A vast portion of Canada and parts of the northern United States drain northward to the Bering Strait, Arctic Ocean, Hudson and James Bay, Hudson Strait and Labrador Sea (Fig. 1). The CAA also drains into the Arctic Ocean but remains largely ungauged (Spence and Burke, 2008), with only two small rivers in this study (see Sect. 3.1). Six separate drainage basins are considered here (from west to east): Bering Strait, western Arctic Ocean, western Hudson and James Bay, eastern Hudson and James Bay, eastern Arctic Ocean, and Labrador Sea. The gauged area totals 5.26 × 106 km2, more than half of the Canadian land surface area (Table 1). The Canadian provinces of British Columbia (BC), Alberta, Saskatchewan, Manitoba, Ontario, Québec and Newfoundland/Labrador along with the Yukon, Northwest and Nunavut territories form part of the study area. Some tributaries of the Nelson River drain a small portion of the north-central United States, namely in Montana, North and South Dakota, and Minnesota. Among the larger systems are the Yukon, Mackenzie, Back, Thelon–Kazan (hereafter collectively referred to as Chesterfield Inlet), Churchill (Manitoba), Nelson, Hayes (Manitoba), Albany, Moose, La Grande, Koksoak and Churchill (Labrador) rivers (Table 1).

The vegetation and land cover varies markedly across the vast area drained by northern Canada's rivers. The northern Rocky Mountains with peaks approaching 4000 m above sea level in the headwaters of the Yukon, Mackenzie and Nelson rivers have bare rocks, glaciers and snow with limited vegetation such as lichens and mosses. Grasslands of the central Canadian Prairies and the American northern Great Plains subject to intense agricultural activity cover the central portion of the Nelson River basin. Further north and to the east, boreal and taiga forests of the Canadian Shield span a vast portion of the study area. Arctic tundra underlain by permafrost covers the northernmost portions of these drainage basins. Several large bodies of water including Great Bear, Great Slave and Reindeer lakes, as well as with the lakes Athabasca, Manitoba, Winnipegosis and Winnipeg and countless smaller lakes, ponds and wetlands, form natural reservoirs in this system. Large artificial reservoirs developed for hydropower production exist in the study area as well, most prominently in the La Grande Rivière, Nelson, and Churchill (both in Manitoba and Newfoundland/Labrador) river basins (see Sect. 2.2 and Table 1).

The climate also varies substantially across the study area. In the mountainous terrain of north-western Canada, mean annual air temperatures remain below 0 ∘C with abundant snowfall dominating the form of precipitation. The Canadian Prairies and northern American Great Plains to the lee of the western Cordillera are relatively warm and dry (mean annual total precipitation of 300–500 mm), with most of the precipitation occurring during summer. The boreal and taiga forests experience relatively cool and wet climate regimes (mean annual total precipitation of 500–1000 mm), with both abundant rainfall and snowfall. On the Arctic tundra, cold temperatures (mean annual air temperature < -10 ∘C) and snowfall dominate the climate. The seasonal snow cover typically lasts 4–6 months on the Canadian Prairies, ∼ 6 months in the boreal forest and 6–8 months in Arctic tundra and mountainous terrain (McKay and Gray, 1981). Given these climate regimes, most unregulated rivers of northern Canada exhibit a nival regime, with low flows in winter when water is stored in the seasonal snowpack, then high flows during the snowmelt-driven freshet in spring and early summer (Déry et al., 2005a). A summer recession driven by high evapotranspiration rates follows, with possible secondary peak flows in fall caused by the frequent passage of synoptic storms (Déry et al., 2005a). High flows at times occur in summer as well in small creeks and rivers associated with severe convective activity or at larger scales when associated with intense synoptic storms. In contrast, some regulated systems exhibit low temporal variability in flows with daily fluctuations arising from hydropower demand and generation (Woo et al., 2008; Déry et al., 2011). In areas affected by permafrost, hydrological responses are relatively rapid given the limited infiltration capacity of frozen soils (Woo, 1986). Glaciers in the northern Rocky Mountains and other mountain chains also supply additional meltwater in late summer and early fall, particularly during warm, dry years (Marshall et al., 2011).

