Any reduction in provisioning costs would likely be very welcome – the expense of the current provisioning system contributes to the high cost of living and high food costs that are problematic for Canada’s Arctic communities (Chan et al., 2006; Brooks and Frost, 2012; Government of Canada, 2015).

Shipping enables the two mining projects currently in production in the study area: the Raglan nickel mine that ships from Deception Bay, near Salluit, and the Meadowbank gold mine that ships from Baker Lake (Gavrilchuk and Lesage, 2014; Mussells et al., 2017). In addition to sealift and mine support, commercial shipping is important to the fishing, tourism, and research industries in the study area, and government shipping is necessary for security and emergency services (Engler and Pelot, 2013; Mussells et al., 2016). Also, prior to 2016, the Port of Churchill was an important economic contributor to the town of Churchill and provided sealift services for communities on the west coast of Hudson Bay, but the Port was put up for sale in 2015 and operations were stopped in 2016 (further discussed under “Churchill” below).

Shipping volumes have been growing in the study area and are projected to continue doing so (Judson, 2010; Étienne et al., 2013). The communities of the study area are experiencing rapid population growth, which will drive growing demand for sealift services (Engler and Pelot, 2013; Statistics Canada, 2016; Andrews et al., 2016). Numerous mining projects currently in the development or construction stages will require shipping through the study area in the coming years in order to progress. These projects include Nunavik Nickel (Ni, Cu, Pd, Pt), the Eldor project (rare earth metals), and Hopes Advance Bay (Fe, Te, V) in Nunavik; Amaruq (Au) and Meliadine (Au) in western Hudson Bay; Roche Bay (Fe) near Hall Beach in Foxe Basin; and Duncan Lake (Fe) near eastern James Bay (Gavrilchuk and Lesage, 2014; Natural Resources Canada, 2016; Agnico Eagle, 2018). A report from the Government of Canada (2015) suggests that if the sealift and resource potential of the study area were fully developed, they could yield a return of well over $10 billion dollars, far exceeding the required investment. Whether the necessary investment will occur, however, remains to be seen. Finally, the tourism industry is expected to continue growing, perhaps (but not necessarily) bringing more cruise ship and pleasure craft traffic to the study area (Stewart et al., 2010; Kelley and Ljubicic, 2012; Engler and Pelot, 2013; Lasserre and Têtu, 2015; Dawson et al., 2017).

There is relatively little industrial activity in the study area at present and shipping may be the most environmentally significant activity for the marine environment. Shipping can impact the marine environment in numerous ways. Key mechanisms include contaminant pollution (e.g., oil spills, bilge release), noise pollution, introduction of invasive species, disturbance of marine mammals, and disruption of sea ice (see Inuit Circumpolar Council – Canada, 2008; Kelley and Ljubicic, 2012; Siders et al., 2016; Zerehi, 2016; Andrews et al., 2016). The potential environmental consequences of shipping in the study area are made considerably more severe by the limited (or non-existent) shipping infrastructure, the difficulties of shipping in ice-infested waters, the relatively low quality bathymetric charts and navigational aids for the area, and the lack of monitoring, vessel support, and emergency response capacity (Kelley and Ljubicic, 2012; Commissioner of the Environment and Sustainable Development, 2014; Government of Canada, 2015; Andrews et al., 2016). These factors also affect the economic feasibility and human risk of shipping in the study area (Commissioner of the Environment and Sustainable Development, 2014; Government of Canada, 2015).

The environment of the study area is a fundamental component of the culture and well-being of its people. The Inuit people of the area are reliant on marine wildlife for a large proportion of their diet and nutrition (Priest and Usher, 2004; Poppel et al., 2007; Ford, 2009; Laidler et al., 2010; Wallace, 2014). This reliance is not simply a case of hunting for food – the marine environment, including sea ice, is an absolutely integral part of the Inuit homeland and identity (Inuit Circumpolar Council – Canada, 2008; Aporta, 2010; Laidler et al., 2010). Also, the coastal waters and estuaries of James Bay and northern Ontario are highly important to the Cree people of that area, supporting key game species (especially migratory waterfowl) and enabling traditional activities (Feit et al., 1995; Ohmagari and Berkes, 1997). The marine environment and the people of the study area should not be considered separately: any major environmental impact in the study area would also be a major social impact.

The communities of Rankin Inlet, Churchill, Kuujjuarapik/Whapmagoostui, and Salluit were selected for an analysis of local sea ice conditions. These communities were chosen in an attempt to provide a cross-section of the 39 communities in the study area with respect to sea ice and shipping. (Note: Kuujjuarapik and Whapmagoostui are neighbouring communities with some shared infrastructure. In this article they are sometimes referred to as a single community for the sake of convenience – thus “four” study communities).

Rankin Inlet, Nunavut (62°N, 92.1°W; Figure 1), is a community of roughly 2,850 people in northwestern Hudson Bay (Statistics Canada, 2016). The community is located on a point at the base of a broad inlet that opens to the east. Rankin Inlet is the administrative centre for the Kivalliq region of Nunavut. The Meliadine Gold Mine, roughly 25 km from the community, is slated to begin production in the coming years and will be shipping to and from Rankin Inlet (Agnico Eagle Ltd, 2012; Canadian Northern Economic Development Agency, 2013; Gavrilchuk and Lesage, 2014). The community currently has no permanent marine infrastructure; sealift is unloaded onto the beach via barge. Some minor shipping infrastructure upgrades are slated to accompany the Meliadine project (Agnico Eagle Ltd, 2012). The Government of Nunavut and the federal government have had discussions about developing a deep-water port at the community (Brooks and Frost, 2012; Gavrilchuk and Lesage, 2014), but there appear to be no concrete plans and no timeline for this project. With respect to sea ice, charts from Environment and Climate Change Canada (2013) indicate that the open water season near the community (approximate sea ice concentration ≤ 10%) typically ran from early July to early November between 1981 and 2010.

