1. Introduction

Rivers provide a natural pathway for freshwater on the land surface and constitute a vital link between the atmosphere, the land surface, and the oceans. Streams and rivers integrate spatially and temporally atmospheric and land surface processes at the catchment scale, providing a mechanism by which climate change may be detected. One area where significant climate change is ongoing is the Arctic (Serreze et al. 2000). Modifications to the high-latitude environment, driven by rising surface air temperatures (Chapman and Walsh 1993), include a decreasing depth and duration of the snowcover (Brown and Braaten 1998; Curtis et al. 1998), a warming and thawing of the permafrost (Osterkamp and Romanovsky 1996; Stieglitz et al. 2003), and increasing storminess and precipitation (McCabe et al. 2001; Walsh 2000). Such changes in the high-latitude environment all affect freshwater discharge in complex ways; however, given that river runoff is driven mainly by precipitation and surface air temperature, recent observations suggest an acceleration of the hydrological cycle in many northern regions, including increasing freshwater discharge (Peterson et al. 2002; Ziegler et al. 2003).

Historical discharge from the Canadian Mackenzie River Basin and from the three largest Siberian drainage basins (Yenisey, Lena, and Ob River Basins) are well sampled and documented (e.g., Stewart et al. 1998; Lammers et al. 2001; Peterson et al. 2002; Yang et al. 2002). By comparison, the Hudson, James, and Ungava Bays (HJUBs) drainage basin has received little attention. Unlike its major Russian and Canadian counterparts, the HJUBs basin consist of a large number of rivers with relatively lower runoff rates. For instance, the Yenisey, Lena, Ob, and Mackenzie Rivers supply no less than 50% of the annual freshwater discharge to the Arctic Ocean. The average annual runoff rates for these four large rivers attains ∼500 km³ yr⁻¹, whereas the greatest annual discharge rate recorded for a single river flowing into HJUBs reaches only 70 km³ yr⁻¹ for the Nelson River (Carmack 2000). Yet the HJUBs drainage system alone equates 18% of the total discharge to the Arctic Ocean, more so than any other large river basin in Eurasia and North America (Shiklomanov et al. 2000). Variability and change in the discharge of freshwater into HJUBs has implications for the transport of nutrients, sediments, and other trace species to high-latitude oceans (Anderson et al. 2004), sea ice formation (Manak and Mysak 1989; Weatherly and Walsh 1996; Saucier and Dionne 1998; Saucier et al. 2004), and the thermohaline circulation (Aagaard and Carmack 1989).

Given the importance of the HJUBs drainage network in the global freshwater budget and the recent decline in the number of monitoring flow gauges in Canada (Shiklomanov et al. 2002), there is a critical need to report historical discharge rates of Canadian rivers with outlets into HJUBs. Furthermore, general circulation models forecast that the Hudson Bay (HB) region will experience some of the more dramatic environmental changes in the coming century (Gough and Wolfe 2001; Gagnon and Gough 2005). The main objective of this paper is to document the characteristics of freshwater discharge rates of 42 rivers that flow into HJUBs and to analyze the data for interannual variability and trends. The study period covers 37 yr (1964–2000) when relatively good spatial and temporal coverage of river discharge data exists.

The paper is organized as follows: Section 2 provides background information on the HJUBs drainage basin, and section 3 introduces the dataset and methods used in this study. Section 4 provides the analysis of our results, which are then compared in section 5 to other studies found in the literature. Section 6 discusses the implications of freshwater discharge into HJUBs on upper-ocean salinity values in the Labrador Current. The paper closes with a summary (section 7) of our major findings and plans for future work.

2. Study area

The HJUBs cover an area of approximately 3.7 × 10⁶ km², or more than one-third of Canada (Fig. 1). The basin collects freshwater over 20° of latitude and over 50° of longitude. The Canadian provinces of Alberta (AB), Saskatchewan (SK), Manitoba (MB), Ontario (ON), and Québec (QC), as well as Nunavut (NU) Territory, all contribute substantial freshwater to the HJUBs watershed. The American states of Montana, North Dakota, South Dakota, and Minnesota are also a source of freshwater for the HJUBs. Forty-two of the most important rivers that drain into HJUBs are listed in Table 1. These rivers are identified by one of the three bays into which they discharge, as well as the province or territory where their outlet in HJUBs is situated.