Several rivers in the study area are not only regulated for hydropower production but also for flood protection, irrigation, industrial and recreational purposes, and are thus considered moderately to strongly fragmented (Dynesius and Nilsson, 1994). The highly fragmented Nelson River basin has a long history of hydropower development, with hydroelectric generation beginning in 1906 on the Pinawa Channel of the Winnipeg River system (Manitoba Hydro, 1998). Since then there has been a proliferation of dams constructed along the Nelson River's main stem and several of its tributaries. Reservoirs such as the artificial Lake Diefenbaker (formerly a section of the South Saskatchewan River) and the natural Lake Winnipeg and Southern Indian Lake allow for seasonal water storage in this system that is managed depending on inflows, hydropower demand, flood protection and governmental regulations. In 1976, Manitoba's Churchill River was partially diverted through the Rat and Burntwood rivers (with water releases controlled at the Notigi Control Structure) for enhanced hydropower production on the lower Nelson River. An additional capacity of 7 km3 of water storage in the Southern Indian Lake was developed in the process as it thereafter became managed (Déry and Wood, 2005). Since then, approximately 75 % of the annual flows into Manitoba's Churchill River have been diverted into the lower Nelson River, greatly diminishing the Churchill River's annual inflows into the Hudson Bay (Newbury et al., 1984).

Another highly fragmented system is La Grande Rivière where the massive James Bay hydroelectric complex was developed in the mid-1970s by Hydro-Québec (Hernández-Henríquez et al., 2010). As a result of this, several large reservoirs with a storage capacity now surpassing 200 km3 have been built and are managed depending on hydropower demand and consumption. Development of the James Bay hydroelectric complex has diverted portions of the Eastmain and Opinaca rivers starting in 1980, the upper Caniapiscau River (a major tributary of the Koksoak River) in 1982 and the Rupert River in 2009 to La Grande Rivière's basin (Déry et al., 2005a). Of note, the Caniapiscau River diversion (area = 36 900 km2) induces an inter-basin transfer of 45 % of its flows or 748 m3 s-1 from the eastern Arctic Ocean toward the eastern Hudson and James Bay system (Roy and Messier, 1989). The overall drainage basin area for La Grande Rivière has now effectively doubled in size to surpass 200 000 km2 (Roy and Messier, 1989; Hydro-Québec, 2008). Just to the east of the Caniapiscau Reservoir lies Newfoundland and Labrador's Churchill River that is also managed for hydropower production. Construction of hydroelectric facilities at Churchill Falls began in 1967 and they have been fully operational since 1974. This has led to the creation of the Smallwood Reservoir with a water storage capacity of 33 km3 (Déry and Wood, 2005). In Ontario, the Moose River and its many tributaries are highly fragmented by a series of 40 hydroelectric dams with development beginning in 1911 (Benke and Cushing, 2005). However, these are mainly run-of-river projects with little storage capacity, exerting less influence on downstream flows. While the Mackenzie River's main stem is unregulated, one of its major tributaries, the Peace River (basin area ∼ 293 000 km2), remains managed for hydropower production. Construction of the W. A. C. Bennett Dam from 1968 to 1972 created the Williston Reservoir with a storage capacity of 74.3 km3. This has led to an attenuation of the seasonal cycle in downstream flows, affecting the recharge of the Peace–Athabasca Delta (Rasouli et al., 2013). Other rivers moderately affected by fragmentation in northern Canada include the Grande Rivière de la Baleine, Nottaway and Albany rivers (Dynesius and Nilsson, 1994).