Churchill, Manitoba (58.7°N, 94.2°W; Figure 1), is a community of roughly 900 people located at the mouth of the Churchill River on western Hudson Bay (Statistics Canada, 2016). The Port of Churchill is the only deep-water port in the Canadian Arctic and is equipped with four loading berths capable of handling vessels of 60,000–80,000 tonnes (Andrews et al., 2017). Churchill is connected to the south via the Hudson Bay Railway. Both the Port and the Railway have been privately owned since 1997, but both assets were put up for sale in 2015 and operations at the Port were stopped in 2016 (Andrews et al., 2016; Robertson, 2017). Prior to its closure the Port typically handled exports of grain and re-supply freight, shipping grain exclusively in open water (non ice-strengthened) vessels (Andrews et al., 2016). In the decade up to 2014 the Port typically exported 400,000–500,000 tonnes of grain and roughly 10,000 tonnes of re-supply freight each year (Andrews et al., 2016). Grain was exported to international destinations in 19–20 shipments per year during a shipping season that typically ran from August to October; re-supply freight was sent to communities in western Hudson Bay (Andrews et al., 2016). According to figures from Environment and Climate Change Canada (2013), the open water season near the community typically ran from early July to early November between 1981 and 2010.

The neighbouring communities of Kuujjuarapik and Whapmagoostui (55.3°N, 77.8°W; Figure 1) are located in southeastern Hudson Bay at the mouth of the Great Whale River. Kuujjuarapik is a village in Nunavik with a population of roughly 690 people; Whapmagoostui is a village in the Cree territory of Eeyou Istchee with a population of roughly 985 people (Statistics Canada, 2016). The two communities (“Kuuj/Whap”) share the use of some infrastructure, including a relatively new marine breakwater and boat ramp (Quebec, 2015). The Government of Quebec has considered developing a deep-water port at Kuuj/Whap; the project was explicitly included in the initial Plan Nord of 2012 but was not in the revised version of 2015 (Gavrilchuk and Lesage, 2014; Quebec, 2015; Government of Canada, 2015). There has also been discussion of building a road to the community from the south, which could be used to bring exports from the planned Duncan Lake Mine (Gavrilchuk and Lesage, 2014; Quebec, 2015). The current government appetite for these projects is not clear and no detailed development plans or timelines could be found. According to figures from Environment and Climate Change Canada (2013), the open water season near the community typically ran from early July to early December between 1981 and 2010.

Salluit, Nunavik (62.2°N, 75.6°W; Figure 1), is a community of roughly 1,485 people located in Sugluk Inlet on the south side of Hudson Strait (Statistics Canada, 2016). The community has a new marine breakwater and a boat ramp (Quebec, 2015). According to figures from Environment and Climate Change Canada (2013), the open water season near the community typically ran from early July to early December between 1981 and 2010. Roughly 20 km east of Sugluk Inlet is Deception Bay; there is no community in the Bay, but there is a port used for shipments to and from the Raglan Nickel mine.

A longer open water season in the study area could be a challenge for the Inuit people who live there – declining sea ice will mean less time for hunting and travel over the ice, and ice use may become more hazardous (Ford et al., 2008). Shipping traffic appears to have doubled in the study area between 1990 and 2015, and longer open water seasons and growing socio-economic demand will provide further stimulus for this trend. There are potential benefits to increased shipping: for example, increased sealift could lead to a lower cost of living; and shipping-enabled growth in the tourism, fishing, and mining sectors could bring welcome employment and economic growth. However, these potential benefits may be offset by the environmental impacts of shipping. A careful approach is needed for development of an environmentally responsible and socially beneficial shipping industry in the study area. Scientific research will be an important component of this approach. In this context, this paper presents our efforts to characterize offshore and local shipping accessibility with respect to sea ice in Hudson Bay, James Bay, Hudson Strait, and Foxe Basin between 1980 and 2016. More specifically, our research examines sea ice timing in offshore waters and in local waters near the four communities of Rankin Inlet, Churchill, Kuujuarapik/Whapmagoostui, and Salluit by analyzing three variables: breakup, freeze-up, and the open water season.

Our examination of sea ice in the study area addresses “offshore” and “local” sea ice using different data and methodologies. For the purposes of this research, “offshore” waters include all waters more than roughly 25–71 kilometres from shore, while “local” waters consist of specific “community areas” with radii of approximately 50 km near the four study communities. We do not distinguish between different types of sea ice in the offshore analysis, but we do distinguish between landfast and pack ice in the local analysis. Our rationale and methodology are explained below.

Offshore shipping accessibility was examined using the “Sea Ice Concentrations from Nimbus-7 SMMR and DMSP SSM/I-SSMIS Passive Microwave Data” dataset (Cavalieri et al., 1996, updated yearly) hosted by the National Snow and Ice Data Centre (NSIDC). This dataset provides daily sea ice concentrations on a 25 km × 25 km grid for 1980 to 2014. The study area defined in this analysis includes 1,559 of the 25 × 25 km “pixels” (Figure 1). The area was also subdivided into three regions for further analysis: Hudson Bay (1,181 pixels), Hudson Strait (197 pixels), and Foxe Basin (181 pixels). The “Hudson Bay” region defined for this analysis includes James Bay.