The mean annual air temperature of the HJUBs varies considerably from north to south, ranging from 4°C in the Canadian Prairies to −12°C in Nunavut (Table 2). The total mean annual precipitation exhibits latitudinal and longitudinal gradients. Average minima of ∼300 mm are found in both extreme northern and southern sections of the drainage basin, whereas average maxima of ∼800 mm are found at intermediate locations in the boreal forest. Annual snowfall amounts range from about 100-mm snow water equivalent (SWE) on average in the Canadian Prairies to 400-mm SWE on average in northern Québec. Onset of the snowcover begins on average by early October in the northern sections of the basin and progresses southward until on average mid-November when all of the HJUBs basin is snow-covered (McKay and Gray 1981). Snowmelt begins in the Canadian Prairies on average by mid-April but does not start until mid-June on average in northern Québec and Nunavut. Continuous and discontinuous permafrost is common at latitudes above 51°N (Woo 1986). Vegetation progresses from prairie grasslands in the headwaters of the Nelson River to the boreal forest across most of HJUBs, and then to the Arctic tundra in Nunavut and northern Québec.

3. Dataset and methods

Freshwater discharge rates for rivers that drain into HJUBs are compiled in Environment Canada’s Hydrometric Database (HYDAT; Environment Canada 1994, 2004b). HYDAT is a comprehensive observational database that provides daily discharge rates (in m³ s⁻¹) for most of the main rivers that flow into HJUBs. At least partial data exist for the 42 rivers listed in Table 1. Additional information on the maximum area over which gauge data are available for each river basin and the geographical coordinates of these recording stations are found in Table 1. The total maximum area that is monitored equals 80% of the overall territory covered by the HJUBs drainage basin. Temporal coverage between 1964 and 2000 is good for the largest rivers; however, smaller streams and rivers are generally not well sampled. Therefore, the following steps are adopted to provide complete temporal coverage of freshwater discharge for the rivers of interest.

Flow rates recorded by the gauge nearest to the rivers’ outlet into HJUBs are first used as the most representative for the entire basin. In situations where there are missing data, measurements from an upstream recording gauge along the river are used if available. In these instances, the river runoff rates are adjusted to reflect the missing contributing area to total discharge such that

View Expanded

where R (m³ s⁻¹) denotes the runoff rate over an area A (m²) and where the subscripts u and d identify the upstream and downstream locations. Table 3 provides a comparison of daily river discharge for the Petite Rivière de la Baleine as measured by a gauge at 55.73°N, 74.69°W and the daily reconstructed river discharge for the same location based on gauge measurements recorded concurrently upstream at 55.68°N, 74.33°W. The error analysis in Table 3 demonstrates that the reconstructed runoff rates match the observations well for the period 1974–89.

In some instances, discharge rates are completely unavailable for certain rivers throughout the period 1964–2000. To fill these gaps in the dataset, an additional procedure is followed. The annual cycle of daily discharge rates for each river is first constructed based on all of the available data for the period 1964–2000. Time series of daily discharge rates are then compiled from all of the available gauge measurements. During the construction of these time series, however, flow rates may be missing for a given river. In that case, discharge rates are estimated from the mean daily value for that river over 1964–2000. However, the streamflow rates for the missing site are adjusted according to the deviation in discharge from the mean of all rivers for which data are available on that specific day. For example, let us assume that the mean runoff for three separate rivers on a given day is , and , respectively. In constructing the time series of river discharge for that day (elements R₁, R₂, and R₃), a missing value may be encountered for station 3. In that instance, the river discharge for that site is estimated from

View Expanded

The accuracy of this methodology over a period of 15 yr is demonstrated in Table 3 for the Petite Rivière de la Baleine (55.73°N, 74.69°W). All of the available discharge data for the Petite Rivière de la Baleine were eliminated from the database and then reconstructed using the above method. Although the reconstructed discharge data are not as accurate as those based on an upstream gauge, they provide reasonable data that ensure continuous time series in each of the 42 rivers of interest.