This study examines 42 main rivers of northern Canada for which daily hydrometric data from gauging stations are available (Table 1). The principal source of the hydrometric data remains the Water Survey of Canada (http://www.ec.gc.ca/rhc-wsc/), with supplemental data from the Direction d'Expertise Hydrique du Québec (http://www.cehq.gouv.qc.ca/) from 2000 to 2013 for rivers in that province. Manitoba Hydro and Hydro-Québec also provide daily hydrometric data for the regulated Nelson River and La Grande Rivière, respectively. Gauges furthest downstream on a river's main stem are chosen to obtain the maximum spatial coverage and most accurate estimates of total inflows to the coastal ocean. Additional criteria used for the selection of the 42 rivers are (1) > 30 years of data availability over 1964–2013 (the study period), (2) gauged area > 1000 km2 and (3) outlets to the coastal ocean in northern Canada. Note that only the Canadian portions of the Yukon River and its tributary the Porcupine River are included here, although additional hydrometric data are available for the former near its outlet to the Bering Strait in Alaska (e.g. Walwoord and Striegl, 2007). This is to establish the direct contribution of Canadian rivers to discharge into the coastal ocean. Apart from the 42 rivers selected for this study, additional hydrometric data for tributaries that flow downstream from a gauge on a river's main stem are also included in the development of the discharge time series (Table 1). In this case, results are presented collectively and the systems are then referred to by the river's main stem. For example, results for the Peel River are added to the Mackenzie River (at Arctic Red River) as their hydrometric gauges are upstream of the confluence of these two rivers. Section 3.2.1 provides details of the construction of the river discharge time series when such situations arise.

While discharge measurements remain highly constrained observational data, errors arise nonetheless during the collection process (Lammers et al., 2001; Shiklomanov et al., 2006). Sources for these errors range from the collection method, sampling frequency, environmental conditions (e.g. under ice cover, backwater effects during ice jams, flood events, beaver dams and vegetation) and the local geography (presence or absence of a flood plain). Errors in measurements typically range from ±2 to 5 % in the absence of both a flood plain and an ice cover (Lammers et al., 2001); however, errors in measurement increase to ±5–12 % in the presence of either a flood plain and/or an ice cover. Errors may reach or even exceed these values during peak and low flows as well (Pelletier, 1988; Di Baldassarre and Montanari, 2009). While a comprehensive analysis of errors in discharge measurements is beyond the scope of this work, it is assumed this study's observational data are subject to similar errors reported by Pelletier (1988), Lammers et al. (2001), Shiklomanov et al. (2006) and Di Baldassarre and Montanari (2009). Caution is also needed in interpreting results for the Thelon and Kazan rivers (Chesterfield Inlet), as a change in recording methodology in the mid-1980s may lead to spurious trends in that system (Déry et al., 2011). Finally, flow measurement error likely decreases over time in this study, as sampling methods become more reliable and increasingly more automated.

The study period covers 50 years, starting in 1964 and ending in 2013. While long-term hydrological records are necessary to distinguish the impacts of decadal climate variability from climate change on streamflow, northern Canada has a paucity of hydrometric data prior to 1964 (Mlynowski et al., 2011). The rapid expansion of northern Canada's hydrometric network in the mid-1960s, particularly on main stem rivers with gauging stations installed near their outlets, allows the study period here to begin in 1964. There are relatively long-term (century scale) hydrometric data in more southern tributaries of some systems, however, including the Mackenzie, Nelson and Moose rivers. Availability of long-term hydrometric data for rivers draining the CAA remains limited to the Freshwater Creek near Cambridge Bay on Victoria Island (gauged area of 1490 km2 draining into the western Arctic Ocean) and the Sylvia Grinnell River near Iqaluit on Baffin Island (gauged area 2980 km2 draining into the eastern Arctic Ocean). Thus, only 0.3 % of the CAA has available hydrometric data, implying the results are not representative of this vast region where glaciers and ice caps are in rapid retreat (Gardner et al., 2011). In fact, most of northern Canada falls well below the World Meteorological Organization (WMO) standards for hydrometric gauge density, imposing limitations on this effort (Coulibaly et al., 2013). More recent hydrometric data (post 2013) remain largely unavailable due to ongoing quality control processes by various governmental agencies; thus, only operational (historical) data from the Water Survey of Canada and its provincial/territorial partners are used in this study since provisional (near-real-time) data posted online have not yet undergone quality control and analysis.

Table 1 lists comprehensive statistics and trend analyses for this study's 42 rivers over 50 years. Mean annual discharge is the highest in the Mackenzie (311.4 km3 yr-1), Nelson (102.7 km3 yr-1), La Grande Rivière (84.2 km3 yr-1) and Yukon (77.3 km3 yr-1) rivers. With the recent diversion of the Rupert River, mean annual discharge in La Grande Rivière regularly exceeds 100 km3 yr-1, with a record 129.2 km3 yr-1 in 2013. Variability (expressed by the CV) in annual discharge remains relatively low (high) in the large (small) basins. The CV in annual discharge remains relatively high in unregulated rivers of the western Arctic Ocean and western Hudson and James Bay with a maximum of 70 % for Manitoba's Churchill River where diverted and regulated flows enhance year-to-year variability. Few statistically significant trends arise for the study period, with the notable exception of rivers affected by diversions and flow regulation.