As can be seen in Figure 1, pixels directly bordering the coast were not included in the analysis. They were excluded for two reasons: to minimize the influence of near-shore landfast ice or any other highly local, coastline ice in our desired offshore results; and to avoid the erroneous data that can sometimes arise in passive-microwave datasets for near-shore locations. Because no pixels bordering the coast were included in the dataset, the offshore analysis excludes coastal marine waters up to a maximum of 71 km from shore, as follows. Pixels have dimensions of 25 × 25 km (and a diagonal length of ~35.4 km), and a pixel containing any land is labeled as “coast”. Thus pixels labeled as coast in the dataset may include marine waters up to 35.4 km offshore. As a result, pixels bordering the coast may actually border a region nearly 35.4 km from shore; the exclusion of these pixels therefore results in a data-less zone of up to 71 km (<35.4 + 35.4 km) from shore.

The timing of open water conditions for each pixel from 1980–2014 was determined using the following method. For each year, breakup was recorded as the first of three consecutive days with ice concentrations ≤ 15%, freeze-up was recorded as the first of three consecutive days with ice concentrations > 15%, and the open water season was the number of days between breakup and freeze-up. The three-day window used for breakup and freeze-up insured that transient ice motion and other short-term variations in ice concentration did not affect our analysis; we experimented with longer time windows (e.g., 7 days), but they made no appreciable difference to the results. The 15% concentration threshold for open water conditions was selected as the best representation of accessible conditions for open water shipping vessels. A 15% ice concentration has been similarly applied as a shipping accessibility threshold in several publications (e.g., Bensassi et al., 2016; Andrews et al., 2017) and more generally as an open water threshold in others (e.g., Stammerjohn et al., 2008; Parkinson, 2014; Stroeve et al., 2016).

Regression analysis was used to examine trends in the timing of breakup, freeze-up, and the open water season for each pixel for 1980 to 2014. Statistical analysis suggested that the data for each variable follow a normal distribution and that no dataset exhibits significant autocorrelation. Multiple types of regression analysis were applied to the data, and linear regression yielded the best fits and the most statistically rigorous results for each variable. As a result, linear regression was used for the trend analysis. It has been suggested that sea ice timing is changing in a non-linear fashion in the study area (Kowal et al., 2017), but we did not observe this pattern in our data. Finally, linear regression was also used to examine the relationships between breakup and freeze-up dates for each pixel.

The Cavalieri et al. (1996) dataset used here and other similar passive microwave-based datasets have been used extensively for sea ice analyses similar to the research we are presenting in this paper (e.g., Stammerjohn et al., 2008; Cavalieri and Parkinson, 2012; Hochheim and Barber, 2014; Parkinson, 2014; Stroeve et al., 2016; Andrews et al., 2017). A more thorough discussion of the strengths and weaknesses of the Cavalieri et al. (1996) dataset in our context can be found in Andrews et al. (2017).

The timing of sea ice near Rankin Inlet, Churchill, Kuuj/Whap, and Salluit, was examined for 1996 to 2016 using ice charts created by the Canadian Ice Service (CIS) for the “Hudson Bay region” (Government of Canada, 2016). The CIS ice charts present the geographic distribution of sea ice within a region, with the ice grouped into “polygons” (i.e., areas) of equal ice concentration. Each polygon is labelled with information on the sea ice within it, including the ice concentration, stage of development, and form. But there is another layer of information: polygons often contain variation in the sea ice within them, and in these cases the different sub-areas of sea ice within each polygon are described. The CIS charts present ice concentrations in tenths (/10ths). Since 1982 the charts have been labelled using the ice “Egg Code” maintained by the World Meteorological Organization (Environment and Climate Change Canada, 2016).

The ice data from CIS charts have been used for numerous quantitative analyses published in the scientific literature but there are some important qualifiers and caveats. Put briefly, CIS ice charts have been created since the 1960s by ice experts using the best-available information from surface observations, aerial observations, and satellite data (both visual and microwave-based). However, “best-available” has changed since the 1960s and so has the demand for ice charts. As a result, the methodology used to create the charts and the frequency of their production has changed considerably over the decades. With respect to methodology, two points in particular are worth elaborating: the proportion of ice chart content derived from satellite observations and the proportion derived from “now-casting”. First, the proportion of ice chart data based on satellite observations grew from 15% prior to 1978, to 50–55% from 1978 to 1995, and finally to 80% in 1996 with the introduction of RADARSAT (Canadian Ice Service, 2006). Second, the CIS has often used a process called “now-casting” to produce ice charts when timely ice observations are not available. When now-casting, CIS experts estimate ice conditions at the time of the chart’s creation by extrapolating from earlier observations (Canadian Ice Service, 2006). The ice estimates based upon now-casting carry considerable uncertainty (Tivy et al., 2011). Prior to 1996, an average of 50% of the data in Arctic ice charts were the product of now-casting based on observations taken an average of 5 or more days prior to a chart’s creation (Canadian Ice Service, 2006). Since 1996, with the introduction of RADARSAT and the more timely availability of data, an average of only 20% of data in the Arctic charts has been produced through now-casting, from observations only 1 day prior to a chart’s creation (Canadian Ice Service, 2006).

Tivy et al. (2011) thoroughly examined the quality of the CIS sea ice data to assess their validity for statistical analysis. The authors discuss now-casting and technological changes and note a further important source of error: the quality of ice observations has not typically been homogeneous across all Arctic regions (or ice chart areas) at any given time. As a result there is variability in the quality of data within each ice chart, with the highest-quality data typically found near communities and shipping routes. For example, we observed that the region near Deception Bay (and the shipping route for the Raglan nickel mine) appeared to have more detailed data than surrounding areas in the ice charts for 2011 to 2016.