In the presentation of our results, we choose not to remove the anthropogenic impact on river discharge. Indeed, seven of the major rivers that drain into HJUBs are impacted by dams, diversions, and/or reservoirs (Table 4). Major dams exist on the La Grande, Moose, and Nelson Rivers, whereas parts of the Koksoak, Churchill, Eastmain, and Opinaca Rivers have been partially diverted to feed other rivers for enhanced hydroelectric power generation (Prinsenberg 1980; Messier et al. 1986). Although dams were constructed prior to 1964 on the Nelson and Moose Rivers, other rivers did not incur similar anthropogenic influences until about 1980. At this time, the Eastmain, Opinaca, and Koksoak Rivers were all partly diverted from their natural courses to create large reservoirs that now feed La Grande Rivière where the massive James Bay (JB) hydroelectric complex was constructed by Hydro-Québec (Messier et al. 1986). Part of the Churchill River was diverted in 1976/77 to enhance the discharge of the Nelson River where Manitoba Hydro operates four large power-generating plants (Prinsenberg 1980).

Discharge data near the mouths of La Grande Rivière, the Opinaca River, and the Eastmain River are unavailable after 1980 in the HYDAT dataset. Therefore, the streamflow computed for these three rivers is based on measurements taken upstream from dams and/or river diversions. Discharge for the Opinaca and Eastmain Rivers after 1980 is therefore overestimated since this water is nearly all diverted into La Grande Rivière system (Messier et al. 1986). Nevertheless, we expect the total estimates for the three rivers to correspond well with the artificially enhanced flow of La Grande Rivière. To test this hypothesis, we obtained a dataset of monthly discharge rates valid for the mouth of La Grande Rivière carried out by Hydro-Québec (R. Roy 2003, personal communication). Figure 2 presents a comparison of the monthly discharge rates for this system based on the Hydro-Québec data that are compared to the sum of the monthly discharge rates for the La Grande, Opinaca, and Eastmain Rivers inferred from the HYDAT data over 1981–94. It shows a good correspondence (r² = 0.87) between the two datasets, with HYDAT yielding only 2% more runoff than the Hydro-Québec dataset. This suggests that the total flow of La Grande Rivière, the Opinaca River, and the Eastmain River as determined from the HYDAT dataset remains accurate even after 1980. In this study, we maintain separate values for the flow of the La Grande, Eastmain, and Opinaca Rivers with the understanding that after 1980, their total is representative of the artificially enhanced discharge in La Grande Rivière. However, care is required in interpreting trends in these three rivers after 1980.

We choose not to remove the anthropogenic effects on HJUBs river discharge since we are interested in examining the actual amounts of freshwater entering high-latitude oceans. The investigation in section 6 of trends in upper-ocean salinity measurements from the inner Newfoundland Shelf requires observed HJUBs river discharge that includes anthropogenic effects. Nonetheless, it is important to distinguish natural trends from those imposed by human activity. Thus where possible, we compare trends in observed river discharge between naturally flowing and artificially influenced systems.

For the investigation of trends in freshwater discharge into HJUBs, we define a normalized runoff anomaly (NRA) as (Myers et al. 1990)

View Expanded

where the subscript i denotes the runoff for an individual year, the overbar denotes the annual average value for the period 1964–2000, and σ_R denotes the standard deviation of R. Positive (negative) values of the NRA indicate above-average (below average) discharge rates for a given river. This quantity allows direct comparisons of streamflow anomalies despite their wide range in absolute values. The method by which trends and their significance are established is given in the appendix.

4. Results

c. Role of snowcover

Snow covers the surface of the HJUBs from 130 to 240 days each year (McKay and Gray 1981). As a result of its accumulation during the cold season, spring runoff is dominated by meltwater that yields an annual peak in total discharge within the HJUBs. Table 6 includes the mean peak flow rates observed each spring and the mean date at which this event occurs. The largest streamflow rates attain 10 275 m³ s⁻¹ on average each spring transition period in the Koksoak, demonstrating the critical role played by snow in the hydrology of high-latitude watersheds. The earliest peak discharge rates occur during the first week of May in several James Bay rivers, whereas the latest peak discharge rates are observed in the Povungnituk (on average in early July). For the system as a whole, maximum flow rates of nearly 0.05 Sv day⁻¹ occur in mid-June.