Table 2 provides aggregated statistics of the mean, SD, CV and trend of annual discharge for six regions of northern Canada from 1964 to 2013. Mean annual discharge ranges from 77.0 km3 yr-1 in the Labrador Sea to 349.9 km3 yr-1 in the western Arctic Ocean, with a total of 1154.1 km3 yr-1 for the gauged area of northern Canada. Considerable inter-annual variability in discharge exists, with the CV ranging spatially from 9 to 16 %, although this value diminishes to 7 % for the system as a whole. Relatively constant discharge to the eastern Arctic Ocean in the 2000s arises from large data gaps in this region and the infilling strategy used in the present study (Fig. 2). There is no significant trend in the 1964–2013 annual discharge to the Bering Strait, western Arctic Ocean, western Hudson and James Bay and Labrador Sea. Nonetheless, a 5-year-running mean applied to the discharge time series has shown rising annual discharge since 1990 for the western Hudson and James Bay. High flows in the late 2000s including record-high annual river discharge (438.8 km3 yr-1) to the western Hudson and James Bay in 2005, follows near-record low annual amounts (263.1 km3 yr-1) in 2003. Similarly, a reversal from record-low river discharge (258.5 km3 yr-1 in 1995) precedes record-high river discharge (419.0 km3 yr-1 in 1997) to the western Arctic Ocean. Persistent low annual discharge to the eastern Arctic Ocean from 1982 onward arises largely from the inter-basin diversion of the Caniapiscau River to La Grande Rivière, enhancing discharge into the eastern Hudson and James Bay. Discharge to the Labrador Sea shows strong decadal fluctuations that may be associated with climate variability such as different phases of the Arctic Oscillation. For northern Canada as a whole, a modest (but insignificant) positive trend of 0.21 km3 yr-1 yr-1 is found, equivalent to a change of < 1 % in mean annual discharge (Fig. 3). The 5-year-running mean shows at least two distinct phases: a declining trend in the first half of the study period followed by increasing discharge until the early 2010s. Indeed, MKT analyses reveal a significant decline of 102.8 km3 25 yr-1 for 1964–1988 followed by a significant increase of 208.8 km3 25 yr-1 for 1989–2013. A case for a possible third phase with relatively stable river discharge across northern Canada could also be argued for the central portion of the record (1985–1995).

Seasonally, spring and summer river discharge to the eastern Hudson and James Bay and the Labrador Sea decline as flows during those seasons are retained in reservoirs and released in winter for hydropower production when demand peaks (Fig. 4). The strong seasonality in flows observed in the 1960s and early 1970s in these two regions nearly vanishes in the 2000s, most notably in the eastern Hudson and James Bay region. For the mostly unregulated Bering Strait and western Arctic Ocean drainage basins, strong seasonality in flows persists through the 50-year study period. There is also a modest, statistically significant increase in winter flows to the western Hudson and James Bay and marked declines in spring flows to the eastern Arctic Ocean. Furthermore, high summer flows to the Bering Sea and to the western Hudson and James Bay arise in the 2000s. Changes in seasonality to the western Hudson and James Bay remain less pronounced than those in the eastern Hudson and James Bay owing to the dominant type of regulation (run-of-river) in the Nelson River versus the large storage capacity of reservoirs in the La Grande Rivière system. For the system as a whole, spring and summer flows are nearly equal (∼ 390 km3 yr-1), given the large (natural and artificial) storage capacity, vast areas and high latitudes of the study basins that delay (into summer) the release of snow meltwater to the Arctic Ocean and adjacent northern seas. A significant decline in summer flows, however, appears over the study period, with perhaps the exception of the 1994–2013 decade (Fig. 5). This is compensated by a gradual and significant increase in winter flows as water releases from reservoirs for hydropower production augments in post-construction decades. In contrast, fall river discharge shows no trend between 1964 and 2013. Autocorrelations between time series of seasonal discharge shows that spring/summer (r= 0.41, p < 0.05), summer/fall (r= 0.60, p < 0.05) and fall/winter of the following calendar year (r = 0.31, p < 0.05) are temporally correlated, showing persistence in seasonal flows. Despite this, there are no statistically significant correlations between time series of river discharge in other seasons, including between spring and fall of a given year.