The CIS’s changing methodology has resulted in an archive of ice data that vary in quality both within and between charts, which affects the suitability of the data for statistical analyses. Nonetheless, after a thorough consideration of uncertainty and error, Tivy et al. (2011) ranked the CIS data for our study area as “fair” to “excellent” on their “Quality Index”, which was created to “portray the variability in data quality over space and time”. Furthermore, the CIS ice chart data has been used for numerous studies examining multi-decadal trends in ice conditions within our study area. For example, Tivy et al. (2011), Galbraith and Larouche (2011), and Kowal et al. (2017) all compare charts from the 1970s or 1980s to the present within their trend analyses. However, we deemed this methodology inappropriate for our application. The aforementioned articles applied the CIS data on a relatively large geographic scale, which may minimize some of the possible error from the sources discussed. Our research, on the other hand, seeks to examine sea ice conditions at a small, local scale. Out of caution, we used only CIS charts for 1996 to 2016 in our analysis; these charts are all within the RADARSAT era, they are largely based on satellite observations (~80%), and they contain relatively little content from now-casting (~20% on average).

Our four study communities – Rankin Inlet, Churchill, Kuuj/Whap, and Salluit – are all encompassed within the area of the CIS ice charts for the Hudson Bay region. These charts were created at a shifting frequency between 1996 and 2016. From 1980 to 2007 the charts were created monthly from January through April, every second week during May, and weekly from June through December. From 2007 to 2011 the charts were created every second week from January through March and weekly from April through December. Since 2008 the charts have been created on a weekly basis throughout the year. These general timing patterns, however, were occasionally interrupted by missing charts or longer time gaps.

The 1996–2016 CIS ice charts for the Hudson Bay region were used to examine the local timing of landfast ice and the open water season for each of the four studies communities. “Local” here refers to the marine waters within a 50-km radius of Churchill and Kuuj/Whap, and within a 50-km radius of the entrance to the Rankin and Salluit inlets (“community areas” shown in Figure 2). Landfast ice is directly labelled in the CIS data, while “open water” was defined using an ice concentration threshold of <20% (i.e., <2 tenths). The 20% concentration threshold was selected because it is the closest option to the offshore threshold of ≤15%.

Analysis was conducted by running the ice chart data through a computer program that indicated the presence/absence of landfast ice or open water conditions within each community area. Breakup for landfast ice and open water conditions were defined as the first week without landfast ice and the first week with no sea ice of concentration ≥20%, respectively. Freeze-up for landfast ice and open water conditions were defined as the first week with landfast ice and the first week with ice concentrations ≥20%, respectively. The open water season was calculated as the number of weeks between open water breakup and freeze-up (i.e., <20% concentration).

In some years multiple “break-up” events occurred in a community area as ice departed, returned, and departed again over a short time frame (e.g., 3 weeks); in these circumstances the final breakup was used for our record. Similarly, in some years multiple “freeze-up” events occurred; in these circumstances the earliest event was used for our record. As a result of this methodology our measurements reflect the latest possible breakup and earliest possible freeze-up, given the data, and therefore produce a relatively conservative estimate for open water conditions at the 20% ice concentration threshold.

Our data from the CIS ice charts were not examined for trends between 1996 and 2011. We did not feel confident that the results would be scientifically rigorous, because of the short 21-year time period and because the timing of the available CIS charts present some difficulties for analysis. The identification of ice timing (i.e., breakup or freeze-up date) using the CIS charts carries a minimum uncertainty of one week, which was the case with all breakup and freeze-up dates for the 20% concentration threshold. Furthermore, the freeze-up and breakup of landfast ice at the four communities occurred quite regularly during months where CIS charts were created only once per two weeks or once per month; in these cases, the identification of ice timing carried a two-week or one-month uncertainty. Finally, occasional ice charts were missing from the record, resulting in increased uncertainty for several measurements of landfast ice timing.

There is no consensus on whether ice concentration from CIS ice charts and passive-microwave observations may be partnered reliably for scientific analysis. Some studies have observed considerable discrepancies between ice concentrations from CIS charts versus passive-microwave methods (e.g., Agnew and Howell, 2003), though others have indicated reasonable agreement in results from the two different data sources (e.g., Tivy et al., 2011; Kowal et al., 2017; Andrews et al., 2017). In this article we do not compare results quantitatively from the two different analytical methods (local versus offshore). To compensate, the CIS charts were used to provide an element of offshore analysis: we tested each community breakup and freeze-up for the presence of a channel of open water extending continuously from the community area boundary to the eastern end of our study area as shown in Figure 1. There was no minimum width for the channel and there was no requirement of a direct route, only continuous open water conditions (ice concentration < 20%) along some channel between the community area and the eastern edge of the study area at the time of local breakup or freeze-up.

Figure 3 shows the median timing of open water conditions in the study area for 1981–1985, 1996–2000, and 2010–2014. The median values for 2010–2014 provide insight into the current timing of open water conditions in the area (Table 1). Breakup for the pixels of the study area varied between 17 May and 19 August, freeze-up varied between 22 October and 30 December, and the length of the open water season varied between 64 and 224 days (Figure 3). These wide ranges indicate the considerable regional variation in ice timing.

Hudson Strait and southeastern Hudson Bay (and James Bay) exhibited the longest open water season (>160 days) as a result of relatively early breakup and late freeze-up. Hudson Bay, with the exception of the southeast, exhibited fairly uniform open water season lengths of 140–160 days; however, that result belies considerable variation within breakup and freeze-up. Regional variation within Hudson Bay includes a relatively early breakup (early June) in the northwestern and southeastern Bay, a steadily later freeze-up from northwestern Hudson Bay (mid-November) towards the southeast (early December), and tight gradients in freeze-up date along the Bay’s western and southern shoreline. Finally, the waters of Foxe Basin exhibited the shortest open water season, a product of relatively late breakup and early freeze-up.