Figure 9 shows that there is a good correspondence (r² = 0.71) between the latitude of the outlets for each of the 42 rivers and the day at which runoff attains its mean annual peak. The linear regression based on these data yields

View Expanded

where ϕ is latitude (degrees north) and where JD represents the Julian day of the observed peak in river runoff. The slope in Eq. (4) suggests that there is a difference of 5 days in the occurrence of peak discharge rates associated with snowmelt for each degree of latitude. Thus snowmelt induces peak discharge rates on average in early May at 50°N but not until late June at 60°N. Owing to its wide range of latitudes, the HJUBs undergo snowmelt for a period of at least two months (mostly in May and June) of each year.

Figure 10a represents the mean departure (in days) from the mean annual date at which the spring peak discharge rates are observed in the HJUBs. Although there is significant variability in the timing of the freshet (±16 days), there is an overall trend toward an earlier snowmelt. The trend analysis shown in Fig. 10 reveals that the onset of melt has advanced on average by 8 days between 1964 and 2000. A similar trend analysis excluding the seven rivers affected by dams, diversions, and reservoirs yields an advance of only 4 days in the timing of spring peak discharge rates. Figure 10b illustrates that the intensity of the observed peak discharge rates during spring has diminished by 420 m³ s⁻¹ over the 37 yr of interest. Similar to the date of snowmelt, there is considerable interannual variability in the flow rates recorded during spring, with anomalies ranging from −400 to 700 m³ s⁻¹. Despite the decreasing trend in this quantity, the largest positive anomalies occurred during the second half of the study period (i.e., 1979, 1985, 1986, and 1992). For the 35 rivers not affected by human impacts, the observed spring peak discharge rates have decreased on average by 302 m³ s⁻¹. As illustrated in Fig. 8, the regulation of water has a tendency to diminish the intensity and alter the timing of the spring freshets (Anctil and Couture 1994; McClelland et al. 2004).

5. Comparison with other studies

A comparison of the mean annual discharge rates obtained in this study with the values compiled by the Natural Resources Canada (2004) in the Atlas of Canada for 22 of the largest rivers that drain into HJUBs is shown in Fig. 11. There is good agreement (r² = 0.85) between the two datasets. Discrepancies may arise at selected time periods (time periods are not specified in the Atlas of Canada). To compare the total annual discharge rates into Hudson Bay and Hudson Strait, we adjust our results to include the missing 20% in contributing area to our calculations. The adjusted total discharge rate of 888 km³ yr⁻¹ for the HJUBs drainage system compares well with the 938 km³ yr⁻¹ reported by Shiklomanov and Shiklomanov (2003) for the period 1966–99.

Trends in Canadian streamflow have recently been assessed by Zhang et al. (2001). The authors employed river runoff data from 243 stations that are part of the Reference Hydrometric Basin Network (RHBN; Harvey et al. 1999) across Canada with temporal coverage beginning in 1947. For the period 1967–96, the authors find a significant upward trend in the discharge into Chesterfield Inlet whereas other rivers in northern Ontario and Québec show declining streamflow rates. This pattern is consistent with the results of Whitfield and Cannon (2000), who used cluster analysis to inspect trends in streamflow between 1976 and 1995 for 650 river basins of Canada. Furthermore, Ingram et al. (1996) reported a diminishing trend in the runoff of the Grande Rivière de la Baleine in northern Québec between 1962 and 1991. These studies are in general agreement with the trend analysis carried here for the 42 rivers draining into HJUBs over the period 1964–2000. Peterson et al. (2002) report a 7% increase in Siberian river discharge between 1936 and 1999. This equates to an additional input of 128 km³ yr⁻¹ of freshwater to the Arctic Ocean by the end of their study period. In Hudson Bay and Hudson Strait, an opposite trend is observed, with a decline of 96 km³ yr⁻¹ or of 13% in freshwater discharge by 2000. Further analysis of the normalized runoff anomalies in HJUBs between 1964 and 1989 reveals a decrease of 148 km³ yr⁻¹ (significant at the p < 0.01 level) that balances the overall change observed in Siberia.