Large, significant trends in the 1964–2013 annual river discharge occur mainly in regulated systems (Fig. 6). A notable exception is the Chesterfield Inlet that shows a 4.9 km3 50 yr-1 increase across the study period, noting however the potential recording issues for this system (see Sect. 3.1). Rivers draining into the eastern Arctic Ocean and Labrador Sea nearly all show declines between 1964 and 2013, in part owing to flow regulation, retention and diversions; nearby unregulated systems also show declines. Couplets of large positive and negative trends to western and eastern Hudson and James Bay arise from diverted flows from one system to another. Otherwise, there are no significant trends in river discharge to the Bering Strait and western Arctic Ocean during the study period. Seasonal analyses reveal a consistent pattern toward greater winter discharge across northern Canada (with a few exceptions) in both regulated and unregulated systems, with few significant changes during the shoulder seasons apart from the strong positive trends in fall in La Grande Rivière and the Nelson River, and strong (but insignificant) spring discharge increases in the Nelson and Mackenzie rivers (Fig. 7). In contrast, there is a general trend toward less river discharge during summer with the exception of La Grande Rivière and the Nelson River where diversions from nearby systems enhance flows in all seasons. Chesterfield Inlet exhibits a strong positive trend in summer discharge but again, care must be taken in interpreting this result given the changes in recording methodology in the 1980s (see Sect. 3.1).

Impacts of regulation on discharge variability are examined by considering inter-decadal differences in the variability of paired regulated and unregulated rivers (see Sect. 3.2.2). Inter-decadal analysis is used because, in some large systems, there has been a stepped introduction of hydroelectric development over the 5 decades in this study (e.g. Nelson River and La Grande Rivière; see Sect. 2.2). Table 3 presents statistics of inter-decadal variability in discharge for eight major regulated rivers and seven of their unregulated counterparts. Regulated rivers in this table are in bold, and their unregulated counterparts are found in the row that follows. Figure 8 shows box-and-whisker plots of CVs for regulated versus unregulated rivers for each season over the 5 decades. Table S1 in the Supplement provides the inter-decadal variability of all 42 rivers across seasons.

The greatest intra-decadal variability is seen from 1974 to 1983, a period of rapid construction and diversions in most of the regulated rivers (Table 3 and Fig. 8). Flow regulation generally suppresses variability compared to matching unregulated rivers, as has been observed in the Eurasian Ob and Yenisei rivers (Yang et al., 2004a, b). This effect is the greatest post-construction during winter and to a lesser extent in the fall, presumably to accommodate higher energy demands (1984–2003; Fig. 8). The Churchill River (Manitoba) is a noteworthy exception to this trend, with increases in inter-decadal CV post-diversion (1984–2013 CV = 0.35–0.67) compared to pre-diversion (1964–1973 CV = 0.11). Much of this increase in variability has occurred during fall and winter over the last 2 decades (1994–2013 CV = 0.94–1.16; Table S1) following a period of relatively lower variability from 1984 to 1993.

Interestingly, enhanced variability arises in both regulated and unregulated rivers in the most recent decade (2004–2013) for all seasons, but most notably during summer (Fig. 8). The unregulated Nottaway River experiences its highest summer variability (by a factor of 2) compared to previous decades (CV = 0.47), and both the regulated Churchill (Manitoba) and Moose rivers experience their greatest flow variability with their matching unregulated rivers experiencing their second greatest flow variability (Table 3). Increasing overall discharge trends reported in this study are likely influenced by increasing mean summer discharge and variability in the most recent decade (2004–2013), which seems to be largely climate driven as it occurs in both regulated and unregulated systems. Particularly in the western Hudson and James Bay region, an increasing number of large summer precipitation and rainfall–runoff events in recent years have, in some cases, yielded annual hydrographs with dual peaks (Ahmari et al., 2016; Blais et al., 2016). Increasing variability may also be influenced by the changing magnitude, timing, frequency and duration of flood events observed in some gauges of our study area (Burn and Whitfield, 2016). Of note is the introduction of regulation on the Rupert River (in 2009) and enhanced variability from 2004 to 2013 not seen otherwise in rivers draining into the eastern Arctic Ocean or Labrador Sea (with the exception of the Alexis River).