Figure 4 illustrates the standard deviation in ice timing in offshore waters between 1980 and 2014. For the study area as a whole, there was a median standard deviation of 14.04 days for breakup, 9.76 days for freeze-up, and 20.58 days for the length of the open water season (Table 2). These values for standard deviation provide an indication of the variability in ice timing in the study area: in general, variability is greatest in Hudson Strait, followed by Foxe Basin, and then by Hudson Bay for all three variables. Also, the timing of breakup is more variable than the timing of freeze-up throughout most of the study area.

Our results for standard deviation indicate that there is substantial variability in the study area, but the 5-year medians displayed in Figure 3 suggest that there has also been directional change between 1980 and 2014. Linear regression for the entire period revealed significant trends for breakup, freeze-up, and open water season length throughout the area, with a few exceptions (Figure 5; Table 2). Figure 5 displays the following considerable and noteworthy spatial variation in these trends.

The 1980–2014 trend magnitudes for each variable (breakup, freeze-up, open water) are consistently highest in Hudson Strait and southern Foxe Basin and are consistently lower in Hudson Bay (particularly the central Bay) than in the other two regions (Table 2). Significant freeze-up trends vary between 0.26 and 1.01 days year^–1. Significant breakup trends vary between 0.29 and 1.79 days year^–1, but only northwestern Hudson Strait and southwestern Foxe Basin exhibited trends greater than 1 day year^–1. Therefore, both breakup and freeze-up trends vary between 0.26 and 1 day year^–1 over most of the study area.

Within Hudson Bay, breakup trends increase towards the western region, which exhibits relatively high trends (0.5–1 days year^–1). Interestingly, a large portion of southeastern Hudson Bay did not exhibit significant trends for breakup. Freeze-up trends increase from the southwest to the northeast, which exhibits trends of ~0.5 days year^–1. Open water trend magnitudes are quite homogeneous, with the relatively highest trends (~1.4 days year^–1) occurring in the west.

Finally, trend analysis for all three variables show a number of pixels without a significant trend along the southern coast of Hudson Bay and in coastal James Bay. Also, southern James Bay exhibited relatively few significant trends.

Roughly 87% of pixels in the study area exhibited a significant (p < 0.05) negative relationship for freeze-up date (Y) as a function of breakup date (X) for 1980–2014, and the percentage was 97% in Foxe Basin (Figure 6; Table 3). The significant relationships exhibited a median slope of –0.36 days day^–1, a median Pearson correlation coefficient (R) of –0.47, and a median coefficient of determination (R²) of 0.22, though with considerable spatial variation. More specifically, Hudson Strait, Foxe Basin, and to a lesser extent the central Hudson Bay area exhibited relatively stronger slopes and larger R and R² values, particularly the Western Strait and southern Foxe Basin. Meanwhile, the pixels of the western and southern coasts of Hudson Bay typically exhibited weaker values for slope, R, and R² and also had a greater proportion of non-significant relationships. Noteworthy pockets of non-significant relationships also occurred in southern James Bay, eastern Hudson Bay, and southeastern Ungava Bay.

Only 54% of the pixels in Hudson Bay exhibited a significant (p < 0.05) relationship for breakup date (Y) as a function of freeze-up date for the preceding year (X). As can be seen in Figure 7, the Hudson Bay pixels with significant relationships were largely in the central bay, with few of the more coastal pixels in the south and west exhibiting significant relationships. In Foxe Basin and Hudson Strait, 65% and 90% of the pixels, respectively, exhibited a significant relationship (Table 3). For the study area as a whole, the significant relationships exhibited a median slope of –0.57 days day^–1, a median R value of –0.48, and a median R² value of 0.19, but with considerable variation between regions. Slopes and R² values were relatively high in Hudson Strait, while R values were relatively high in Foxe Basin.

With respect to open water conditions (ice concentrations < 20%), the Rankin Inlet community area exhibited a 2010–2014 median breakup of week 29 (~17–23 July), freeze-up of week 45 (~6–12 November), and open water season of 119 days (the shortest of the four study communities; Table 4). Between 1996 and 2016, breakup timing exhibited a mean of week 28.8 with standard deviation of 1.00; freeze-up timing, a mean of week 45.1 with standard deviation of 1.28; and the open water season, a mean of 16.4 weeks with standard deviation of 1.53 (Figure 8; Table 5). A continuous channel of open water ran from the edge of the community area to the eastern edge of the study area at the time of 48% of breakup events and 100% of freeze-up events (Table 6). Finally, the 1996–2016 average breakup and freeze-up dates with standard deviation for landfast ice were week 27.9 ± 1 and week 47.0 ± 1, respectively (Figure 9; Table 7).

The Churchill community area exhibited a 2010–2014 median breakup of week 27 (3–9 July), median freeze-up of week 44 (30 October–5 November, earliest of the four study communities), and median open water season of 126 days (Table 4). Between 1996 and 2016, breakup timing exhibited a mean of week 28.6 with standard deviation of 1.86; freeze-up timing, a mean of week 44.7 with standard deviation of 1.46; and the open water season, a mean of 16.1 weeks with standard deviation of 2.62 (Figure 8; Table 5). A continuous channel of open water ran from the edge of the community area to the eastern end of the study area at the time of 43% of breakup events and 100% of freeze-up events (Figure 8; Table 6). Finally, the 1996–2016 average breakup and freeze-up dates with standard deviation for landfast ice were week 26.0 ± 1 and week 49.0 ± 1.1, respectively (Figure 9; Table 7).