6. Discussion

River discharge into high-latitude oceans affects the salinity and the stability of the upper ocean, and hence sea ice formation and the global thermohaline circulation (Aagaard and Carmack 1989; Manak and Mysak 1989). Sutcliffe et al. (1983) proposed a link between freshwater discharge into Ungava Bay and upper-ocean salinity anomalies observed four months later on the Newfoundland Shelf in the northwestern Atlantic Ocean. Myers et al. (1990) investigated this link in greater detail and established a negative correlation between enhanced Hudson Bay and James Bay river discharge and the upper-ocean salinity observed nine months later at station 27, an oceanographic station located at 47.5°N, 52.6°W on the inner Newfoundland Shelf.

To explore further the findings of Sutcliffe et al. (1983) and Myers et al. (1990), a dataset of ocean salinity covering the period 1964–94 has been acquired from the Canadian Marine Environmental Data Services (MEDS). Salinity measurements taken within the upper layer (<50 m) of the ocean for a 25 km² grid box that encompasses station 27 are extracted from the MEDS archive. An algorithm that follows the approximate path taken by HJUBs freshwater into Hudson Bay, Hudson Strait, and then out into the Labrador Current is also developed. We compute the mean distance between the outlet of each river to points along the dominant wind-driven currents in the area. In Hudson Bay, the surface currents form a cyclonic gyre with a mean current speed of ∼0.04 m s⁻¹ (Prinsenberg 1980, 1986a). Arriving in Hudson Strait, the water accelerates along the Québec coastline at a speed of ∼0.15 m s⁻¹ (Prinsenberg 1986b; Drinkwater 1986) and then deflects southward into the Labrador Current where the water travels along the Labrador and Newfoundland shelves at about 0.5 m s⁻¹ (Reynaud et al. 1995; LeBlond et al. 1981, 1996). The tracking algorithm follows the spring/summer pulse of freshwater that arises from continental snowmelt with realistic contributions and delays for each of the 42 rivers that drain into HJUBs. This methodology serves as a useful diagnostic tool to estimate the timing and intensity of downstream freshwater pulses and salinity anomalies in the top layers of the ocean.

Figure 12 provides the contributions of four selected regions along the perimeter of HJUBs to the pulse of freshwater inferred for station 27 between 1970 and 1972. The time period is restricted here to 3 yr to improve clarity of the results. It shows a phasing of the peaks in river runoff from three sections along the perimeter of the drainage system: Ungava Bay, James Bay, and Western Hudson Bay. Discharge from Ungava Bay exhibits the strongest seasonal cycle and leads to the sudden peaks in the freshwater pulses. The seasonal cycle of river discharge into James Bay and Western Hudson Bay is not as pronounced but still displays a significant peak arising from snowmelt that is in phase (with a lag of 1 and 2 yr, respectively) with that of Ungava Bay. Table 7 lists the 42 rivers of interest in order along the perimeter of Western Hudson Bay, James Bay, Eastern Hudson Bay, and Ungava Bay; the approximate number of days that water from the river outlets takes to reach station 27 on the inner Newfoundland Shelf; and the distance traveled over that time. The table also includes the Julian day at which a crest in river discharge induced by snowmelt reaches this ocean location. Of interest is the phasing of the contributions from Western Hudson Bay (with more than a 2-yr delay), James Bay (with more than a 1-yr delay), and Ungava Bay (with a 3-month delay) to the freshwater pulse arriving at station 27 during summer. As many as 22 rivers with outlets into HJUBs see their meltwater peaks reach station 27 on average between 1 July and 31 September.

Figure 13 depicts the upper-ocean salinity observed at station 27 between 1966 and 1994 in addition to the pulse of freshwater that arises from river runoff into HJUBs as computed by the tracking algorithm for this location. Freshwater pulses for 1964 and 1965 are omitted owing to the 2-yr delay required for some of the water to reach station 27. There is a strong annual peak in the pulse of freshwater that resembles high discharge rates associated with snowmelt. Peak transport values typically reach 0.05 Sv each year during summer; the peaks correspond well with the annual minima in upper-ocean salinity observed at station 27. This supports the results of Sutcliffe et al. (1983) and Myers et al. (1990) that suggest a negative correlation between continental runoff into HJUBs and ocean salinity of the inner Newfoundland Shelf. However, the tracking algorithm suggests that the freshwater pulses are composed of meltwater from three successive winter seasons rather than a single season as proposed by Sutcliffe et al. (1983) and Myers et al. (1990).