Climatological hydrographs of daily river discharge for each of the six regional drainage basins exhibit relatively consistent patterns characteristic of nival regimes, with low flows in winter, a spring freshet induced by snowmelt, followed by a recession toward lower flows in late summer and fall (Fig. 9). Higher winter flows to western and eastern Hudson and James Bay and the Labrador Sea occur partly because of strong regulation and flow retention in reservoirs during other seasons. Noticeable reductions in streamflow input to the Hudson and James Bay that occur at the start and near the end of the calendar year happen in association with New Year's, Christmas and Boxing days. Reductions in hydropower demand and consumption during these Canadian statutory holidays abruptly decrease flows into Hudson Bay from the highly regulated Nelson, Moose and La Grande Rivière systems.

Milliman and Farnsworth (2011) provided a comprehensive table of mean annual discharge for most rivers included in this study. Results herein are generally consistent with the mean annual discharge from Milliman and Farnsworth (2011) with variations likely dependent on the selected study period and basin area under consideration. Exceptions arise in some rivers, most notably for Nunavut's Coppermine River where Milliman and Farnsworth (2011) reported a mean annual discharge of 11 and 2.6 km3 yr-1 before and after regulation, respectively. Results from this study yield a mean annual discharge of 8.8 km3 yr-1 for the Coppermine River – with no known flow regulation over 1964–2013 according to the Water Survey of Canada. Milliman and Farnsworth (2011) also reported mean annual discharge rates of 23 km3 yr-1 and 15 km3 yr-1 before and after regulation, respectively, for northern Ontario's Severn River (despite this system being unregulated), with our results indicating mean annual discharge of 21.9 km3 yr-1. The present study also reports underestimates of ∼ 7–13 km3 yr-1 for the Albany, Attawapiskat and Winisk rivers of northern Ontario compared to Milliman and Farnsworth (2011), perhaps owing to different study periods and/or basin areas under consideration. Discrepancies in mean annual discharge for La Grande Rivière, the Eastmain and Rupert rivers are likely due to recent water management practices to enhance power production at the James Bay hydroelectric complex. Statistics reported in this study are based on hydrometric data provided directly by Hydro-Québec (post development of hydropower facilities and infrastructure) that better reflect the current level of regulation in these systems.

Benke and Cushing (2005) provided mean annual streamflow statistics for 11 rivers examined in this study (i.e. the Porcupine, Yukon, Chesterfield Inlet, Seal, La Grande, Koksoak, Mackenzie, Moose, Nelson and Churchill (both in Manitoba and Labrador) rivers). Our results are generally consistent with Benke and Cushing (2005) with two notable exceptions. First, mean annual discharge for the Mackenzie River is ∼ 27 km3 yr-1 greater in this study, perhaps due to the exclusion of the Peel River's contribution to overall Mackenzie River discharge by Benke and Cushing (2005). Second, Benke and Cushing (2005) reported a mean annual discharge of ∼ 200 km3 yr-1 for the Yukon River, although this considers both the Canadian and American contributing area (total of 839 200 km2), rather than just the upstream part in Canada examined here (gauged area = 288 000 km2). Discrepancies also arise for the Koksoak River and La Grande Rivière as mean annual discharge values reported by Benke and Cushing (2005) reflected conditions prior to the diversion of the upper Caniapiscau River to the La Grande Rivière system.