The Kuujuarapik/Whapmagoostui community area exhibited a 2010–2014 median breakup of week 27 (3–9 July), freeze-up of week 50 (11–17 December, the latest of the four communities), and open water season of 161 days (the longest of the four study communities by over 20 days; Table 4). Between 1996 and 2016, breakup timing exhibited a mean of week 28.1 with standard deviation of 2.32; freeze-up timing, a mean of week 50.2 with standard deviation of 1.86; and open water season, a mean of 22.1 weeks with standard deviation of 3.37 (Figure 8; Table 5). A continuous channel of open water ran from the edge of the community area to the eastern end of the study area at the time of 33% of breakup events and 14% of freeze-up events (Table 6). Finally, the 1996–2016 average breakup and freeze-up dates with standard deviations for landfast ice were week 24.3 ± 1.5 and week 52.6 ± 1.9, respectively (Table 7).

The Salluit community area exhibited a 2010–2014 median breakup of week 29 (17–23 July), freeze-up of week 48 (27 November–3 December), and open water season of 140 days (Table 4). Between 1996 and 2016, breakup timing exhibited a mean of week 29.0 with standard deviation of 1.12; freeze-up timing, a mean of week 48.1 with standard deviation of 2.31; and the open water season, a mean of 19.1 weeks with standard deviation of 2.72 (Figure 8; Table 5). A continuous channel of open water ran from the edge of the community area to the eastern end of the study area at the time of 81% of breakup events and 62% of freeze-up events (Table 6). Finally, the 1996–2016 average breakup and freeze-up dates with standard deviation were week 23.6 ± 1.5 and week 52.6 ± 1.9, respectively (Table 7).

If we assume that ice concentrations of ≤15% accurately represent navigable conditions for open-water vessels, our results clearly indicate a growth in the length of the shipping season in the offshore waters of the study area. Over the 35-year time period (1980–2014) the median trends correspond to breakup 20.24 days earlier, freeze-up 16.4 days later, and an open water season 34.0 days longer. There is, of course, considerable regional variation in these trends (Figure 5; Table 2), and the trends are superimposed upon substantial variability (Figure 4; Table 2).

Our trend results are largely in agreement with similar studies that used passive microwave-based datasets or CIS ice charts. For example, Galbraith and Larouche (2011) reported 1979–2009 trends in breakup date of –3.2 days decade^–1 for Hudson Bay, –5.6 days decade^–1 for Hudson Strait, and –4.9 days decade^–1 for Foxe Basin. These results are similar to our median trends of –3.6 days decade^–1 for Hudson Bay, –7.4 days decade^–1 for Hudson Strait, and –4.9 days decade^–1 for Foxe Basin (Table 2). Our stronger trend for Hudson Strait could be a product of trend acceleration in the region, which was reported by Galbraith and Larouche (2011). In another example, our results for Hudson Bay agree very closely with the 1971–2011 trends reported by Kowal et al. (2017). Furthermore, Galbraith and Larouche (2011), Tivy et al. (2011), and Kowal et al. (2017) describe significant changes in breakup date in central and western Hudson Bay and less significant or nonsignificant results in the east. These patterns are also shown in our results (see Figure 5). Finally, our results agree fairly well with those of Hochheim and Barber (2014), particularly with respect to the relatively rapid change in ice timing for Hudson Strait and southern Foxe Basin. However, our results do not show the rapid change in breakup and open water season length in eastern Hudson Bay reported by those authors.

Figure 3 indicates that the regional variation in ice timing trends between 1980 and 2014 has changed the spatial patterns of breakup and freeze-up in the study area. For example, Hudson Strait is now (2010–2014) exhibiting the earliest breakup and latest freeze-up in the region, whereas in 1981–1985 the ice timing in the Strait was similar to that of Hudson Bay at the same latitudes. Also, the relatively strong breakup trends in northwestern Hudson Bay have led to a new breakup pattern for the Bay (Figures 3 and 5); Tivy et al. (2011) also reported relatively strong trends in northwestern Hudson Bay.

The comparison of medians for 2010–2014 and 1996–2000 in Figure 3 reveals some surprising results: in several regions of Hudson Bay and Hudson Strait freeze-up was later in 1996–2000 than in 2010–2014 (Table 1), while in some parts of the Bay breakup was earlier, despite the significant trends for 1980 to 2014. The time series of Andrews et al. (2017) for Hudson Strait and Hudson Bay also indicate similar results, but the authors tested those time series for inflection points and non-linear behaviour and found no significant result. The late freeze-up and early breakup of 1996–2010 in the Bay and Strait appear to be variation about a linear trend, though we cannot express this view with complete certainty. Research into the conditions of ice-forcing factors, such as atmospheric temperatures and wind, in the study area for 1996–2000 might provide some insight into the seemingly anomalous ice timing.

Our results suggest a significant relationship between breakup and freeze-up over 87% of the offshore waters of the study area (Figure 6; Table 3). Based on the median slope values for each region, breakup one day earlier corresponds to a delay of 0.34 to 0.45 days in freeze-up date. Figure 6 shows considerable variation in the Pearson correlation coefficient (R) and the coefficient of determination R² values of the relationship across the study area. The variation in the R values indicates that breakup and freeze-up are more closely correlated in Hudson Strait, Foxe Basin, and central Hudson Bay (R < –0.5) than in other regions, which suggests that the timing of breakup could be used more accurately to predict the timing of freeze-up in these regions. The variation in R² values is scientifically interesting: the higher values for Hudson Strait and Foxe Basin indicate that breakup timing is responsible for a greater proportion of the variance in freeze-up timing in these regions. This indication suggests that the “climate memory” of breakup (via heat absorbed in the surface layer) has a greater role in freeze-up timing in Hudson Strait and Foxe Basin. That said, no region exhibited median R² values greater than 0.35, indicating that the majority of variance in freeze-up date in the study area is driven by factors other than breakup date.