Figure 14 presents the magnitude of the annual freshwater pulse between 1 July and 31 September and the corresponding upper-ocean salinity values. Accompanying the diminishing discharge rates into HJUBs between 1966 and 1994 are increasing upper-ocean salinity values on the inner Newfoundland Shelf. Over the 29-yr period, the summertime freshwater pulse has declined by 0.003 Sv and has resulted in an upper ocean that is 0.2‰ (parts per mil) more saline at station 27. The analysis presented in this section implies that interannual to decadal variations and trends in precipitation, snowcover, and river runoff over North America may play a key role in governing the state of the upper layers of the Arctic Ocean and northwestern Atlantic Ocean.

Despite its simplicity, the methodology used here to track freshwater pulses has limitations. It assumes no dispersion of the freshwater along a trajectory and does not consider tidal mixing, precipitation, or ice melt that exert a significant control on the upper layers of the ocean in the Labrador Current (Prinsenberg 1984; Mertz et al. 1993; Myers et al. 1993). In other words, we assume the freshwater discharge to be a passive tracer in the analysis. A more comprehensive approach will require the use of coupled atmosphere/ocean/sea ice/land surface models to better track the evolution and impact of freshwater discharge from HJUBs to the Labrador Current and at downstream locations. In future work, we will therefore conduct simulations with a global climate model to establish the precise role of the HJUBs’ freshwater discharge on the state of the Arctic and Atlantic Oceans and its contribution to the global thermohaline circulation.

7. Summary and future work

We examined the characteristics and trends of observed river discharge into the Hudson, James, and Ungava Bays for the period 1964–2000. Forty-two rivers with outlets into these bays contribute on average 714 km³ yr⁻¹ (= 0.023 Sv) of freshwater to high-latitude oceans. For the system as a whole, discharge attains an annual peak of 4.2 km³ day⁻¹ on average in mid-June, whereas the minimum of 0.68 km³ day⁻¹ occurs on average in the first week of April. The Nelson River contributes as much as 34% of the daily discharge for the entire system during winter, but diminishes in relative importance during spring and summer. Runoff rates per contributing area are highest (lowest) on the eastern (western) shores of the Hudson and James Bays. Linear trend analyses reveal decreasing discharge over the 37-yr period in 36 out of the 42 rivers, 33 of which are not affected by dams, diversions, and/or reservoirs. By 2000, the total annual freshwater discharge into the Arctic Ocean diminished by 96 km³ from its value in 1964, equivalent to a reduction of 0.003 Sv. The annual spring peak discharge rates associated with snowmelt has advanced by eight days between 1964 and 2000 and has diminished by 0.036 km³ day⁻¹ in intensity. There is a direct correlation between the timing of peak spring discharge rates and the latitude of a river’s mouth; the spring freshet varies by 5 days for each degree of latitude. Continental snowmelt induces a seasonal pulse of freshwater from HJUBs that is tracked along its path into the Labrador Current. It is suggested that the annual upper-ocean salinity minimum observed at station 27 on the inner Newfoundland Shelf can be explained by freshwater pulses composed of meltwater from three successive winter seasons in the river basins draining into HJUBs. A gradual salinization of the upper ocean during summer over the period 1966–94 at station 27 is in accord with a decadal trend of a diminishing intensity in the continental meltwater pulses.

A recent study by Déry and Wood (2004) documents a strong anticorrelation between the Arctic Oscillation (AO; Thompson and Wallace 1998) and HJUBs river discharge. Déry and Wood (2004) suggest that the climatological sea level pressure patterns during the alternating phases of the AO influence the dominant air masses affecting the basin. During the positive (negative) polarity of the AO, relatively cool, dry (warm, wet) air masses impact the eastern half of the HJUBs leading to substantial decreases (increases) in overall basin discharge. The recent trend toward the positive phase of the AO has cooled air temperatures and diminished precipitation in the eastern half of HJUBs (e.g., Serreze et al. 2000). This suggests that the observed decrease in HJUBs river discharge is driven by large-scale atmospheric anomalies such as the AO. A comprehensive water budget of the HJUBs has been initiated to explore in detail the link between the AO and river discharge and will be reported in a future study.