While Déry and Wood (2005) first reported a 10 % decline in river discharge across northern Canada from 1964 to 2003, this effort finds a remarkable reversal to that trend in expanding the study period by only a decade. In the second half of the study period, river discharge in northern Canada increased by 18.1 % (relative to its overall mean annual discharge). This is in stark contrast to the first half of the study period during which river discharge declined significantly. Rood et al. (2016) also documented a statistically significant (p < 0.05) ∼ 1.5 % decade-1 increase in Mackenzie River discharge over 1939–2013, consistent with the pattern observed in this study; however, our abbreviated study period reveals an insignificant trend in this system (Table 1). St. Jacques and Sauchyn (2009) likewise reported significant increases in winter (base) flows for the Mackenzie and other rivers of the Northwest Territories in possible response to permafrost degradation. These findings suggest that rivers in northern Canada are now responding similarly to rising air temperatures as those in Eurasia (Peterson et al., 2002), in accord with climate change projections (Milly et al., 2005; van Vliet et al., 2013). In fact, the rate of increase of 8.4 km3 yr-1 yr-1 for river discharge across northern Canada from 1989 to 2013 exceeds the overall trend of 6.3 km3 yr-1 yr-1 for 16 Eurasian rivers draining into the Arctic Ocean from 1964 to 2000 (McClelland et al., 2006). The period of record remains relatively short and care must be taken in interpreting these findings; nonetheless, attempts should be made to reconcile observed trends in northern Canada and Eurasia using identical study periods and methodologies. Decadal climate variability associated with the Arctic, Pacific Decadal, and Atlantic Multidecadal oscillations, among other large-scale modes of climate variability, are known to affect river discharge in northern Canada (Déry and Wood, 2004; Kingston et al., 2006; Assani et al., 2010; Rood et al., 2016). We stress that it remains important to continue monitoring river discharge in northern Canada (and expand where possible to attain WMO standards) as this region is expected to continue warming rapidly in the twenty-first century (Coulibaly et al., 2013; Gough and Wolfe, 2001; Gagnon and Gough, 2005).

While rising air temperatures and changing precipitation patterns are key factors in altering northern Canadian river discharge, another control remains anthropogenic activities such as water retention, regulation and diversion. The development of large hydroelectric complexes in northern Québec, Ontario, Manitoba and across the Canadian Prairies into BC has significantly altered the seasonality of flows in northern Canada, most notably to western and eastern Hudson and James Bay. River diversions and flow regulation typically do not influence overall flow volumes to the coastal ocean (McClelland et al., 2006); however, pronounced changes in seasonality accompany regulation, especially in systems with large storage capacity such as the Mackenzie (Peace) River and La Grande Rivière. Furthermore, short-term (i.e. 1–5 years) declines in river discharge to the coastal ocean can arise from the filling of large reservoirs for hydropower production. Across northern Canada, > 300 km3 of water storage capacity has been developed since 1964 (Lee et al., 2012). This may lead to both a short-term decline in observed flows to the coastal ocean while reservoirs are filled, and “aging” of water in storage affecting its properties, such as biochemistry and temperature, while enhancing evaporative losses (Vörösmarty and Sahagian, 2000). Perhaps lesser known are the impacts of land cover and land use change on river discharge in northern Canada. Deforestation through wood harvesting depresses water demand by vegetation while increasing soil moisture and runoff generation (Boon, 2012). In contrast, the intensification of agricultural activities, particularly in the Canadian Prairies, increases water demand for irrigation. Anthropogenic activities play a major role in changing pan-Arctic hydrology that requires special attention when assessing observed changes. Thus, the use of land surface or hydrological models is particularly useful in identifying the individual roles of climate change and anthropogenic activities on streamflow variability and trends that otherwise may be masked in the observational data.

Changes in seasonal river conditions induced by climate change and river regulation affect the physical regime of coastal estuaries by modifying salinity levels and the input of nutrients and sediments (Gillanders et al., 2011). For instance, the Eastmain River estuary experienced an increase in salinity after diversion into the La Grande Rivière system (Messier et al., 1986; Drinkwater and Frank, 1994), concurrently lowering salinity in La Grande Rivière's estuary during winter (Whittaker, 2006). Increasing river discharge in northern Canada strengthens ocean stratification, thereby suppressing deep-water formation in the Labrador Sea (Myers, 2005). This, in turn, weakens the thermohaline circulation that is responsible for the transport of heat and nutrients in the North Atlantic Ocean (Ogi et al., 2001; Rennermalm et al., 2007). Increasing winter discharge in northern Canada also delivers sensible heat to the coastal ocean, promoting the ablation of sea ice in estuaries and delta regions (Kuzyk et al., 2008; Searcy et al., 1996). This emphasises the need for basin-scale numerical modelling of the coupling between freshwater fluxes and the marine environment (e.g. Saucier et al., 2004).