Figure 7 shows significant relationships between freeze-up and breakup the following year over 60% of the offshore waters of the study area (Table 3). Large portions of the area did not exhibit a significant relationship, including northern Foxe Basin, western, southern, and northeastern Hudson Bay, and all of James Bay (Figure 7). In the Belcher Islands and western and southern Hudson Bay the persistent coastal polynyas (areas of open water in sea ice) and counter-clockwise currents (Joly et al., 2011; Hochheim and Barber, 2014; Andrews et al., 2016) could disrupt the relationship between breakup and freeze-up. In James Bay, polynyas and high winter and spring river input (Gough et al., 2005; St-Laurent et al., 2011; Andrews et al., 2016) could have a similar impact.

For the significant relationships that do exist between freeze-up and breakup the following year, the median slope values suggest freeze-up one day later corresponds to breakup 0.54–0.78 days earlier the following year. The observed correlation strengths (R values) suggest slightly stronger correlation (and thus predictive ability) in Foxe Basin, followed by Hudson Strait, and then Hudson Bay (Table 3). Finally, the R² values indicate that the role of “climate memory” from freeze-up timing is a relatively more important factor in breakup timing in Hudson Strait than in other regions, though median values were ≤0.35 in all regions.

Local ice timing (20% concentration threshold) was similar in Rankin Inlet and Churchill between 1996 and 2016. The two communities exhibited nearly identical averages for breakup (week 28.8 versus 28.6), freeze-up (week 45.1 versus 44.7), and open water season length (16.4 versus 16.1 weeks). Salluit averaged later breakup, later freeze-up, and a longer open water season than Churchill and Rankin Inlet between 1996 and 2016. Kuuj/Whap averaged the earliest breakup, latest freeze-up, and longest open water season.

The variation in ice timing (suggested by the standard deviation values in Table 5) was not uniform across the four study communities. Salluit and Rankin Inlet exhibited greater variation in breakup than freeze-up, while Churchill and Kuuj/Whap exhibited the opposite. Rankin Inlet exhibited the lowest variation for all variables, Kuuj/Whap exhibited the greatest variation in breakup, and Salluit exhibited the greatest variation in freeze-up and open water season length (Table 5).

The average breakup and freeze-up timing of landfast ice for 1996–2016 followed a uniform pattern across the four communities: average breakup decreased and average freeze-up increased from Rankin Inlet, to Churchill, to Salluit, to Kuuj/Whap. Also, landfast ice arrived after and broke up before the mobile pack ice in every year from 1996 to 2016 at all four communities (Figures 8 versus 9; Tables 4 versus 7).

The shipping seasons for the communities of the study area are a product of offshore and local accessibility. Both Figures 3 and 7 suggest that the Churchill and Rankin Inlet community areas commonly breakup after the nearby offshore waters and freeze-up before. This ice timing can be seen in Figure 3, in the CIS ice charts, and in the 1980–2010 “Sea ice Climatic Atlas” from Environment and Climate Change Canada (2010): during spring breakup, open water begins to appear in the offshore waters of northwestern Hudson Bay while sea ice remains intact along the coast. Also, relatively high ice concentrations often linger in the southwestern and southern Bay; this lingering is likely caused by the counter-clockwise currents of Hudson Bay moving mobile sea ice into the region during spring breakup (Hochheim and Barber, 2014). Churchill is sometimes within the western edge of the region where mobile ice becomes concentrated and persists. During freeze-up, sea ice typically grows southward in a narrow band along the west and southwestern coast of Hudson Bay, freezing the coastal waters before the nearby offshore waters. This pattern is just visible in the offshore waters shown in Figure 3, where freeze-up is earliest in the westernmost and southernmost areas of Hudson Bay.

Put briefly, our results suggest that shipping accessibility to Churchill and Rankin Inlet is limited by the timing of local sea ice (i.e., within the community areas). We find that the two community areas both have open water seasons (<20% ice concentration) of roughly 16 weeks, and that these approximate 16 weeks are often encompassed by the open water season in the nearby offshore waters and along a corridor to the eastern end of the study area. We also find that the offshore waters near Churchill and Rankin Inlet exhibited significant trends towards a longer open water season between 1980 and 2014.

The offshore waters of western Hudson Strait, near Salluit, have exhibited remarkable change in ice timing in recent decades. As can be seen in Figure 3, in 1981–1985 the waters of the western Strait broke up and froze at a similar time to waters at similar latitudes in Hudson Bay. By 2010–2014 western Hudson Strait exhibited the earliest breakup in the study area and a much later freeze-up than similar latitudes in the Bay. A qualitative examination of the CIS ice charts for 1980–2016 suggests an interesting change in patterns: the northwestern Strait has begun to break up progressively earlier and the open water often expands south and east after appearing, driving earlier breakup in that direction. Freeze-up, meanwhile, initially progressed from Foxe Basin southward into Hudson Bay and western Hudson Strait, but over the years the progression of freeze-up has begun to bypass Hudson Strait, moving from Foxe Basin to the Bay while leaving the Strait unfrozen until a later date. What has driven these spatially variable ice timing changes in the central part of the study area is not clear.

Between 1996 and 2016, the Salluit community area exhibited an average open water season of 19.1 weeks, the second longest of the four study communities. Figure 3 suggests that the community area breaks up considerably later than the nearby offshore waters but freezes up with similar timing. Furthermore, Figure 8 suggests that Salluit is typically accessible to shipping throughout the length of the open water season in its community area: local breakup was preceded by a continuous corridor of open water to the east end of our study area in each of the final ten years in the dataset (2007–2016), while local freeze-up occurred with a corridor still remaining in eight of the final ten years. The local community area for Salluit includes Deception Bay (Figure 2), thus these findings are also relevant for the port located there.

With respect to the waters near Kuuj/Whap, the CIS charts for recent years show that breakup typically progresses north along the coast from eastern James Bay and southeastern Hudson Bay, while freeze-up often occurs throughout Foxe Basin and much of Hudson Bay before reaching the coastal waters near the community. These patterns are well reflected in the 2010–2014 medians shown in Figure 3. The Kuuj/Whap community area appears to breakup considerably later than the nearby offshore waters but freezes with similar timing, resulting in a slightly shorter open water season for the community area (Figure 3). Between 1996 and 2016, the Kuuj/Whap community area exhibited an average open water season of 22.1 weeks, the longest of the study communities. These local results suggest that, with respect to sea ice, Kuuj/Whap is more accessible for sealift and could be a relatively good location for a port. However, the offshore waters may mitigate the local accessibility advantages of Kuuj/Whap: sea ice in the community area often broke up before a corridor of open water had formed through the Bay and Strait (67% of the time) and closed up after the Bay and Strait had closed to open water shipping (86% of the time; Table 6). Figure 3 suggests that shipping accessibility to Kuuj/Whap is limited by ice timing in the waters of northeastern Hudson Bay.

We did not conduct trend analysis with the 1996–2016 data for the community areas, and thus cannot comment directly on changes to local shipping accessibility. However, regardless of whether ice timing is changing in the community areas, it is conceivable that the trends in the offshore waters could lead to increased accessibility to all four study communities. With earlier breakup and later freeze-up in the offshore waters, vessels travelling to or from communities within the study area may be able to approach and depart the community areas earlier and later than historically possible.

Although the averages, multi-year medians, and trends used for our discussion of shipping accessibility thus far do not directly present the inter-annual variability or extreme events in sea ice timing, these factors have an important impact on shipping accessibility. The standard deviation in the 1996–2016 sea ice timing in the community areas provides some idea of inter-annual variability: the variability in the length of the open water season of the Kuuj/Whap community area was greatest (standard deviation of 3.37 weeks), followed by Salluit (2.72 weeks), Churchill (2.62 weeks), and finally Rankin Inlet (1.53 weeks).

Interannual variation and extreme events in sea ice timing may be particularly relevant to the shipping accessibility of Kuuj/Whap, as suggested by the relatively high standard deviation values associated with ice timing in the Kuuj/Whap community area. At irregular intervals since 1996, large volumes of sea ice have become concentrated in south and southeastern Hudson Bay (including the Kuuj/Whap community area) during spring breakup. At times, this sea ice has persisted in high concentrations into late July or August, effectively barring marine access to Kuuj/Whap. This situation happened most recently in 2015, when abnormally high ice concentrations in the southeastern Bay delayed shipping accessibility to numerous communities in the area (Government of Canada, 2016; CBC News, 2015). These extreme events may be caused by an atypically voluminous and mobile ice pack in combination with the westerly winds and counterclockwise currents of Hudson Bay. These events are difficult to predict and could present a challenge for the long-term planning of shipping operations at Kuuj/Whap.

The timing of sea ice in the study area has changed since 1980. Offshore shipping accessibility (ice concentrations ≤15%) increased between 1980 and 2014, with significant trends towards earlier breakup, later freeze-up, and longer open water seasons observed for much of the study area. These trends exhibited considerable spatial variability, resulting in a shift in the patterns of breakup and freeze-up in the study area: breakup in Hudson Bay now progresses from the southeast and the northwest; Hudson Strait now exhibits considerably longer open water seasons than similar latitudes in Hudson Bay.

Between 1996 and 2016, the Churchill and Rankin Inlet community areas both exhibited an average open water season (ice concentration < 20%) of roughly 16 weeks. Our research suggests that the nearby offshore waters were typically less ice-restricted than these community areas, and therefore we conclude that the shipping accessibility for both communities is typically limited by the timing of ice within the community areas. The Salluit community area exhibited a 1996–2016 average open water season of 19.1 weeks. We observed that the open water season for the Salluit community area is typically encompassed by the open water season in the offshore waters, as in Rankin Inlet and Churchill. Thus we conclude that shipping accessibility to Salluit is also typically limited by its local ice timing. Lastly, the Kuujjuuarapik/Whapmagoostui community area exhibited a 1996–2016 average open water season of 22.1 weeks. This period of time is the longest average open water season of our study communities, but we do note that interannual variability and irregular ice timing may be particularly problematic for shipping accessibility at Kuuj/Whap. Our results also suggest that the timing of ice in the Kuuj/Whap community area is fairly similar to ice timing in the nearby offshore waters, and that the community’s shipping accessibility may typically be limited by ice timing in northeastern Hudson Bay.

The shipping accessibility of communities and resource projects in the study area is a product of their local and regional accessibility. Most of the offshore waters of the study area exhibited growing open water seasons between 1980 and 2014, and these trends may enable earlier and later access to the 39 communities of the study area. Such access would have considerable implications for the people of the study area. For example, growing shipping accessibility could increase sealift, which could reduce the cost of goods in northern communities and facilitate resource projects and economic growth. However, the decline in sea ice represents the decline of a cultural cornerstone for the Inuit people and the loss of travelling and hunting opportunities. With both shipping accessibility and shipping traffic on the rise in the study area, the potential impact on the environment and the people of the area must be kept in mind.

We thank Dr. John Iacozza at the University of Manitoba’s Centre for Earth Observation Science for his assistance with the statistical analysis conducted for this paper.

Jonathan Andrews: The National Science and Engineering Research Council of Canada’s Canadian Graduate Scholarship – Masters ($17, 500).

The authors have no competing interests to declare.

• Contributed to conception and design: JA, DBabb, DBarber

• Contributed to acquisition of data: N/A – data downloaded from NSIDC website.

• Contributed to analysis and interpretation of data: JA, DBabb, DBarber

• Drafted and/or revised the article: JA, DBabb, DBarber